STARTPAGE - numerical.rl.ac.uk · eigenvectors, singular values and singular vectors for largescale...
Transcript of STARTPAGE - numerical.rl.ac.uk · eigenvectors, singular values and singular vectors for largescale...
ARTDECO
STARTPAGE
HUMAN RESOURCES AND MOBILITY (HRM)ACTIVITY
MARIE CURIE ACTIONSResearch Training Networks (RTNs)
“Interdisciplinary and Intersectorial”
Call: FP62005Mobility1
PART B
STAGE 2 – FULL PROPOSAL DESCRIPTION
“ARTDECO”
Part B – Page 1 of 45
ARTDECO
Table of ContentsPARTNER LIST.................................................................................................................................................4B.1 Scientific quality of the project....................................................................................................................5
B.1.1. Research topic.....................................................................................................................................5B.1.2. Objectives of the Network..................................................................................................................5B.1.3 Scientific originality.............................................................................................................................6B.1.4. Research method.................................................................................................................................7B.1.5. Work plan............................................................................................................................................7
B.2 Training and transfer of knowledge activities...........................................................................................10B.2.1. Content and quality of the training and transfer of knowledge programme....................................10B.2.2. Planned recruitment of earlystage and experienced researchers....................................................13B.2.3. Training Work plans.........................................................................................................................13
B3 QUALITY/CAPACITY of the Network PARTNERSHIP........................................................................16B.3.1. Existing collaborations......................................................................................................................16B.3.2. List of Publications...........................................................................................................................16B.3.3 Individual contributions.....................................................................................................................20Team 1 ERCIM EEIG..............................................................................................................................20Team 2 CCLRC RAL CSE Numerical Analysis Group.........................................................................20Team 3 Oxford University Computing Laboratory, Numerical Analysis Group...................................21Team 4 IMATI CNR, Pavia, Italy ..........................................................................................................23Team 5 University of Pavia, Departments of Mathematics and Structural Mechanics .........................23Team 6 Parallel Algorithms team of CERFACS....................................................................................24Team 7 IRITENSEEIHTINPT “Algorithmes Parallèles and Optimisation”.......................................25Team 8 INRIA SophiaAntipolis ............................................................................................................26Team 9 FUNDP, Department of Mathematics, Numerical Analysis Unit, Namur ...............................27Team 10 Academy of Sciences of the CR, ICS, Department of Computational methods.....................28Team 11 Delft University of Technology...............................................................................................29Team 12 SEPRA......................................................................................................................................30
B4. Management and feasibility.......................................................................................................................32B4.1 Proposed management and organisational structure..........................................................................32B4.2. Management knowhow and experience of network coordinator ERCIM......................................39B.4.3. Budget...............................................................................................................................................40
B.5 Added value to the community and relevance to the objectives of the activity........................................41B.5.1 Connections with other FP6 thematics.............................................................................................42
INDICATIVE FINANCIAL INFORMATION ...............................................................................................44OTHER ISSUES...............................................................................................................................................44ETHICAL ISSUES CHECKLIST....................................................................................................................45
Part B – Page 2 of 45
ARTDECO
PARTNER LIST
Participant number Participant Organization legal name and Department Country
1 GEIE ERCIM F
2Council for the Central Laboratory of the Research
Councils, Rutherford Appleton LaboratoryComputational Science and Engineering Department
UK
3Oxford University
Computational LaboratoryUK
4IMATICNR
PaviaI
5University of Pavia
Department of Mathematics andDepartment of Structural Mechanics
I
6CERFACSToulouse
F
7INPTIRITToulouse
F
8 INRIA
SophiaAntipolisF
9FUNDP, Department of Mathematics, Numerical
Analysis Unit, Namur B
10Academy of sciences of the Czech republic
Institute of Computer ScienceCZ
11Delft University of Technology, Faculty of Electrical
Engineering, Mathematics and Computer Science
NL
12 SEPRA NL
Table 1 List of Participants
Part B – Page 3 of 45
ARTDECO
“All the life that whirls about us, runs, and stops is not only dependenton mathematics for its comprehensibility, but has effectively come intobeing through it and depends on it for its existence, …” (Robert Musil,“The Mathematical Man”, 1913.)
B.1 SCIENTIFIC QUALITY OF THE PROJECT
B.1.1. Research topic
To compete in the new global market, European industry must take advantage of the competitiveedge that would be gained from using European expertise in applied mathematics and scientificcomputing. Engineers, applied scientists and mathematicians working in industry can greatlybenefit from close collaboration with academics and mathematicians with skills and knowledgerelevant to their applications. Moreover, there is a shortage of mathematicians within the Europeanindustry, particularly in the sectors of computational mathematics. Thus, there is the need for a newaction to bring industry up to date with stateoftheart mathematical ideas, methodologies, toolsand techniques. Academic resources in Mathematics for Industry are also scarce and fragmentedacross Europe, while industrial demands are widely spread. Exchange and interaction are necessaryin training, research and industrial arenas.
In 2002, ERCIM (European Consortium for Informatics and Mathematics), a grouping of leadingresearch institutes from eighteen European countries, created a Working Group on “Application ofNumerical Mathematics in Science” to support and implement cross fertilisation of numericaltechniques in different areas and by different teams. This Working Group has decided to submit this proposal for the creation of a “MarieCurieResearch and Training Network” aiming to become an international leader and a reference pointwithin Europe on this underpinning theme of computational mathematics. A survey of the active researchers in the laboratories of the organisations that plan to participate inthe Network indicates that the following major fields, in the area of Largescale ScientificComputing, have strategic interest and strong interactions:
• Numerical Linear Algebra;
• Numerical Solution of Differential Equations;
• Continuous Optimization and Optimal Control.
Each of these fields frequently uses techniques developed in one of the others. By creating anetwork of European researchers, the increased collaboration and communication between the fieldswill further strengthen this research and enhance its impact on science and industry. Moreover, wewill have the opportunity to develop a new generation of mathematical software based on modularcomponents.
B.1.2. Objectives of the Network
The global aspiration of the network is to link together mathematicians working in National
Part B – Page 4 of 45
ARTDECO
Laboratories, Universities, and R&D groups in Industry. We now briefly describe the specific topics of research in each field:• Numerical Linear Algebra. The topics range from sparse matrix theory, direct and iterative
solvers for large and sparse linear systems of equations, to the computation of eigenvalues,eigenvectors, singular values and singular vectors for largescale and /or structured matrixproblems, including the use of symbolic manipulation techniques for the solution of polynomialsystems of equations. Parallel and grid computing should be added to them.
• Numerical Solution of Differential Equations. The topics of major interest are finiteelementmethods, finitevolume methods, mesh generation, multigrid methods, wavelets, spectralmethods, multilevel optimization for discretized problems and timestepping methods.
• Continuous Optimization and Optimal Control. The topics of interest are interiorpointmethods for largescale linear, quadratic and nonlinear programming and SQP methods fornonlinear programming and optimal control.
Within these fields, we have identified some integrated actions around which we can propose aninterdisciplinary programme of research that will integrate the current activities focusing them on acommon interest “kernel” target. In particular, we have identified among all our research topics thefollowing transverse research lines as scientifically relevant for our cross fertilization project:1. Saddle points The computation of saddle points is a key problem in many fields. Several
participants in the existing Working Group are already studying this problem in the context ofdifferent applications, thus building multiple perspectives for its handling and solution. Theseapplications include mixed finiteelement approximation of partial differential equations inelasticity and fluid dynamics, and interiorpoint and SQP algorithms for optimization. Parallelcomputing can be the only way to approach the solution of these problems where the number ofparameters to determine is exceedingly large.
2. Optimal design and shape optimization The design of new materials and the optimization ofartefacts (airplane, electrical devices, cars) in engineering requires the combined efforts ofoptimization, approximation, and linear algebra experts in order to simulate and solve thecomplex phenomena under scrutiny.
For the first line of research, we plan to update existing and to develop new software in order tohave efficient and flexible modules that, to address the second line of research, will be combined bythe design of specific software interfaces.
B.1.3 Scientific originality
The scientific originality of the project is in the cross fertilization between several fields of researchand in its clear impact on European industry. Under the largescale scientific computing title, weoften find problems that are solved with mathematical techniques that are not at the cutting edge.The time needed for the implementation of new successful ideas in industrial software is too longeven without taking into account the necessity of a complete validation of the new release. We believe that a possible cure for this defect is the training of a new generation of young applied
Part B – Page 5 of 45
ARTDECO
mathematicians with a strong field of specialization, combined with an outstanding knowledge inseveral other fields of applied and computational mathematics. This can be achieved using ournetwork of universities, research laboratories, and industrial research and development centreswhere the young researcher will be involved with the cutting edge research and softwaredevelopment.In particular, we hope to exploit new optimization techniques that allow us to solve problems with avery large number of degrees of freedom (> 107). We expect that several application areas will benefit from the results and the activities of theNetwork: we mention, for example, simulation of electromagnetic phenomena, electrical circuittheory, computational chemistry, computational biology, computational materials, CFD andstructural engineering, optimal design, mathematics for financial derivatives, finiteelementmodelling for medical simulation, and environmental modelling.
B.1.4. Research method
The main project objective is to integrate several new algorithms and numerical techniques inScientific Computing, in order to characterize and increase the accuracy of the models and to extendtheir field of applicability. We intend to achieve the target of integrating the expertise of each teamby using young scientists and by organizing meetings and schools, where each team can present insimple terms their problems or their algorithms. Currently, the ERCIM working group alreadyorganizes similar activities. Nevertheless, we feel the need for a more sustainable and institutionaleffort. The successes of integrated initiatives and the long standing record of cooperation betweenseveral of the teams, stimulates our proposal. Moreover, we think that the area of Numerical Linear Algebra represents the natural “glue” that willmaintain the cohesion between the teams. Its fundamental value is in its immediate applicability toseveral problems and in its closeness to the software. Therefore, we plan to use the expertise inNumerical Linear Algebra resident within the majority of the teams to ensure a threadsafeapproach to the implementation of the research topics described in Section 1.2 of this proposal.
B.1.5. Work plan
The objectives described in Section 1.2 are subdivided in work packages and then into several tasks.We describe the contents of each task and, in Table2 we relate the teams to the tasks. Moreover, weadd an objective related to the dissemination and the management of the network.
WP1.Saddle point problems:
Task-1.1.Direct methods: comparison between existing direct solvers and reordering techniques
for fill-in control, parallel computing and grid computing. On the basis of this evaluation,improvement of existing, and development of new techniques.
Task-1.2.Iterative solvers: Evaluation and improvement of existing saddle point preconditioning
Part B – Page 6 of 45
ARTDECO
techniques in combination with Krylov and Chebyshev methods. Study and evaluation of thecombination of saddle point preconditioners with spectral preconditioning (deflation of thesmallest eigenvalues in absolute value). Study and evaluation of hierarchical solvers such asmultigrid and Domain Decomposition.
Task-1.3.Mixed approach: the study of mixed direct and iterative approaches and theircomparison on very large problems with direct and iterative solvers, Null and Range Spacealgorithms.
Task-1.4. Applications: application and comparison of these methods, taking advantage of thestructural properties of the discrete operators to the solution of mixed finiteelementapproximations. Applications could include Darcy’s laws, Biharmonic problems,Computational Fluid Dynamics, and NavierStokes equations. Application of interiorpointalgorithms for largescale problems in the numerical solution of PDEs.
Task-1.5.Aposteriori error estimator and stopping criteria: study of suitable error formulaein selfadaptive discretization methods. Identification of stopping criteria suitable foriterative methods applied to PDEs. Study of stopping criteria with respect to the aposteriorierror estimation and the error control in relation to the underlying PDEs.
WP2.Optimal design and shape optimization
Task2.1. Algorithms for elliptic control problems: use of interiorpoint techniques for thesolution of constrained optimization problems. Study of the discretization of the differentialoperators.
Task2.2. Multiscale optimization: the use of multigrid/multilevel techniques in the context ofoptimization problems arising from PDEs.
Task2.3.Adaptation of optimization algorithms to optimal design: study and design ofimproved general software in constrained optimization to the case where the constraints arePDEs.
Task2.4.Applications: identification of suitable models, their approximation and solution inelectromagnetism, aerodynamics, and in shapememory alloys simulation.
Task2.5. Sensitivity analysis: evaluation by Automatic Differentiation and use of adjointoperators in the study of the stability and sensitivity of the solution to the design parametersand in the identification of the optimal conditions.
WP3.Dissemination and Management
Task-3.1.Internal Dissemination, providing collaborative tools, ensuring a coherent informationflow within the network and supporting the organisation of summer schools, workshop andconferences.
Task-3.2.External dissemination, through dedicated support ranking from ARTDECO web site,to newsletter, ERCIM news and publications, and the participation in international
Part B – Page 7 of 45
ARTDECO
conferences. The objective is not only to advertise the network’s activities but also to attractnew candidates while promoting ARTDECO achievements on a global scale.
Task-3.3.Administrative coordination, to ensure the completion of the work plan, theproduction of reports and deliverables, to handle the administrative dimension of thetraining programme (student contracts), to act as an interface with the EuropeanCommission and manage arising issues.
Task-3.4.Financial coordination, to ensure funding redistribution among the network teams,and to perform the management of financial resources for and towards the students onfellowships.
For each scientific task, the major deliverable will be trained students, technical reports andscientific publications. We assume that some of the results will be incorporated before the end ofthe project in numerical software libraries. This will be consistent with the tradition of severalteams in building and maintaining software libraries. The general activity reports and annual reportswill be delivered through Task 3.
T a s k
Team
1.1 1.2 1.3 1.4 1.5 2.1 2.2 2.3 2.4 2.5 3.1 3.2 3.3 3.41 x x x x2 x x x x x x x x x x x x3 x x x x x x x x x x4 x x x x x x5 x x x x6 x x x x x x x x x x7 x x x x x x x x x8 x x x x x x x x x9 x x x x x x x10 x x x x x x x x x11 x x x x x x x12 x x x x x
Table 2 Team vs task connection
Part B – Page 8 of 45
ARTDECO
B.2 TRAINING AND TRANSFER OF KNOWLEDGE ACTIVITIES
B.2.1. Content and quality of the training and transfer of knowledge programme
It is widely acknowledged today that young scientists represent one of the most rapid and costeffective means of dissemination and diffusion of new knowledge, methodologies and concepts tothe industrial world. The ARTDECO network will create an environment that will facilitate thisprocess investing the majority of the funds in a fellowship programme and in summer schools. Itwill also provide the framework for early interaction between young PhDs, postdoctoralresearchers and industry. This interaction will enlighten researchers to the workings of industry, andwill put industry in contact with potential new staff recruits specializing in areas that are of itsspecific interest. Furthermore, the involvement of industrial lecturers in the training and teachingprogramme will help academic researchers to select the key and difficult industrial problems thatrequire new techniques.
Based on the experience developed in the context of ERCIM, we propose a fellowship programmewhere a young mathematician will spend part of his/her threeyear fellowship in at least twoacademic institutions and another part in an R&D centre if his/her research studies are of interest toan industry associated or cooperating with a participant. A Steering Committee will evaluate andrecommend the value of the applications.
The partners involved in this proposal already form a multidisciplinary consortium where theexisting teams can bring their expertise to the training network and to the students. ERCIM will bethe management structure ensuring the administrative and financial coordination of the network,while the other teams involved will focus on the scientific agenda and the training.
The ARTDECO management organisation will rely on a coherent architecture organised around thework packages: WP1 and WP2 will address the joint research and training agenda, backed up by theResearch and Training Coordination Board (RTCB); WP3 will be dedicated to the disseminationefforts to be conducted within and outside the network and will ensure the general management ofthe project, performing the administrative and financial coordination.The RTCB is composed of the scientific coordinator, Dr Mario Arioli, who also chairs the ERCIMWorking Group “Applications of Numerical Mathematics in Science”, and all the work packageleaders. The scientific coordinator chairs the board and all decisions are taken by a majority vote.The RTCB will consider also the possibility of inviting an external member of internationaloutstanding mathematical prestige and educational experience from outside the network and the EUas an advisor on international cooperation with nonEuropean institutions.RTCB will supervise and stimulate the overall research and training activities while fosteringexchanges and interactions among the scientific Work Packages. In particular, the RTCB willcoordinate the activities of the PhD students with special attention to the legal and institutional
Part B – Page 9 of 45
ARTDECO
aspects concerning the fulfilment of their duties visavis the European University responsible forawarding the title of Doctor of Philosophy. Moreover, when legally possible, we will encourage therecognition of the PhD title by another European University participating in the Network andinvolved in the training of the candidate. The RTCB will define the student selection requirementsfor all partners and will cooperate with the ERCIM office in charge of implementing the selectionand integration of young researchers. The RTCB is also to discuss the multidisciplinary scientificorientations and will assess the quality of the network’s achievements as an internal peer reviewcommittee.
The project involves institutions with an established record in research and in education. Amongthem we have eleven research centres and laboratories with a vast range of expertise andcompetence in numerical linear algebra (CCLRCRAL, Academy of Sciences of the CzechRepublic, CERFACS, INPTIRIT, Delft University of Technology), in the numerical solution ofpartial differential equations (IMATICNR, INRIA, Oxford University, Delft University ofTechnology, SEPRA), and in optimization (CCLRCRAL, University of Namur, CERFACS,Academy of Sciences of the Czech Republic, INRIA, INPTIRIT). They are some of the leadingcentres for research and education in Europe.The fields of expertise of the University participants cover all the sectors indicated in Section B.1.2.Moreover, all the Universities involved have both the administrative and the training skills requiredby the project.
Industry, in particular our industrial partner SEPRA, will be involved in establishing the lines ofresearch and in the preparation of the calls for application. In particular, some of the partners(ERCIMW3C, CERFACS, Delft University of Technology, INRIA, and CCLRCRAL) have along standing cooperation with industries potentially interested in the outcome of the project. Thiscooperation will facilitate the task of transferring the results and the knowhow. The allocation ofsome of the experienced researchers to industry will be decided in a later phase of the project whentheir training will reach an appropriate level. The subdivision of the fellowships among the partners is based on the level of personal involvementdeclared by each of them. We divide the project into two phases:
1. In the first year we will recruit six experienced researchers to guarantee a sound basis for theactivities. The subdivision among the partners will reflect the resulting levels of qualification ofthe researchers that we recruit.
2. The recruitment of eleven earlystage researchers will start after the first year of the project.This delay will allow the academic institutions responsible for their PhD to fulfil the formalitiesrequired by the local University and by the national state legislation. The RTCB will take chargeof the scientific coordination between the hosting institutions and the originating one.
The training of the earlystage and experienced researchers will be based both on individual transferof knowledge and on the participation in summer schools where all the partners in turn will
Part B – Page 10 of 45
ARTDECO
contribute with their personal involvement. Early-stage researchers will also be encouraged to take
courses in the specialized Graduate Schools organized by the university partners (thus earning the ECT
credits associated with these courses).
During the four years of the project, we plan to organise two summer schools that will cover, duringa three to four weeks period, the research topics described in Section 1.5 and other topics that mightarise in the future. Each team will have the opportunity to lecture on their research expertise.The first summer school will serve as a bridge between the expertises on saddlepoint computationin optimization, perturbation analysis, and the numerical solution of partial differential equations.The second summer school will be organized around the use of advanced computationaloptimization, optimal design and their applications.We will require the speakers and all the team members to produce teaching material (slides orlecture notes) that will be used during the summer schools. This material and all the publicationsrelative to the project produced by the participants in ARTDECO will be made available on theInternet through the home web page of the project.Furthermore, we will organise workshops and integrated teaching programmes involving industrialand national laboratory lecturers. We will take full advantage of electronic communication creatingweb pages where the scientific activities, databases of test problems, and a newsletter are presented.The Network will also provide the organizational support for the movement of the participatingresearchers between the institutions. Moreover, we will encourage all the earlystage andexperienced researchers to present the results of their activities during the workshops organised bythe project and we will support their participation in international conferences in their specificsector.
We anticipate that the use of the less traditional forms of communication (such as articles for nonspecialists, web pages), as well as other types of activities, will enhance the appreciation ofcomputational mathematics in general, and will render the role and the contribution of the proposedNetwork in science and technology transparent for the general public.The ratio between individual and networkwide training is a delicate issue that involves carefulidentification of the individual psychological and scientific needs of the earlystage researchers.The English language will be the official language of the network and we must ensure that theearlystage researchers are fluent in it, taking into account that they are supposed to visit the otherteams involved in their training. Thus, the balance between the individual training given by theinstitution legally responsible for the PhD and the period of individual training abroad must bedecided by the tutors responsible for the earlystage researcher. The RTCB will arrange that all theearlystages and the experienced researchers have the possibility of participating in the commonsummer schools and in the common workshops.Finally, the presence of national laboratories equipped with large computing facilities such as theHPCx at CCLRC will give the possibility for all the participants and especially for the earlystageand experienced researchers to be trained on the use of stateoftheart parallel techniques andadvanced numerical software.
Part B – Page 11 of 45
ARTDECO
B.2.2. Planned recruitment of earlystage and experienced researchers
From the experience of earlier participation in European networks, we believe that there is a needfor an important recruitment effort at the beginning of the project. A slow start to the recruitmentcan otherwise become a major difficulty. Experience shows that recruiting young researchers fromwestern European countries is not an easy task. We are convinced, however, that extending ourrecruiting areas to the new member states will ensure that it is possible to recruit excellentcandidates. We will start recruiting as soon as possible by advertising vacancies on the web (Cordis,ARTDECO Web Page and our institutions' Web Pages), in journals and electronic newsletters.ARTDECO will also contact colleagues throughout Europe, including the new member states, andwe expect that this will lead to the main source of candidates.
Owing to the realistic difficulty in the recruiting of earlystage researchers, we plan to proceedduring the first year with the recruitment of six experienced researchers that will help the startupprocess and will sustain the crossfertilisation effort during the first two years. We will use all theelectronic (Nanet the numerical analysis network and OptNet and SIAGOpt the optimizationnetworks) and classical channels of information for advertising the positions. The recruitment willalso be sought via the extensive European Grid network. Many Grid applications involve numericalcomputing, and this will provide a useful recruitment and dissemination channel for a network ofthis kind.The experience of ERCIM in running its fellowship programme will be crucial in this phase. Duringthe first year, the university teams will proceed with the selection of suitable candidates with thenecessary intellectual qualities and suitable background. The ERCIM consortium advertises postdoctoral fellowships throughout Europe twice a year. This campaign results in roughly 150candidates in all areas of applied mathematics and computer science, including discretemathematics, systems and control and computer science. This constitutes an excellent source ofcandidates for experienced researcher’s positions.Particular care will be paid to ensure that all candidates get equal opportunities. Moreover, allpartners are well aware of the necessity to promote the gender issue during the recruitmentprocedure. It will be clearly indicated in the advertisement that particular attention will be given toencourage (and give opportunities to) young women researchers to pursue their career in the field.We believe that the training programme will initiate a permanent flow of young European experts inthe area of Computational Mathematics. This objective will be achieved by making the researcharea more visible in Europe, by the organisation of workshops and summer schools that will beopen to the entire European research community, by the dissemination of our lecture notes to allnetwork institutions, and by the close interactions between early stage and experienced researcherswithin the network. Thus, the total personmonths requested for experienced researchers is 216months (6×36 months) and for earlystages researchers is 396 months (11×36 months).
B.2.3. Training Work plans
In the first six months, in addition to the recruitment of the experienced researchers, we plan to
Part B – Page 12 of 45
ARTDECO
prepare a syllabus related to the prerequisites that we assume the earlystage researchers mustpossess or acquire during the initial months of their training. We intend to use the local trainingoffer by the Universities involved in ARTDECO in order to fulfil the development of thecomplementary skills required. After this preliminary phase, we will organize the first SummerSchool. This school will be oriented to the introduction of the numerical methodologies in the fieldsof Numerical Linear Algebra and Optimization required in the solution of large scale industrialproblems (Aeronautics, Material Science).A second Summer School will be oriented to more advanced topics related to WP2 (Optimal Designand Shape Optimization). Finally, we plan three workshops oriented to specific thematics connected with more advancedtopics of research and/or industrial problems. All the experienced and earlystage researchers willparticipate in the workshops. They will be exposed to the frontend line of research and they willfind a forum in which present their activities. In Table 2, we summarize the tentative timescale of the following training task packages:
TP1. Summer schools and training
Training Task1. Syllabus
Training Task2. First Summer School: the main topics will be
1. Introduction to direct and iterative solvers for Saddle Point Systems (WP1
Task-1.1, Task-1.2, Task-1.3, Task-1.5)
2. Optimization algorithms (WP2 Task-2.1, Task-2.2)
3. Introduction to applications (WP1 Task-1.4 and WP2 task-2.4)
Training Task3. Second Summer School: the main topics will be:
1. Optimal design and shape optimization algorithms (WP2 Tasks 2.2, 2.3 2.4)
2. A-posteriori error estimators and Sensitivity Analysis (WP1 Task-1.5, WP2
Task-2.5)
3. Industrial Applications
TP2. Workshops
WS Task-1. Automatic Differentiation and Sensitivity Analysis
WS Task-2. Matrix free iterative solvers.
WS Task-3. State of the Art on Saddle-Point algorithms and their Applications in Optimization
Part B – Page 13 of 45
ARTDECO
TrainingTask 1
TrainingTask 2
TrainingTask 3
WSTask1
WSTask2
WSTask3
3 X6 X9 X1215 X18 X2124 X2730 X33 X36 X39424548
Table 2
Part B – Page 14 of 45
ARTDECO
B3 QUALITY/CAPACITY OF THE NETWORK PARTNERSHIP
B.3.1. Existing collaborations
The partners involved in this proposal already have a long history of collaboration amongthemselves and with several laboratories in Europe and in the USA. In order to prove our expertisein the tasks we described in Section B.1.5, we give a list of selected publications where, when it ispossible, the existing collaborations are highlighted (see Table1 for the team number). In thesubsequent text, the publication reference numbers from this list are used for reference.
B.3.2. List of Publications
1. P. Amestoy, I.S. Duff, and C. Voemel, Task Scheduling in an asynchronous distributed multifrontalsolver. SIAM Journal on Matrix Analysis and Applications, Vol 26(2) pp 544565 (2005). (Teams:2, 6, 7).
2. M. Arioli and G. Manzini: A Network Programming Approach in Solving Darcy's Equations byMixed FiniteElement Methods. Electronic Transactions on Numerical Analysis, Special Volume onSaddle Point Problems: Numerical Solution and Applications 22, (2006). (Teams: 2, 4).
3. M. Arioli, D. Loghin, and A. Wathen, 2005, Stopping criteria for iterations in finiteelementmethods. Numer. Math. 99, pp 381410, (Teams: 2, 3).
4. M. Arioli, J. Maryška, M. Rozložník, and M. Tůma: Dual variable methods for mixedhybrid finiteelement approximation of the potential fluid flow problem in porous media. ElectronicTransactions on Numerical Analysis, Special Volume on Saddle Point Problems: NumericalSolution and Applications, 22, (2006). (Teams: 2, 10).
5. F. Auricchio, P. Bisegna, and C. Lovadina, Finite element approximation of piezoelectric plates, Int.J. Numer. Methods Eng, 50, 2001, 14691499. (Teams: 4, 5)
6. F. Auricchio, L. Beirão da Veiga, C. Lovadina, and A. Reali, A Stability Study of some MixedFinite Elements for Large Deformation Elasticity Problems, Comput. Methods Appl. Mech. Engrg.194, 2005, 10751092 (Teams: 4, 5).
7. F. Auricchio, F. Brezzi and C. Lovadina, Mixed Finite Element Methods, in Encyclopedia ofComputational Mechanics, Vol. 1 Chapter 9, Wiley and Sons. Editors: E. Stein, R. de Borst, T.J.R.Hughes, 2004. (Teams: 4, 5).
8. F. Brezzi and M. Fortin, Mixed and hybrid finite element methods, Springer Series inComputational Mathematics, 1991.
9. F. Brezzi, T. J. R. Hughes, D. Marini, A. Russo, and E. Süli. A priori error analysis of a finiteelement method with residualfree bubbles for advection dominated equations. SIAM Journal onNumerical Analysis, 36 (1999) 6, 19331948. (Teams: 3, 4).
10. F. Brezzi, L.D. Marini, and E. Süli. Discontinuous Galerkin methods for firstorder hyperbolicproblems. M3AS: Mathematical Models and Methods in Applied Sciences. 14(12), 18931903,2004, (Teams 3 and 4).
11. F. Brezzi, D. Marini, and E. Süli. Residualfree bubbles for advectiondiffusion problems: thegeneral error analysis. Numerische Mathematik. 85 (2000) 1, 3147, (Teams : 3 and 4).
Part B – Page 15 of 45
ARTDECO
12. F. Bastin, C. Cirillo, and P. L. Toint, Convergence theory for nonconvex stochastic programmingwith an application to mixed logit. To appear in Mathematical Programming B (2006), (Teams: 6and 9).
13. J. B. Caillau, J. Gergaud and J. Noailles, 3D Geosynchronous Transfer of a Satellite:Continuation on the Thrust. Journal of Optimization Theory and Applications, Vol 118(3)(2003), pp 541565.
14. B. Carpentieri, I. S. Duff, and L. Giraud. A class of spectral twolevel preconditioners. SIAMJournal on Scientific Computing, vol. 25, n. 2, pp. 749765, 2003, (Teams: 2, 6,7).
15. A. R. Conn, N. I. M. Gould, and Ph. L. Toint, Trustregion methods'', SIAM/MPS Series onOptimization, SIAM, Philadelphia (2000), ISBN 0898714605, (Teams: 2, 9).
16. F. Courty, A. Dervieux, B. Koobus, and L. Hascoet. Reverse automatic differentiation for optimumdesign: from adjoint state assembly to gradient computation , Optimization Methods and Software,5, 18, 2003, p. 615627.
17. H.S. Dollar, N.I.M. Gould, and A. J. Wathen, On implicitfactorization constraint preconditioners,to appear in proceedings of “Large Scale Nonlinear Optimization”, Erice, Italy, June 2004, KluwerAcademic.
18. I. S. Duff and S. Pralet. Strategies for scaling and pivoting for sparse symmetric indefiniteproblems. To appear in SIMAX. (Teams 2 and 7).
19. H.C. Elman, D.J. Silvester, and A.J. Wathen, Finite Elements and Fast Iterative Solvers: withapplications in incompressible fluid dynamics, OUP, Oxford, 2005.
20. Y.A. Erlangga and C. Vuik and C.W. Oosterlee On a class of preconditioners for solving theHelmholtz equation Appl. Num. Math., 50, pp. 409425, 2004.
21. L. Giraud, J. Langou, M. Rozlozník, and J. van den Eshof. Rounding error analysis of theclassical GramSchmidt orthogonalization process. Numerische Mathematik, 101(1):87100,2005. (Team 6,7, 10)
22. L. Giraud and S. Gratton, On the sensitivity of some spectral preconditioners. To appear in SIAM J.Matrix Analysis (2006), (Teams 6 and 7).
23. N. I. M. Gould, D. Orban, A. Sartenaer, and Ph. L. Toint, Superlinear convergence of primaldualinterior point algorithms for nonlinear programming''. SIAM Journal on Optimization 11 (2001)9741002. (Teams: 2, 9)
24. N. I. M. Gould, C. Sainvitu and Ph. L. Toint, A filtertrustregion method for unconstrainedoptimization, SIAM Journal on Optimization, vol. 16(2), pp. 341357, 2006. (Teams 6 and 9).
25. S. Gratton, A. Sartenaer and Ph. L. Toint, On Recursive Multiscale TrustRegion Algorithms forUnconstrained Minimization, in Oberwolfach Reports: Optimization and Applications, F. Jarre,C. Lemaréchal and J. Zowe, eds., to appear, 2006, (Teams 6 and 9).
26. R. Hauser. "The NesterovTodd direction and its relation to weighted analytic centers''. Found.Comput. Math. 4 (2004), no. 1, 140. Winner of the SIAM Activity Group on Optimization Prize2005.
27. J. van Kan, A. Segal, and F. Vermolen, Numerical Methods in Scientific Computing First edition2005, VSSD, Delft, The Netherlands. ISBN 9071301508, NUR 956. (Teams 11 and 12)
Part B – Page 16 of 45
ARTDECO
28. C. Keller, N. I. M. Gould, and A. Wathen, Constraint preconditioning for indefinite linear systems,SIAM J. Matrix Anal. Appl. 21, 2000, pp. 13001317, (Teams 2 and 3).
29. D. Loghin, M. van Gijzen, and E. Jonkers, Bounds on the eigenvalue range and on the Field ofValues of nonHermitian and indefinite finite element matrices, Journal of Computational andApplied Mathematics, Vol 189/12, 304323, 2006, (Teams 6 and 11).
30. M. Michieli de Vitturi, F. Beux, G. Lombardi, and A. Dervieux. Optimum shape design forturbulent viscous flows around complete configurations of 2D flying sails. Journal ofComputational Methods in Sciences and Engineering, volume 4, numbers 12 , 4355, 2004.
31. J. Nedic and R. Hauser, The continuous NewtonRaphson method can look ahead". SIAM J. Optim.15 (2005), no. 3, 915925.
32. S.P. van der Pijl, A. Segal, C. Vuik, and P. Wesseling, A massconserving LevelSet method formodelling of multiphase flows, International Journal for Numerical Methods in Fluids, 47, 339361, 2005 (Teams 11 and 12).
33. M. Rozložník and V. Simoncini. Krylov subspace methods for saddle point problems withindefinite preconditioning, SIAM J. Matrix Anal. and Appl. (2002), Vol. 24, No. 2, pp. 368391.(Teams: 4, 10).
34. E. Schall, D. Leservoisier, A. Dervieux, and B. Koobus. Mesh adaptation as a tool for certifiedcomputational aerodynamics. I. J. Num. Meth. Fluids 2004; 45:179196
35. M. Vazquez, B. Koobus, and A. Dervieux. Multilevel optimisation of a supersonic aircraft . FiniteElement in Analysis and Design, Vol.40, 21012124, 2004.
36. M. Vazquez, A. Dervieux , B. Koobus, A methodology for the shape optimization of flexible wings,to be published by Eng. Comp. 2005.
37. F.N. van de Vosse, J. de Hart, C.H.G.A. van Oijen, D. Bessems, T.W.M. Gunther, A. Segal,B.J.B.M. Wolters, J.M.A. Stijnen and F.P.T. Baaijens, Finite Element Based ComputationalMethods for Cardiovascular FluidStructure Interaction J. Eng. Math. 47, 335368, 2003. (Teams 11and 12).
38. P. Wesseling. An Introduction to Multigrid Methods. John Wiley & Son, Philadelphia, 2004(corrected reprint).
Part B – Page 17 of 45
ARTDECO
Connectivity graph of the teams
Part B – Page 18 of 45
ARTDECO
B.3.3 Individual contributions
Team 1 ERCIM EEIG
The European Consortium for Informatics and Mathematics (ERCIM) will be the coordinator ofARTDECO. ERCIM is an EEIG composed of seventeen different national research organisations inas many European countries. GEIE ERCIM members are all research organisations, with strongactivity in I.T. research. Members are (in alphabetical order): AARIT (Austria), CCLRC (U.K),CRCIM (Czech Republic), CWI (The Netherlands), CNR (Italy), FNR (Luxembourg), FORTH(Greece), FhG (Germany), INRIA (France), NTNU (Norway), SARIT (Switzerland), SICS(Sweden), SPARCIM (Spain), FNRS (Belgium), SZTAKI (Hungary), Irish Universities Consortium(Ireland), VTT (Finland). ERCIM has a leading expertise in the coordination of FP4, FP6 and FP6European Projects.ERCIM’s central Office acts as a frontend to access to support the scientific teams of its members(for example, the Office is running the ERCIM Fellowship Programme). The Office has set upstrong internal and external channels for coordination, solicitation and dissemination. Members of the team and their fields of expertise:Bruno Le Dantec (Deputy Manager): Project coordination, European contract regulations,administrative and financial supervision of projects, lawRemi Ronchaud (Project Manager): Project coordination, management, training anddissemination organisation, international businessEmma Liere (Assistant Manager): Administration, Fellowship programme managerCeline Bitoune (Assistant Manager): administration, financial coordinationIn addition, The ERCIM Office team will be completed by the ERCIM HRM task force, dedicatedto coordinate and manage the training and dissemination activities within the EEIG and across thedifferent ERCIM member institutes. This dedicated team will complement the administrative andfinancial coordination of the Office with expertise in education, and training, to provide an efficientlearning frame to the student while implementing all the necessary measures to ensure their lastingintegration into the scientific community (career development programme).Partner 1 Contribution to WP3 (Dissemination and Management) tasks:Task3.1 – Internal dissemination among Workpackages, teams, students and managementTask-3.2 - External dissemination to promote ARTDECO to students, ICT stakeholders and thescientific community at large.Task-3.3 – Administrative coordination, supervision of the project and management of the
ARTDECO proposal as a whole.
Task-3.4 – Financial coordination, transfer towards ARTDECO participants, grant payment to
students and supervision of the allocation of resources.
Team 2 CCLRC RAL CSE Numerical Analysis Group
The Numerical Analysis Group is based at the Rutherford Appleton Laboratory and is part of the
Part B – Page 19 of 45
ARTDECO
Computational Science and Engineering Department of CCLRC. The CSE Department is in chargeof the High Performance Computing activity and runs the HPCx system based on IBM’s p690highend compute nodes (1280 nodes) with a peak performance of 7.7 TeraFlop/s (3.5 Tera Flop/ssustained). The team will grant access to the ARTDECO partners to our software libraries HSL andGALAHAD (http://www.cse.clrc.ac.uk/nag/hsl/hsl.shtml ,http://www.cse.clrc.ac.uk/nag/galahad/galahad.shtml ) and to the HPCx facility.
Members of the team and their fields of expertise:
Dr. Mario Arioli (senior researcher): error analysis, iterative solvers, preconditioning, and
stopping criteria applied to the numerical solution of PDEs.
Prof. Iain S. Duff (team leader): Sparse direct solvers for linear systems and parallel computing
Prof. Nicholas I. Gould (senior researcher): Optimization theory and algorithms.
Dr. Jennifer Scott (senior researcher): Sparse direct solvers for linear systems and sparse
eigenvalue problemsGroup members have been heavily involved in the Institute of Mathematics and its Applications,the IMA Journal of Numerical Analysis, SIAM organisation, SIAM J. of Optimization, theMathematics College of EPSRC, the organization of national and international conferences,refereeing and editing, seminars and teaching at universities and summer schools, supervision andexamination of research students, writing books and papers, developing mathematical software, andadvising UK and overseas governments on high performance computing. The group membersactively collaborate with several of the teams involved in the project: University of Namur, IMATI-CNR, CERFACS, IRIT-INPT, the Institute of Computer Science of the Academy of Sciences of theCzech Republic, and the University of Oxford.
Partner 2 contributions to the WP1 and WP2 tasks:
Task-1.1 Direct solvers for saddle point linear systems including an implementation of an out-of-
core Gaussian decomposition.
Task-1.2 Spectral preconditioning techniques for iterative Krylov and Chebyshev methods.
Numerical stability of Krylov methods. Algebraic preconditioning.
Task-1.3 Analysis of null and range space algorithms for quadratic linearly constrained problems.
Task-1.5 Identification of stopping criteria for iterative methods applied to the solution of the
finite-element approximation of PDEs (in collaboration with Team 8).
Task2.1 Implementation of interior point algorithms and use of GALAHAD.Publications: See [1], [2], [3], [4], [14], [15], [17], [18], [23] and [24] in Section B.3.2
Team 3 Oxford University Computing Laboratory, Numerical Analysis Group
The Numerical Analysis Group in the Computing Laboratory (the Oxford University ComputerScience Department) has one of the larger groups of active Numerical Analysis researchers in theUK. There are currently 8 permanent members of staff and many research students, postdocs andvisitors. Together with the Oxford Centre for Industrial and Applied Mathematics (which is in theMathematics Department), the Group runs two Master of Science courses in “MathematicalModelling and Scientific Computing” and “Applied and Computational Mathematics”. These
Part B – Page 20 of 45
ARTDECO
courses which are supported by significant funds from the Engineering and Physical SciencesResearch Council (EPSRC) and attract extremely well qualified students from Britian, manycountries of the European Union, North America, and more occasionally from Asia and Africaalso. Currently there are 30 students on these courses. The Group currently has 21 PhD students.Members of the team and their fields of expertise:The research interests of the group cover most of the areas of Numerical Analysis: DifferentialEquations, Linear Algebra, Optimization and Approximation Theory.Professor Nick Trefethen has broad interests in Eigenvalue problems, nonnormality, spectralmethods and Complex approximation.Professor Mike Giles heads the University Technology Centre in Computational Fluid Dynamicsand is an Associate Director of the Oxford eScience Centre. He has research interests also inComputational Finance.Dr Raphael Hauser is an expert in Optimization. Dr Ian Sobey works on numerical techniques for Incompressible Fluid Flow problems.Professor Endre Süli is an authority on the analysis of numerical approximation methods fordifferential equations, in particular finite element methods.Dr Andy Wathen works on Numerical Linear Algebra and methods for Partial DifferentialEquations, in particular Iterative methods and Preconditioning. The various members of the Numerical Analysis Group have significant research experience andexperience of organizing and collaborating research. They have significant involvement with theSociety for Industrial and Applied Mathematics (SIAM), and amongst the group a large number ofeditorial and advisory roles. Being teaching staff in a large University, to varying degrees themembers have significant to less significant teaching responsibilities. The Group has many researchvisitors from all over the world with stays varying from a day to a year. These have included inrecent years members of Team 4. Many oneday workshops and meeting are organized in Oxfordby members of the group, with roughly 3 or 4 in any one year. Different members of the Group have collaborative research projects with groups in France,Sweden, Germany, Italy, Spain as well as other in the UK within the European context. There areseveral current research grants of varying size, those associated with eScience and Life Sciencesbeing the most significant.Partner 3 contribution to the WP1 and WP2 tasks:Task 1.2: preconditioning and iterative solvers for saddlepoint problems. Block preconditioningand Constraint preconditioning.Task 1.3: direct factorizations in preconditioning, Schilders Factorization in preconditioning.Task 1.4: incompressible flow applications. oil reservoir applications.Task 1.5: natural norms for PDEs and their linear algebra analogues.Task 2.1: use of Schilders factorization in PDEs constrained optimization problems.Task 2.3: PDE constrained optimizationTask 2.4: applications in flow control.
Part B – Page 21 of 45
ARTDECO
Publications: See [3], [9], [10], [11], [17], [19], [26], [28], and [31] in Section B.3.2
Team 4 IMATI CNR, Pavia, Italy
Partner 4 (Instituto Matematica Applicata e Tecnologie Informatiche Consiglio Nazionale delleRicerche) was founded in 2000 and started its research activities in 2002. The current structure ofthe Institute results from merging three existing Institutes, which are well known in the field ofMathematics and its Applications: IAMI (Institute for the Application of Mathematics andComputer Science) in Milan, IAN (Institute of Numerical Analysis) in Pavia, and IMA (Institute ofApplied Mathematics) in Genoa. Research is mainly carried out in the field of differential modeling,which is considered from different points of view: theoretical (existence, uniqueness and stability ofthe solutions), numerical (approximation schemes, stability and adaptability), and computational(algorithms and computing methodologies). The result of these basic research programs are widelyused to deal with application problems coming from, for instance, fluiddynamics, analysis ofelectromagnetic and semiconductor devices, and the study of the elastic properties of materials.Members of the team and their fields of expertise:Prof. Franco Brezzi (full professor): is the director of IMATI and coordinator of the PhD programof the Institute of Advanced Studies of the University of Pavia. His scientific activities are devotedto the studies of finite element methods for Partial Differential Equations (PDEs).Dr. Gianmarco Manzini (researcher): finite volume methods for PDEs and porous mediaapplications.Dr. Micol Pennacchio (researcher): Numerical linear algebra with particular attention to the studyand development of Krylov and substructuring preconditioners for the discretization of cardiacmodels.Dr. Sergio Rovida (researcher): development and implementation of High Performance Computing(HPC) techniques. Dr. Gianni Sacchi (research director): finite element methods for PDEs with particular attention toimplementation issues in HPC environments.Partner 4 contributions to the WP4:Tasks 1.21.4: Development and analysis of efficient and reliable finite element techniques forstructural and fluid problems with attention to fluid dynamics models for porous media.Implementation and optimization of these methods in HPC environments.Publications: see [2], [7], [8], [9], [10], and [11] in Section B.3.2.
Team 5 University of Pavia, Departments of Mathematics and Structural Mechanics
Partner 5 (Department of Mathematics and Department of Structural Mechanics – University ofPavia) has a recognized experience in the modelling, analysis and numerical simulation of variousengineering problems. In particular, several problems in mixed form have been extensively studiedby the Team members. A nonexhaustive list comprehends: nearly incompressible elasticity, plateand shell problems, Stokes problem, Darcy's flows. Recently, a special attention has been paid tothe possibility of proposing robust numerical schemes to be used in the solution of large scale
Part B – Page 22 of 45
ARTDECO
problems and to the applicability of standard theories to the modeling of living soft tissues (such asarteries).Members of the team and their fields of expertise:Prof. Ferdinando Auricchio (full professor): Modelling and numerical simulation of ShapeMemory Alloys and elastoplastic materials.Prof. Carlo Lovadina (associate professor): Finite element analysis of problems involving thinstructures, such as plates and shells. Aposteriori error analysis of elliptic problems in mixed form. Dr. Alessandro Reali (postdoc student): Modelling and numerical simulation of Shape MemoryAlloys. Nonlinear Finite element analysis.Prof. Auricchio is the coordinator of GIMC (Italian Group of Computational Mechanics);moreover, he has been the coordinator of two Italian research projects on Shape Memory Alloysand biomedical applications: COFIN2002 and COFIN2004.Prof. Auricchio and Prof. Lovadina are involved in the “Rose School” of Pavia, a European Schoolfor advanced studies in reduction of seismic risk. Partner 5 contributions to the WP1:Tasks 1.31.4: Development and analysis of efficient and reliable finite element techniques forStructural Mechanics problems in mixed form: plates and shells, and large deformation elasticityproblems. Modelling and numerical simulation of innovative materials, to be used for the design ofhitech medical devices.Publications: see [5], [6], [7] in Section B.3.2.
Team 6 Parallel Algorithms team of CERFACS
The Parallel Algorithms team of CERFACS is concerned both with underlying mathematical andcomputational science research, the development of new techniques and algorithms, and theirimplementation on a range of high performance computing platforms. Members of the team playtheir full part in the wider academic and research community. They are involved inProgramme Committees for major conferences, are editors and referees for frontline journals,and are involved in research and evaluation committees. At the level of international efforts forstandards in numerical linear algebra, the team has been involved in the development of a newstandard for the Linear Algebra Subprograms. The team is also distributing popular direct anditerative solvers for linear systems (MUMPS, Krylov methods), in collaboration with ENSEEIHTIRIT and INRIA. Members of the team and their fields of expertise :Prof. Iain Duff is the Project Leader of the Parallel Algorithms Team at CERFACS inToulouse. He is a leading expert in parallel computing and in sparse solvers for linear systems. Prof. Francoise ChaitinChatelin is Professor at the Université de Toulouse I and thehead of the Qualitative Computing Group hosted in the Parallel Algorithms Team at CERFACS.She is a world expert in the reliability of finite precision computations. Dr. Serge Gratton is a senior researcher in charge of the data assimilation and optimizationresearch. Application areas of his work include atmosphere and ocean data assimilation as well as
Part B – Page 23 of 45
ARTDECO
gravity field computation.Dr. Fabian Bastin is a postdoctoral researcher. His scientific interests include nonlinearprogramming, stochastic optimization and discrete choice modeling Dr. Xavier Vasseur is a postdoctoral researcher. He has worked on parallel algorithm for thesolution of PDE's. Application of his work can be found in computational fluid dynamics andelectromagnetics.Marc Baboulin (PhD student) is working on the solution of large inverse problems. MelodieMouffe (Phd student) is working on multigrid techniques in optimization. Azzam Haidar (Phdstudent) is working on hybrid solvers for PDE's and Tzvetomila Slavova (Phd student) is involvedin the outofcore solution of sparse linear systems.Partner 6 contributions to the WP1 and WP2 tasks:Task1.1 Optimization of the performance of outofcore solvers by using task schedulingtechniques.Task1.2 Development of blockKrylov techniques. Study of inexact Krylov methods wherematrixvector products are approximate. Use of spectral preconditioner to accelerate block methods.Task1.3 Combination of direct and iterative solution method for the solution of PDE's usingdomain decomposition techniques for 3D problems.Task1.4 Solution of complex symmetric systems arising form the discretization of wavepropagation problems (electromagnetics and acoustics).Task2.2 Solution of bound constrained optimization problems using trust region techniques. Publications: see [1], [12], [14], [21], [22], and [25] in Section B.3.2.
Team 7 IRITENSEEIHTINPT “Algorithmes Parallèles and Optimisation”
The Parallel Algorithms and Optimization at ENSEEIHTIRIT has a strong experience in the designof numerical software, in highperformance computing, and more recently in grid computing. Themain research areas are : Computational kernels for linear algebra and nonlinear optimization,Optimal Control, Sparse direct linear solvers, Iterative linear solvers, PDES and domaindecomposition, Parallel computing and Grid computing.The team has been involved in the design of several software packages that are available to thescientific community. Included among these are: • Iterative solvers (use of block conjugate gradient, elementbyelement preconditioners, row
projection techniques, ...), Implementation of Krylov packages• MUMPS is a public domain package, based on public domain software developed during the
Esprit IV European project PARASOL (19961999) by CERFACS, ENSEEIHTIRIT and RAL.Since this first public domain version in 1999, the developments are supported by the followinginstitutions: CERFACS, ENSEEIHTIRIT, and INRIA (see http://www.enseeiht.fr/apo/MUMPSor http://graal.enslyon.fr/MUMPS).
• The codes MA41 (LU factorisation), and MA49 (QR factorisation) for shared memorymultiprocessors available on the Web and/or in HSL (formerly the Harwell Subroutine Library).
• The optimal control software TfMin: Minimum Time Orbit Transfer Computation and MfMax:
Part B – Page 24 of 45
ARTDECO
Maximum Final Mass Orbit Transfer Computation The team is also involved in two majors grid projects funded by the French Minister of Research:the GridTLSE project that aims at designing a Web expert site for sparse linear algebra and theGRID'5000 wide experimental Platform for Grid Computing.Members of the team involved in the Project and their fields of expertise:Prof. Patrick Amestoy: Sparse direct solvers, grid and parallel computing.JeanBaptiste Caillau (senior lecturer): Optimization and optimal control of systems governed byordinary differential equations with applications to Celestial Mechanics.Prof. Michel Daydé: Computational kernels, grid and parallel computing.Joseph Gergaud (senior lecturer): Optimization and optimal control of systems governed byordinary differential equations with applications to Celestial Mechanics.Prof. Luc Giraud: Numerical linear algebra, domain decomposition, iterative and hybrid methods,high performance computing and grid computing.Ronan Guivarc'h (senior lecturer):.Parallel numerical computation and distributed computation,Parallelization of Scientific Software.Prof. Joseph Noailles: Optimization and optimal control.Dr. Daniel Ruiz (senior lecturer): Iterative solvers and preconditioning.Team members have been heavily involved in the organization of national and internationalconferences, refereeing and editing, seminars and teaching at universities and summer schools,supervision and examination of research students, developing mathematical software, and involvedin reviewing projects on high performance computing and grid computing both at national andinternational levels. The team members actively collaborate with several of the teams involved inthe project: University of Namur, CERFACS, and RAL. Partner 7 contributions to the WP1 and WP2 tasks:Task1.1 Parallel distributed implementation of scalable sparse direct techniques with outofcorefeatures and facilities for tight coupling with hybrid approaches, new ordering techniques.Task1.2 Spectral preconditioning techniques for iterative Krylov and Chebyshev methods.Algebraic preconditioning. Block Krylov technique.Task1.3 Study of parallel scalable hybrid direct/iterative linear solvers suited for the solution oflarge 3D problems on large clusters of SMPs; extend to general sparse linear systems some of thenumerical approaches developed in domain decomposition.Task1.4 Applications could include Computational Fluid Dynamics, and NavierStokes equations,wave propagation problems, 3D structural analysis problems. Task2.4 Applications: use of optimization techniques for optimal control in ODE and applicationto orbital transfer.Publications: See [1], [12], [13], [14], [18], [21], and [22] in Section B.3.2
Team 8 INRIA SophiaAntipolis
Partner 8 is the Tropics team of INRIA at SophiaAntipolis. It has a long experience of
Part B – Page 25 of 45
ARTDECO
Computational Fluid Dynamics and Optimal shape design in CFD, and in AutomatedDifferentiation. Recent contributions are related to Reverse or adjoint Automatic Differentiation ofprograms, and to the resolution of saddle points systems formulated as minima with equalityconstraint and resulting of adjoint based automated sensitivity. These systems are complex toassemble and since they involve the exact sensitivity of complex nonlinear approximations(Riemann solvers, local reconstructions, TVD limiters in CFD)), they are more difficult to solvethan usual approximate linearisation.Members of the team and their fields of expertise:Dr Laurent Hascoet (hab.) is head of the Tropics team, specialist of Automatic Differentiation. Dr Alain Dervieux (hab.) is Senior Scientist, specialist in CFD and in Optimal Shape Design,advisor of 30 PhD. Partner 8 contributions to the tasks:Task 1.2: Iterative solvers: Iterative matrixfree solvers are generally compulsory for adjointbasedOptimal Control. Multilevel additive preconditioners [30, 35] can be adapted to complex saddlepoint research on unstructured meshes thanks to the volumeagglomeration method. We shall applythis to a saddle point system related to mesh adaptation.Task 1.4: Application: application to Computational Fluid Dynamics: optimal meshes for SonicBoom prediction as a sequel of [35, 36]Task 1.5: Aposteriori error estimator and stopping criteria: the introduction of Partner 2 (RAL)methods into the optimisation loop will be studied in cooperation with Partner 2.Task 2.2: Multiscale optimization: this is covered by the work planned in Task 1.2.Task 2.3: Adaptation of optimization algorithms to optimal design: Partner 8 will work on stoppingcriteria related to the quality of descent directions in relation with Task 1.5.Task 2.4: Applications: in shape optimization in aerodynamics. Partner 8 will experiment the newalgorithms developed in collaboration with partners of Task 1.5 on a set of difficult systems arisingin sonic boom reduction for 3D aircraft shapes.Task 2.5: Sensitivity analysis [16, 36]: This is the main contribution of Partner 8 to ARTDECO:Adjoint methods and applications, in cooperation with Partner 3.. The Automatic Differentiator (AD)TAPENADE is usable and downloadable from Partner 8's web site:http://wwwsop.inria.fr/tropics/ where additional documentation and references can be found.Partner 8 will be responsible of coordination of Task 2.5, Sensitivity Analysis. Partner 8 will focuson the derivation of adjoint and optimality systems for optimal control or optimal design problems.A workshop for the application of AD tools will be organised by Partner 8 during the first year ofARTDECO. A second workshop will be organised by Partner 8 at year 2 of ARTDECO and willfocus on matrixfree iterative solvers for adjoint systems produced by AD or direct development.Publications: see [16], [30], [34], [35], and [36] in Section B.3.2 .
Team 9 FUNDP, Department of Mathematics, Numerical Analysis Unit, Namur
The Numerical Analysis Unit is part of a department of applied mathematics. The research group
Part B – Page 26 of 45
ARTDECO
has a long experience in numerical optimization, both from the theoretical and applied point ofview. It has invested in the past in the whole range from algorithm development and design toapplication in fields as diverse as surface design in applied optics, econometric estimation, trafficand transportation engineering. It has taken part in several project on numerical optimization, suchas CUTE(r), LANCELOT and GALAHAD (all three in collaboration with CCLRC RAL), PSDFO(partially separable derivative free optimization and others. Its current work include an ongoingproject with CERFACS and the Czech Academy of Sciences on multilevel optimization algorithms.Members of the team and their field of expertise:
Prof. Philippe Toint (director): theory and algorithms for numerical optimization, derivative freemethods, scientific software, traffic engineering Prof. Annick Sartenaer (senior researcher): numerical optimization, applications of optimizationin industry. The group also currently include 6 PhD students (M. Weber, E. Wanufelle, K. Demaseure, C.Sainvitu, D. Tomanos, J. TshimangaIlunga, and M. Mouffe (joint with CERFACSINPT)), ofwhich 2 are nearly finished. A. Sartenaer and Ph. Toint have been involved in national andinternational institutions, such as the Belgian Mathematical Society, the SIAM Journal onOptimization, the Quarterly Journal of Operations research, Mathematical Programming,Operations Research, the SIAM Activity Group on Optimization, the SMAI, IFIP TC7, or theFoundation for Computational Mathematics (FoCM). They have also taken part in severalinternational scientific committees (CERFACS, KUL Center for Optimization, Matheon Berlin, ...) Collaboration is currently ongoing with several of the teams involved in the project includingCCLRC RAL, CERFACS and INPT. Partner 9 contributions to the WP1 and WP2 tasks: Task 1.2: Krylov subspace methods, linear multigrid algorithms. Task 1.3: Largescale optimization applications (optics, variational surfaces) Task 2.1: Interior point techniques and filter methods, preconditioning. Task 2.2: New multilevel/multigrid methods in optimization and optimal control (theory, softwareand applications) Task 2.3: Multilevel approach to optimal control problems described by PDEs. Publications: See [12], [23], [24], and [25] in Section B.3.2.
Team 10 Academy of Sciences of the CR, ICS, Department of Computational methods
The Numerical Linear Algebra and Optimization group is a part of the Department ofComputational Methods of the ICS AS CR. Its research effort is focused on theory of convergenceand numerical behaviour of Krylov subspace methods; development, analysis, implementation andapplication of scalable algorithms for solving extremely large generally sparse linear algebraicsystems including preconditioning and implementation on parallel computer systems; mathematicalfoundations and numerical methods in errorinvariables modelling; development and analysis ofmethods in numerical optimization. Members of the team and their fields of expertise:
Part B – Page 27 of 45
ARTDECO
Prof. Ladislav Lukšan (senior researcher): Nonlinear (smooth and non smooth) optimization,quasiNewton methods, interiorpoint methods.Dr. Miroslav Rozložník (head of the department): Numerical stability analysis, saddlepointiterative solvers, preconditioning.Prof. Zdeněk Strakoš (team leader): Theory of convergence of Krylov subspace methods, erroranalysis, stopping criteria, numerical methods in errorsinvariables modelling. Dr. Petr Tichý (junior researcher): Convergence of Krylov subspace methods,error analysis, stopping criteria.Prof. Miroslav Tůma (senior researcher): Preconditioning of iterative methods, theory andimplementations. Applications in preconditioning of saddlepoint problems arising from mixedfinite elements. In addition to research, publications and presentations of results, editorial activities (SIAM J.Matrix Anal. Appl., Electronic Trans. Num. Anal., Linear Algebra Appl., BIT Numer. Math.),involvement in the organizing activity important for the numerical linear algebra communityworldwide (Householder Symposium), the team members regularly teach and supervise PhDstudents at several Czech universities. They organize yearly winter schools within the projectMSTEP of the National Program of Research, see http://www.cs.cas.cz/~mweb. With the five yearperiod the team organizes the high profile conference with the focus significantly overlapping withthe subject of this proposal, and with the regular participation of members of many teams involvedin the proposed project, see http://www.cs.cas.cz/~harrachov. The group members collaborate withseveral of the teams (Team 2, Team 4, Team 6), and also with several leading research groupsoutside Europe (McGill University, University of Maryland, Emory University etc.) Partner 10 contribution to the WP1 and WP2 tasks:Task 1.3: Preconditioning of saddlepoint problems combining combinatorial and numerical toolsin mixed direct/iterative approach.Task 1.4: Applications of algorithms to Darcy's law. Applications in numerical optimization (incollaboration with Team 2)Task 1.5: Analysis and stopping criteria in Krylov space methods (in collaboration with Team 2)Task 2.1: Interiorpoint methods in constrained optimization.Publications: See [4], [21], and [33] in Section B.3.2.
Team 11 Delft University of Technology
The Numerical Analysis group of the Institute of Applied Mathematics of the Delft University ofTechnology consists of 9 permanent staff members and presently hosts 8 PhDstudents and onepostdoctoral researcher. The head of the group is Professor Piet Wesseling, who is a worldspecialist in Computational Fluid Dynamics. The research of the group focuses on advanceddiscretisation techniques for PDE's and on fast solvers for linear and nonlinear systems. Theresearch has a strong emphasis on applications. All PhDs are funded by external money. There areongoing collaborations with industry, among others with Philips and Shell.Members of the team and their fields of expertise:
Part B – Page 28 of 45
ARTDECO
Prof. dr. ir. Piet Wesseling is head of the Numerical Analysis group. He is a world specialist inComputational Fluid Dynamics and was among the first researchers interested in the multigridsolution method. Professor Wesseling has been the advisor of 30 completed PhD theses. He ischairman of the Delft Center for Computational Science, member of the board of the J.M.BurgersCenter and chairman of the Computational Fluid Dynamics Panel of the European Community onComputational Methods in Applied Sciences (ECCOMAS).Dr. ir. Kees Oosterlee is Associate Professor. His research concentrates on fast iterative solutionmethods in science and engineering, in particular multigrid methods, with application among otherthings in financial mathematics. Dr. ir. Kees Vuik is Associate Professor. His research concerns preconditioned Krylov methods forsolving linear systems of equations and the numerical solution of free and moving boundaryproblems.Dr. ir. Martin van Gijzen is Assistant Professor. His research focuses on preconditioned Krylovsolvers and Finite Element techniques and on parallel and grid computing. His application areas ofinterest include acoustics and ocean circulation.Dr. ir. Fred Vermolen is Assistant Professor. He is a specialist in free and moving boundaryproblems with application to among others wound healing and homogenization of alloys.Partner 11 contributions to the tasks:Task.1.2. Use of multigrid techniques in saddle point algorithms.Task.1.5. Stopping criteria and multigridTask.2.1. Numerical solution of free and moving boundary problems.Task.2.2. Use of multigrid/multilevel techniques in solving PDEs.The group is member of the Research School J.M.Burgers Center, which is the National ResearchSchool for Fluid Dynamics. PhDstudents of the group are encouraged to take part in the PhDcourses that are organised by the Burgers Center, which are in part given by the members of thegroup. The group belongs to the Center of Excellence 'Multiscale Phenomena in Fluids and Solids'of the three Universities of Technology in the Netherlands. The group is also member of the DelftCenter for Computational Science. This centre offers additional PhDcourses and provides aplatform for collaboration with the other groups in the centre at all levels. Members of the groupare active in the organisation of (international) scientific conferences. For example, in 2005 thegroup organised the European Multigrid Conference and in 2006 the group has the lead in theorganisation of the ECCOMAS CFD 2006 Conference. All permanent staff members of the group(co)supervise PhD and Master students on a daily basis. Publications: see [27], [29], [32], [37] and [38] in Section B.3.2 .
Team 12 SEPRA
SEPRA is a small consultancy and numerical software company that works for both the Dutchgovernment and large industrial companies. It develops and maintains the Finite Element packageSEPRAN (see http://ta.twi.tudelft.nl/sepran/sepran.html) that is used at more than ten universities inthe Netherlands, Germany and France, and at a similar number of international companies. At the
Part B – Page 29 of 45
ARTDECO
TUDelft SEPRAN is also used for educational purposes in a number of advanced numericalanalysis courses.Members of the team and their fields of expertise:Dr. Guus Segal: cofounder and codirector of SEPRA. He is a specialist in the Finite Elementmethod, and development of numerical software. He also holds a part time position as AssociateProfessor at the TUDelft where he is responsible for an advanced master class on the FiniteElement method. He is also responsible for a PhD course on CFD. Guus Segal has supervised alarge number of master students and cosupervised more than ten PhDstudentsDr. Niek Praagman: the other cofounder and codirector of SEPRA. He is a specialist in theFinite Element method and in numerical software development. Both Niek Praagman and Guus Segal are active in consultancy for industry.Partner 12 contributions to the tasks:Task.1.3 and Task.2.2 SEPRA is highly interested in new techniques for solving saddle pointproblems faster to enable the simulation of larger and more complicated problems, and thusyielding more accurate predictions. Faster methods for solving saddlepoint problems will alsoenable parameter optimization procedures, which involves the subsequent solution of a largenumber of saddlepoint problems. This capability does not yet exist in SEPRAN.Task.1.4 SEPRAN is a numerical software package for solving Stokes and NavierStokes flowproblems involving free and moving surfaces. A typical application is the simulation of mixing offluids for manufacturing of glass. Another application is the simulation of blood flow through aheart, as part of the acceptance procedure of artificial valves. The above problems all lead to large saddlepoint problems of which the solution is a bottle neckfor large 3D problems. An additional complication that arises when the method of fictitiousdomains is used to model the free boundary is that in order to couple the fluid with the structureadditional constrains have to be imposed using a Lagrange multiplier technique, which yields asaddle point problem that can be very ill conditioned.SEPRA will contribute to the training program of the young researchers in ARTDECO byproviding up to three months long internalships during which the young researchers apply thetechniques that are developed within ARTDECO to relevant applications such as the two mentionedabove. Additional training will be provided in collaboration with the TUDelft. For example, theyoung researchers will be encouraged to participate in the PhDcourses that are organized by theResearch School for Fluid Mechanics J.M. Burgerscenter. Publications: see [27], [29], [32] and [37] in Section B.3.2 .
Part B – Page 30 of 45
ARTDECO
B4. MANAGEMENT AND FEASIBILITY
B4.1 Proposed management and organisational structure
The partners involved in this proposal already form a multidisciplinary consortium relying onteams assigned to bring their expertise to the training network and to the students. While theresearch teams involved will focus on the scientific and training agenda within ARTDECO, ERCIMwill ensure the administrative and financial coordination of the network, with the support of thededicated management structure. The general management structure of ARTDECO can be outlinedas follows:
Part B – Page 31 of 45
WP1 Saddle point problems•Task 1.1 Direct methods•Task 1.2 Iterative solvers•Task 1.3 Mixed approach•Task 1.4 Applications•Task 1.5 Aposteriori error estimator and stopping criteria
WP2 Optimal design and shape optimization
•Task 2.1 Algorithms for elliptic control problems•Task 2.2 Multiscale optimization•Task 2.3 Adaptation of algorithms to optimal design•Task 2.4 Applications•Task 2.5 Sensitivity analysis
WP3 Dissemination and Management•Task 3.1 Internal Dissemination•Task 3.2 External dissemination•Task 3.3 Administrative coordination•Task 3.4 Financial coordination
Research & Training coordination Board
ARTDECO
European CommissionEuropean Commission
International IST
Community
HRM Office
Executive Task force
Academy and Industry
ARTDECO
The ARTDECO management organisation will rely on a simple and coherent architecture organisedaround three core workpackages. To support this simple architecture, the project has adopted alight but efficient management scheme.WP1 and WP2 will address the respective research and training agenda, backed up by the ResearchTraining Coordination Board (RTCB) acting as the cornerstone of all of the research and trainingactivities.
WP3 will be dedicated to the coordination and dissemination efforts to be conducted within andoutside the network. With the support of the HRM Office, the WP3 will also ensure the generalmanagement of the project, and will perform the administrative and financial coordination.
The different organisations and actors composing the ARTDECO management are closelyintertwined to allow WP1, WP3 and WP3 to complete their respective mission. Their descriptionand interrelations as described hereafter:
• Research and Training Coordination Board (RTCB)• Workpackages• Scientific coordinator• Administrative and Financial Coordinator
Research and Training Coordination Board (RTCB)Composition: The RTCB is composed of the scientific coordinator, Dr Mario Arioli, who alsochairs the ERCIM Working Group “Applications of Numerical Mathematics in Science”, and of arepresentative of each partner institute involved in ARTDECO. The scientific coordinator chairs theboard and all decisions are taken by a majority vote. The RTCB will consider also the possibility ofinviting an external member of international outstanding mathematical prestige and educationalexperience from outside the network and the EU as an advisor on international cooperation withnonEuropean institutions.Objectives: The RTCB will supervise and stimulate the overall research and training activitieswhile fostering exchanges and interactions among the scientific WorkPackages. In particular, theRTCB will coordinate the activities of the PhD students with special attention to the legal andinstitutional aspects concerning the fulfilment of their duties visavis the European Universityresponsible for awarding the title of Doctor of Philosophy. Moreover, when legally possible, wewill encourage the recognition of the PhD title by another European University participating in theNetwork and involved in the training of the candidate. The RTCB will define the student selectionrequirements for all partners and cooperate with the dedicated HRM Office in charge ofimplementing the selection and integration of young researchers. The RTCB will also discuss themultidisciplinary scientific orientation and will assess the quality of the network’s achievements asan internal peer review committee. Tasks: The major role of the RTCB will be to identify, define and stimulate initiatives that spanacross several areas, to select common research activities and dissemination initiatives such as
Part B – Page 32 of 45
ARTDECO
summer schools and the financial support for the young researchers to attend internationalconferences. Moreover, we must be prepared to foresee the possibility of critical situationsconcerning the fellowships and, thus, the RTCB will also take care of scientific arbitration andadvice to the young researchers. Finally, the RTCB will perform an internal assessment of theARTDECO achievements. In addition to the monitoring of activities, the underlying objective is toidentify successes and best practices in the project and to share these with the different teamsinvolved. The internal dissemination undertaken in WP3 will come as a direct support to thisparticular task.Means: The RTCB members will participate in biannual meetings organised to assess the overallachievements of the network. The board will also rely extensively on audioconferencing (once amonth) and use a dedicated mailing list to ensure regular communication. Board members willperform an overview of the WP activities and the Board will provide general recommendations toimprove both training and research. Decisions will be taken by majority vote, with one vote perpartner, plus an additional casting vote for the coordinator if necessary. To implement its decisionsin the project, the RTCB will rely on the Executive Task Force. This task force is composed of thethree Workpackage Leaders and will be in charge of implementing the board’s decisions and willrelay these directives to the different tasks that comprise every WorkPackage.Assessment: It is expected that the RTCB will monitor its own activity. Nevertheless, the projectcoordinator will review the RTCB members’ commitment and the relevance of theirrecommendations.WorkPackages : Composition: A WorkPackage is a coherent cluster of tasks. Each WP is headed by a WorkPackage Leader responsible for the work to be carried out. WorkPackage Leaders report to thescientific coordinator. Objectives: The WP leaders are responsible for the set of activities assigned to them in the workplan, and are in charge of the corresponding reports and deliverables. The WP leaders will bemembers of the RTCB and will be members the Executive task force. They are also expected tocollaborate and exchange views with the other WorkPackages leaders for improved coordinationof the different network activities. Tasks: Each WP is composed of several sub tasks that have to be conducted as described in thework plan. Each task is headed by a task leader that reports directly to the WP Leader.Means: The activities carried out in the different WPs will rely on the students and on the expertiseprovided by the ARTDECO partner institutes to fulfil their objectives.Assessment: The WP Leaders will perform a regular followup of the tasks in their WP. The RTCBwill perform a general assessment of each WP activity.Scientific Coordinator
Dr Mario Arioli will be in charge of the scientific coordination of the network. He will be carryout this role by chairing the RTCB, in collaboration with high profile experts including Prof. FrancoBrezzi (Team 4) as the expert for the fields of partial differential equations and their applications,and Prof. Philippe Toint (Team 9) as the expert in the field of Optimization. His duties also involve
Part B – Page 33 of 45
ARTDECO
the monitoring of WP activities and the validation of all scientific deliverables. He will play animportant role in identifying the best practices implemented in ARTDECO and in supporting theirbroader adoption by the rest of the project participants. Finally, he will be the scientificrepresentative of the project at International conferences and events, and as such, he will also haveto play a significant role in fostering collaboration with Industrial partners, with the support of theentire consortium.Project coordinatorWhile the scientific coordinator Dr Arioli will focus on the scientific coordination, the projectcoordinator, Bruno Le Dantec, will be assisted by four dedicated assistants composing theERCIM team to concentrate its activities on the overall management and administrative andfinancial coordination of ARTDECO.This will imply directing the administrative work of the project, managing the Human Resources,the Mobility Program requirements and ensuring institutional exchanges with the EuropeanCommission representatives. The coordinator is also responsible for managing the progress andmanagement reports, monitoring the project costs and signing the agreements with the selectedresearchers. His specific responsibilities include:
• Official point of contact with the European Commission• Contract signatory for the contract with the European Commission• Preparation of European Commission and Closure Reports• Supervising efficient Human Resources Management• Point of contact for conflict resolution• Establishing efficient internal management and control procedures including budgets
He will also head WP3 and monitor the ARTDECO the dissemination activities. Further details onthe dissemination efforts are presented in the following sections. The ERCIM team will assist him in his daytoday management. This team will be completed withmembers of the ERCIM HRM task force to compose the HRM Office. The HRM Office will beresponsible for recruiting earlystage or experienced researchers chosen by the scientists, ensuringcoherent HR logistics and policies, and for the preparation of an efficient followup of theresearchers to set their careers on optimal tracks. This will include the placement of theseresearchers in promising positions through networking and jobposting within the scientificcommunity. Being an EEIG of 18 research institutes in Europe, ERCIM will initiate this networkingwith the support of the ARTDECO consortium.Therefore the HRM Office will also ensure that a constant contact will be maintained with theselected researchers.4.2 Administrative coordination (Task 3.3 of WP3)Within ARTDECO the administrative coordination is two fold:1.General coordination the project2.Administrative Coordination of Human Resources to provide optimal support to the studentsinvolved in project
Part B – Page 34 of 45
ARTDECO
The general coordination will be carried out by the administrative and financial coordinator,
Bruno Le Dantec (ERCIM), with the assistance of the ERCIM Office team composed ofexperienced project managers and assistants. The daytoday coordination consists of a wide arrayof tasks such as:
1. Administrative and financial coordination within and across workpackages and specifictask forces,
2. Quality Management ,3. Interactions and negotiations with the European Commission, 4. Preparation and delivery of periodic reports, organisation of the periodic meetings,5. IPR issues and publication policy6. Definition of the Consortium Agreement if required, 7. Collection and preparation of deliverables, 8. Contract amendments,9. Audit Certificates
In addition the administrative coordination is responsible for establishing a coherent informationflow among the different ARTDECO stakeholders and workpackages. Several measures havealready been considered to avoid any communication bottlenecks, in particular with the adoption ofcollaborative tools, dedicated internal communication web pages and newsletterMoreover, the coordination will ensure the continuity and consistency of the ARTDECO initiativeafter the contractual end of the projects. The coordination will therefore implement severalmeasures that will ensure the distribution of information and support even after the end of theproject:
Maintaining a central archive of all documents produced within ARTDECO; Ensuring lasting contact between promising researchers and institutions through thepreservation of mailing lists; Updating job position offers and monitoring the backlog of researchers looking for aposition (job posting)
ERCIM has a long experience in managing EC projects. The central office of ERCIM, located inFrance, is a small and flexible administrative unit, which acts as a frontend to access the scientificexpertise of its members. Since 1990, ERCIM has been involved in over 30 European projectsunder the ESPRIT, IST FP5, Telematic, INCODC, IST, HCM and TMR Programmes. In theseprojects, the ERCIM office takes care of the financial and administrative tasks. This distribution ofwork has been a valuable asset, allowing the research institutes and the other partners to focus onthe scientific tasks at the core of the project. Within FP6, ERCIM has already been praised for theexcellence of its management and will be coordinating two Networks of Excellence in Europe,MUSCLE (Semantic web) and DELOS (Digital Libraries).
The Administrative Coordination of Human Resources encompasses several tasks, amongwhich:
Part B – Page 35 of 45
ARTDECO
Providing guidelines for the selection of high profile students Administration & Payment of the student grant Follow up of the students (administratively, financially and in terms of career development) Provide a suitable environment for career development (job posting, relations with researchinstitutes and private companies, career development help and guidance) Lasting contacts and support after the end of projects Organisation of the promotion ARTDECO and of its students to scientific communities in Europeand worldwide
The ERCIM HRM task force will complement this administrative team by providing advanced HRexpertise. The HRM task force has a proven expertise in managing international human resources inparticular with the management of running the ERCIM fellowship programme for more than 10years, of the DELOS and MUSCLE fellowship programmes, and with the successful internalmobility scheme within ERCIM.
The selection procedure described hereafter will be validated and implemented by the HRMOffice. The positions will be advertised internationally relying essentially on the project web siteand on all the networking the project can muster. The ARTDECO contact point in the HRM Officewill receive the student applications. The Office will interact with the students requesting theappropriate documents to ensure a thorough evaluation. All this information will be made availableon the ARTDECO Web site for the consortium to consult. Every application will be reviewed byone or more senior scientist in each partner institute. The RTCB then decides on the finalrecruitment taking into account:
• The qualification of the applicant, • The overlap of interest between the applicant and the hosting institutes,• The available funding
All partners are well aware of the necessity to promote gender parity during the recruitmentprocedure. The HRM Office has already been implementing the European directives in its internalERCIM programmes. It will be clearly indicated in the advertisement that the consortium willensure the participation of at least one third of each gender in the project, and evaluators will bebriefed before any evaluation of the candidates.
In managing training and educational initiatives, the ERCIM Human Resources Managementexpertise will be a valuable asset. Indeed, ERCIM has also a great experience in the HumanResources and Mobility management through different activities such as the internal mobilityscheme, the Fellowship Programme, and the PhD mobility Programme. The main assets of ERCIMare the human potential (in number and quality) of its 18 member institutes and the portfolio ofR&D topics offered to the outside word. The ERCIM Fellowship Programme was launched in 1990and it has been running independently ever since to enable young scientists to perform research at
Part B – Page 36 of 45
ARTDECO
different ERCIM institutes across Europe. The fellowships are open to PhD holders from all overthe world. To promote crossfertilisation and cooperation between research groups working insimilar areas in different laboratories, the fellow is hosted by two ERCIM Members for a ninemonth period each. With twelve years of previous experience, more than 150 fellows have beenawarded through this ERCIM Fellowship Programme.Financial coordination (Task 3.4 of WP3)
Within ARTDECO, ERCIM will perform the interface between the European Commission and theARTDECO partners, redistributing funding efficiently to avoid otherwise lengthy and costlyprocedures.
This task will be directly handled by the ERCIM Office team which has a long expertise in thefinancial coordination of European Projects (STREPs, IPs NoEs in FP6 and also in previous FP4and FP5 programmes)
The financial coordination will ensure:• The preparation of the annual cost statements in line with the European Commission
expectations & procedures;• the payment and fund transfers to all the ARTDECO participants;• Implement regular payments of grants to the early stage and experienced researchers;• Monitor the ARTDECO expenses and assess their relevance against the workplan;• Perform the internal financial audit of the project for enhanced visibility;
• If necessary, prepare internal budget redistribution among tasks or/and partners for
optimal allocation of the financial resources in the project; and• Clear financial reporting with the European Commission.
ERCIM team has a proven expertise in the financial management of projects and networks. TheFP5 RESET network on smart cards involved 120 members managed by ERCIM and within FP6,ERCIM is managing three networks of Excellence, namely CoreGRID, MUSCLE and DELOS.
Internal Dissemination (Task 3.1 of WP3)Beyond the administrative and financial coordination, ERCIM will also have to provide the bestenvironment for both research and training within ARTDECO.
The internal dissemination will therefore essentially focus on three priorities:• internal information flow• collaborative work environment• supporting international dissemination
This will imply providing the network with reliable working tools and a collaborative environment.The underlying idea is to eliminate any internal communication bottlenecks and to stimulateexchanges not only among partners, but also towards students. This task will
• Provide an overview of the project and of its subtasks
Part B – Page 37 of 45
ARTDECO
• Present the expected achievements• Prepare the ground for cooperation of different teams from different countries around
the same objective• Ensure a coherent internal information flow and supporting working environments and
collaborative tools.• Promote and diffuse knowledge about ARTDECO internally• Provide all the necessary support material to ensure a broad international dissemination.
External Dissemination (Task 3.3 of WP3)Finally, ERCIM will also ensure the external dissemination. The objective here is to Disseminate information on ARTEDCO activities, Prepare a wide array of dissemination vectors (logo, web site, poster, flyer, electronic newsletters) Attract high profile experts and institutes or companies to support this initiative, Attract promising young researchers, Advertise events to the scientific community (academy and industry), and Promote the ARTDECO students towards potential employers from both industrial and academicworlds.
Particular efforts will be paid to identify clear dissemination targets in the scientific community inboth industrial and academic worlds to ensure the optimal placement of the young ARTDECOresearchers. Moreover this activity will:
1. Organise the ARTDECO summer schools 2. Activate networking capabilities of all the partners to promote the project and students 3. Organise workshops to present the achievements of the network 4. Carry out lasting dissemination, even after the end of the project 5. Publish papers at conferences and in journals; 6. Support the mobility of earlystage and experienced researchers among the network
participants; 7. Perform broad dissemination to the International Scientific Community, to radiate
beyond Europe; 8. Reach out to the industrial stakeholders to promote industrial careers, 9. Establish and update a dedicated web site for job posting and to give information on the
project10. Attend International conferences11. Include articles in ERCIM News, with more than 11000 copies distributed worldwide. 12. Update an electronic newsletter dedicated to the network.
B4.2. Management knowhow and experience of network coordinator ERCIM
The ARTDECO Research Training Network will be coordinated and managed by EEIGERCIM,the European Research Consortium for Informatics and Mathematics. ERCIM is an organisation
Part B – Page 38 of 45
ARTDECO
dedicated to the advancement of European research and development in information technology andapplied mathematics. Its eighteen national member institutions aim to foster collaborative workwithin the European research community and to increase cooperation with European industry. Assuch it has a long and established expertise as a project coordinator and as a training institution forPhD students in different institutes across Europe. To conclude, the coordinator is not only abreeding ground to attract the best young and experienced researchers in this field via his currentactivity and his network of centres of excellence, but is also able to bring its experience ofmanagement at an international level to this Research Training Network.
B.4.3. Budget
We emphasize that 80% of the budget will be entirely dedicated to the financial support of theearlystage and experienced researchers stipends and mobility.
Budget Breakdown
Part B – Page 39 of 45
ARTDECO
B.5 ADDED VALUE TO THE COMMUNITY AND RELEVANCE TO THE OBJECTIVESOF THE ACTIVITY
The use of computational mathematics plays a vital part in technological leadership. In the USA,the NSF budget for computational mathematics has increased for the last three years and aconsiderable part of this budget targets advanced training. In Europe, computational mathematicsexpertise is scattered and fragmented and lacks the recognition awarded in the USA. By gatheringtogether the disseminated expertise in Europe into a European training environment, we intend toensure a future and lasting flow of young experts in the area. Because some regions actually sufferfrom a lack of means and expertise, an important aspect of the proposed network is to developdoctoral level courses not typically available in all European universities. The constituent membersof ERCIM are involved in sponsoring the mathematical activities within the national laboratories.The ambition of our network is to start the seminal work of involving the national laboratories withEuropean universities into a common educational project involving top quality research andsoftware development.Furthermore, we also want to promote an interdisciplinary action within the field of computationalmathematics. In addition to a strong specialist training, we will stimulate all the players involved tohave a sounder knowledge of all the research developed in the network. This will hopefully preparea new generation of researchers inclined to exercise more lateral thinking in their work. Beyond thisdirect result, ARTDECO will also support the European scientific community in several lessexplicit ways.
Firstly, the added value of such an European training network is a real opportunity for the new andfuture member states. The European Research Area will directly benefit from this initiative as itwill help new member States to augment their expertise. Indeed, new member States have often avery strong mathematical tradition with excellent educational programmes with a high participationrate, but it appears that they often lack good programmes for training in applied mathematics andapplied research. This situation not only explains why these countries will profit from the networkactivities, but also highlights the fact that ARTDECO will also benefit from the local expertise thesecountries have preserved and developed as a tradition.
Another side effect of ARTDECO will be to improve the strength of research and support theexistence of a Europeanwide network. This will prove essential in our effort to keep the bestresearchers in Europe. Most of the world leading researchers in mathematics are European, yet toooften these researchers work in the USA. This project will contribute to reducing or reversing thistraditional brain–drain. It is the network’s intention to capitalise on the existing Europeanexcellence in the field to further develop the expertise of this research community.
Moreover, we want to promote the image of mathematics beyond the borders of our project makingthe material used for the training of our young researchers available to a larger sector of the public
Part B – Page 40 of 45
ARTDECO
by the creation of specialized web pages where the topics of our research are presented in simplebut precise terms to all the nonspecialist scientific community and to the interested public.
As mentioned previously, part of the impact of this training programme is to ensure a future andpermanent flow of young experts in the field. This goal is also part of the rationale behind theproposed mix of earlystage and experienced researchers. By bringing together postdocs and earlystage researchers, we hope to enrich the community with shared expertise and create links amongdistinct generations of researchers, as a first step towards future collaboration. The underlying ideais to get experienced researchers to assist earlystage researchers in the development of their futurecareers. This will act as an early transfer of knowledge which the network believes researchers willrely on to pass on knowledge and expertise.
The subdivision between earlystage and experienced researchers and their distribution among theteams has been motivated by additional technical and institutional reasons. Some nationallaboratories do not have the juridical authority to confer academic titles. Nevertheless, they are therepositories of extremely valuable specialist competence and of cutting edge computing equipment.These two qualities add value to the network and give the opportunity to the young researchers tobecome familiar with world class supercomputers and to be involved in the theoretical and practicaldesign of advanced numerical software.We anticipate that the impact of 17 new researchers in the field of computational mathematics willimprove the human resources of the teams involved. The successful achievement of the researchobjectives and the transfer of the relative knowhow and of the skill in using the related numericalsoftware will enhance further the reputation of the teams involved and will promote and assist thefuture career of the earlystage and experienced researchers at the international level. In this respect,we plan to invite among the speakers at our workshops and the teachers at our summer schoolsselected international nonEuropean specialists. This will give the opportunity to build additionalpersonal relationships between our young researchers and the international scientific community.
Finally, ERCIM will take responsibility for advertising the results and the transfer of the knowhowat the international, the European and the national level through its institutional channels.With its European dimension and joint programme for postdocs and earlystage researchers, thismultidisciplinary training network will promote excellence across boundaries, between disciplines,educational systems and age generations.
B.5.1 Connections with other FP6 thematics
The ARTDECO proposal has evident links to other research areas supported by the EuropeanUnion. More specifically, our proposal can be seen supporting some of the FP6 workpackages suchas:• Aeronautics and Space (Thematic Priority 1.4) in the sectors:
• Strengthening competitiveness (1.3.1.1)
Part B – Page 41 of 45
ARTDECO
• Improving environmental impact with regard to emissions and noise (1.3.1.2)• Integration of experimental and analytical research capacities for fixed wing aircraft (13)
• Knowledgebased multifunctional materials, new production processes and devices(Thematic Area 3) in the sectors:
• Knowledgebased Multifunctional Materials (3.4.2)• Development of fundamental knowledge (3.4.2.1)• Technologies associated with the production, transformation and processing of• Knowledgebased multifunctional materials (3.4.2.2)• Engineering support for materials development (3.4.2.3)
• New Production Processes and Devices (3.4.3)• Development of new processes and flexible & intelligent manufacturing systems
(3.4.3.1)• Systems research and hazard control (3.4.3.2)• Optimising the lifecycle of industrial systems, products and services (3.4.3.3)
Moreover, several of the tasks described in Section B.1.5 match the themes described in theNETIAM report “New and Emerging Themes in Industrial and Applied Mathematics” regarding the“Challenges in visualization, simulation and design for virtual porous materials”1. In particular, we emphasise the objectives described in WP1 are tightly connected to the study ofnew technologies appropriate for the modelling and numerical simulation of porous media.
This positions ARTDECO at the cutting edge of new technologies, exposing to them both the earlystage researchers and the experienced researchers involved in the training.
1 see Report of a workshop held Kaiserslautern, Germany, on 29th & 30th September 2004 athttp://www.smithinst.ac.uk/Events/Kaiserslautern/KLReportFull Prepared by the Smith Institute on behalf of theNETIAM Project
Part B – Page 42 of 45
ARTDECO
INDICATIVE FINANCIAL INFORMATION
Indicative financial information on the network project (excluding expenses related to the recruitment of earlystage and experienced researchers)
NetworkTeamNo.
Contribution to the research/ training /transfer of knowledge expenses
(Euro)
Managementactivities
(including auditcertification)
(Euro)
Other types ofexpenses / specific
conditions
(Euro)
(A) (B) (C) (D)
1.2.3.4.5.6.7.8.9.10.11.12.
120463278109
84278 84278 201285 289010 113560 84725 159238 174460
10000
252460 50000
Totals 1425061 10000 252460 50000
OTHER ISSUES
Nil
Part B – Page 43 of 45
ARTDECO
ETHICAL ISSUES CHECKLIST
Table A. Proposers are requested to fill in the following table
Does your proposed research raisesensitive ethical questions related to:
YES NO
• Human beings X
• Human biological samples X
• Personal data (whether identified byname or not)
X
• Genetic information X
• Animals X
If you answer “YES” to any of the above, please include in your proposal section B7 the more detailedversion of Table A (“Crucial information”) obtained from:http://europa.eu.int/comm/research/sciencesociety/ethics/rules_en.html and also incorporate in section B.7 and in other appropriate parts of your proposal commentscorresponding to the detailed instructions given in sections CD at the above address
Table B. Proposers are requested to confirm that the proposed research does not involve:
13. Research activity aimed at human cloning for reproductive purposes,
14. Research activity intended to modify the genetic heritage of human beings which could makesuch changes heritable2
15. Research activity intended to create human embryos solely for the purpose of research or for thepurpose of stem cell procurement, including by means of somatic cell nuclear transfer.
Confirmation : the proposed researchinvolves none of the issues listed in Table B
YES NOX
Further information on ethics requirements and rules are given at the science and ethics website athttp://europa.eu.int/comm/research/sciencesociety/ethics/ethics_en.html
2 Research relating to cancer treatment of the gonads can be financed
Part B – Page 44 of 45
ARTDECO
ENDPAGE
HUMAN RESOURCES AND MOBILITY (HRM) ACTIVITY
MARIE CURIE ACTIONSResearch Training Networks (RTNs)
“Interdisciplinary and Intersectorial”
Call: FP62005Mobility1
PART B
STAGE 2 – FULL PROPOSAL DESCRIPTION
“ARTDECO”
Part B – Page 45 of 45