1. M. Stupazzini M. Stupazzini Grenoble 3D benchmark July 2006 23 rd of July 2006 Kinsale – Ireland (3rd SPICE workshop) Politecnico di Milano Dep. of Structural Engineering Ludwig Maximilians University Dep. of Earth and Environmental Sciences - Geophysics Center for Advanced Research and Studies in Sardinia

2. Seismic wave Propagation and Imaging in Complex media: a European network MARCO STUPAZZINI Experienced Researcher Host Institution: LMU Munich Date of Birth: 26 / 10 / 1974 Place of Origin: Milano, Italy Key Words: Computational Seismology, Spectral Element Method, Visco Plasticity, Soil-Structure Interaction Appointment Time: May 2004 Task Groups: TG 2: Numerical Methods MARCO STUPAZZINI Experienced Researcher Host Institution: LMU Munich Date of Birth: 26 / 10 / 1974 Place of Origin: Milano, Italy Key Words: Computational Seismology, Spectral Element Method, Visco Plasticity, Soil-Structure Interaction Appointment Time: May 2004 Task Groups: TG 2: Numerical Methods

3. 3D numerical simulation of seismic wave propagation in the Grenoble valley (M6 earthquake)

4. 36 km 30 km

10. SD Elements # Nodes # Memory [Mb]  t simulation [sec.] time steps # Total time simulation [s.] Total CPU time (10 CPUs) [min.] 32169726.05·10 6 52820.246E-03121951303070.45 421697214.2·10 6 112110.154E-031951873011177.35

13. Synthetic seismograms

14. Peak ground velocities

15. Spectral ratios for the fault parallel component

16. Max. Displacement E. Faccioli and A. Rovelli: Project S5 of “DPC-INGV” "Definizione dell’input sismico sulla base degli spostamenti attesi" (1 giugno 2005 - 30 giugno 2006)

17. Max. Displacement E. Faccioli and A. Rovelli: Project S5 of “DPC-INGV” "Definizione dell’input sismico sulla base degli spostamenti attesi" (1 giugno 2005 - 30 giugno 2006)

18. Conclusions www.spice-rtn.org www.stru.polimi.it/Ccosmm/ccosmm.htm

19. 1 st 2 nd 3 rd Subsidence Liquefaction Landslides General problem

20. A sub-structuring method : the Domain Reduction Method (Bielak et al. 2003) Local geological feature P e (t) Soil-Structure interaction Inner region External region EFFECTIVE NODAL FORCES P Boundary region Method for the simulation of seismic wave propagation from a half space containing the seismic source to a localized region of interest, characterized by strong geological and/or topographic heterogeneities or soil-structure interaction.

21. The free field displacement u 0 may be calculated by different methods Step I ( AUXILIARY PROBLEM ) The auxiliary problem simulates the seismic source and propagation path effects encompassing the source and a background structure from which the localized feature has been removed. P e (t) Analytical solutions (e.g.: Inclined incident waves) Numerical method (e.g.: FD, SEM, BEM, ADER-DG) DRM : 2 steps method

22. The reduced problem simulates the local site effects of the region of interest The input is a set of equivalent localized forces derived from step I The effective forces act only within a single layer of elements adjacent to the interface between the external and internal regions where the coupled term of stiff matrix does not vanish EFFECTIVE NODAL FORCES uiui wewe ubub Inner region External region Boundary region Inner region Boundary region External region Step II ( REDUCED PROBLEM ) DRM : 2 steps method

23. Study case railway bridge

24. Wave propagation in 2D “ Site effects “ & “ Soil Structures Interactions “ “Source“ & “ Deep propagation“ Fault zoom zoom

25. Computational comparison:Simulation # elem. Memory[Mb]  t simulation [sec.] # time steps Tot. CPU time [min.] Single model 279015 1.177 10 -5 570 620 190.0

26. Computational comparison:Simulation # elem. Memory[Mb]  t simulation [sec.] # time steps Tot. CPU time [min.] Single model 2790151.177 10 -5 570 620190.0 DRM 1 st step 2370145.5 10 -4 18 3625.5

27. Computational comparison:Simulation # elem. Memory[Mb]  t simulation [sec.] # time steps Tot. CPU time [min.] Single model 2790151.177 10 -5 570 620190 DRM 1 st step 2370145.5 10 -4 18 3625 DRM 2 nd step 585181.177 10 -5 570 62064 + The computation with DRM is 2.8 times faster

28. Kinematic source: Kinematic source: Seismic moment tensor density (Aki and Richards, 1980): M W = 4.2, slip = 50 cm Dynamic rupture modelling (Festa G., IPGP) Interface behavior via friction Slip weakening law + Stress distribution Initial Principal stresses : 4.0 10 7 Pa  1 1.8 10 8 Pa  3 100° Orientation 0.67 Static friction 0.525 Dynamic friction 0.4 m D C 150-300m Cohesive zone thickness

29. Comparison

31. Outlook DRM Study case GeoELSE

32. GeoELSE (GEO-ELasticity by Spectral Elements) GeoELSE is a Spectral Elements code for the study of wave propagation phenomena in 2D or 3D complex domain Developers: -CRS4 (Center for Advanced, Research and Studies in Sardinia) -Politecnico di Milano, DIS (Department of Structural Engineering) Native parallel implementation Naturally oriented to large scale applications ( > at least 10 6 grid points)

33. Formulation of the elastodynamic problem Dynamic equilibrium in the weak form: where u i = unknown displacement function v i = generic admissible displacement function (test function) t i = prescribed tractions at the boundary  f i = prescribed body force distribution in 

34. Time advancing scheme Finite difference 2 nd order (LF2 – LF2) Spatial discretization Spectral element method SEM (Faccioli et al., 1997) Courant-Friedrichs-Levy (CFL) stability condition

35.  Suitable for modelling a variety of physical problems (acoustic and elastic wave propagation, thermo elasticity, fluid dynamics)  Accuracy of high-order methods  Suitable for implementation in parallel architectures  Great advantages from last generation of hexahedral mesh creation program (e.g.: CUBIT, Sandia Lab.) Why using spectral elements ?

36. Why using spectral elements ? acoustic problemn=1 Acoustic wave propagation through an irregular domain. Simulation with spectral degree 1 (left) exhibits numerical dispersion due to poor accuracy. n=2 Simulation with spectral degree 2 (right) provides better results. Change of spectral degree is done at run time.

37. Navier’s equation: Fault Internal domain External domain Internal domain: External domain: DRM : 2 steps method

38. u j o = vector of nodal displacements j = i, b, e P b o = forces from domain  + to  0 AUSILIARY PROBLEM (Step I) Faglia Internal domain (0) External domain (0) Mass and stiffness matrices do not change because properties in  + do not change External domain (0): Change of variables : DRM : 2 steps method

39. External domain - External domain (0): Dominio interno: DRM : 2 steps method

40. M and K matrices of the original problem P localization within a single layer of elements in  + adjacent to  (Step II) REDUCED PROBLEM (Step II) DRM : 2 steps method

41. Non linear properties in the internal domain The effectiveness of the method depend on the accuracy of the absorbing boundary conditions DRM : 2 steps method

42. DRM : 2D Validations using Spectral Elements (GeoELSE) Homogeneous valley in a layered half space Mechanical properties V S [m/s]V P [m/s]  [m/s] Valley451051000 Half space 501001200 801401600 1001801800

43. DRM : 2D Validations using Spectral Elements (GeoELSE) Relative displacements (w) Total displacements (u=w+u o ) Homogeneous valley in a layered half space Internal points External points

44. Canyon in a homogeneous half space Mechanical properties V S [m/s]V P [m/s]  [m/s] Canyon501001200 Half space801401600 DRM : 2D Validations using Spectral Elements (GeoELSE)

45. Relative displacements (w) Total displacements (u=w+u o ) Internal points External points Canyon in a homogeneous half space DRM : 2D Validations using Spectral Elements (GeoELSE)

46. Calculation of effective forces P b and P e ORIGINAL PROBLEM II STEP Analysis of wave propagation inside the reduced model. Interface elements Nodes e Nodes b P Calculation of u 0 for a homogeneous model I STEP Analytical solution Numerical methods (Ex. Hisada, 1994) Same method used for step II (ex. SE) Oblique propagation of plane waves inside a valley DRM : 2 steps method

47. Comparison

48. Conclusions Capabilities of DRM to handle „source to structure“ wave propagation problem with reduced CPU time Dialog between numerical codes oriented for different purposes Kinematic model are satisfactory to describe the low frequency bahaviour (e.g.: PGD and PGV) while PGA seems to be overestimated (nucleation, constant rupture velocity and instantaneous drop of the slip on the fault boundaries?).