A Priori Error Estimation for Random Field Generation and a Method to Make Random Generation Scalable in Massively Parallel Implementations

Luciano de Carvalho Paludo (1) , Régis Cottereau (1) , Victor Bouvier (1)

1 MSSMat - Laboratoire de mécanique des sols, structures et matériaux

This work analyses the error committed when sampling a random field with the spectral representation method. We conclude the error decreases proportionally to the size of the domain. The problem is that the cost of generation a random field scales as O ( N log ( N )) , where N is the number of points in the simulation. We proposes a subdivision-method that maintain where we can glue together several sample blocks (generated with a O ( N log ( N )) complexity) into one single field. This allows to have a O ( N ) scalability of the system, saving much computational effort and suited to massively parallel architectures.

Mots clés: A Priori Error Estimation; High Performance Computing; Gaussian Random Fields

Domaines: Mécanique des structures [physics.class-ph]

Luciano de Carvalho Paludo, Régis Cottereau, Victor Bouvier. A Priori Error Estimation for Random Field Generation and a Method to Make Random Generation Scalable in Massively Parallel Implementations. 13e colloque national en calcul des structures, Université Paris-Saclay, May 2017, Giens, Var, France. ⟨hal-01926332⟩