Validating IM-based methods for probabilistic seismic performance assessment with higher-level non-conditional simulation P. Franchin, F. Cavalieri & P.E. Pinto Department of Structural Engineering & Geotechnics, University of Rome “La Sapienza”, Via Gramsci 53, 00197, Rome, Italy SUMMARY: Probabilistic seismic performance assessment consists of evaluating the probability that a structure exceeds a limit state at least once within a given time interval (risk). Classical methods to evaluate it are those of structural reliability. For dynamic problems such as the seismic one, recourse to simulation methods is the most common choice. So far, the main obstacles to their widespread adoption have been the limitations of stochastic ground motion models and the computational cost. The last fifteen years have seen a considerable amount of research devoted to the development of approximate affordable alternatives, employing some assumptions to dramatically reduce the effort and requiring only accessible tools and notions in probability/statistics. These methods can be collectively denoted as “IM-based”, since they rely on the hazard-fragility split of risk, which are expressed in terms of an Intensity Measure (IM). The paper critically compares results of the two classes of methods. Keywords: Monte Carlo, Importance Sampling, Synthetic records, IDA, RC frame 1. INTRODUCTION This paper presents a comparison of the results of probabilistic seismic performance assessments carried out with reliability methods and with so-called IM-based methods. The latter methods, emerged in the last years as affordable alternatives to the former ones (Cornell, 1996, Deierlein et al, 2003), derive their name from the main assumption employed to reduce the computational cost associated with simulation. The assumption is that the probability of exceeding a limit state, conditional on a measure of the local ground motion intensity (the IM) at the site of the structure, is independent of other earthquake properties such as, notably, magnitude, source-to-site distance, faulting style, etc. For the same reason the reliability methods are called herein non-conditional, to underline the fact that they do not rely on the above assumption. In both classes of methods there is a spectrum of alternatives. The paper considers two such alternatives for the IM-based methods, and three for the non-conditional methods. It is important to highlight the complementary criticalities of the two classes. Whereas the non- conditional methods can claim to explore in a probabilistically consistent manner the whole random space and thus lead to unbiased estimates of the probabilities of interest, the IM-based ones cannot and, further, the validity of their main assumption (conditional independence) is structure- and IM- dependent. On the other hand, IM-based methods employ real recorded ground motions while non- conditional methods require stochastic models that simulate synthetic motions starting from basic random variables. The capability of the latter to represent real motions has been under scrutiny and the “final” satisfactory model has not emerged yet. Also, uncertainty in hazard prediction, i.e. due to imperfect knowledge in the source characterization (source boundaries, source activity parameters) as well as in the source-to-site portion of the ground motion generation (different ground motion prediction equations), can be easily/economically accounted for in IM-based methods (usually through a logic tree on alternative models and assumptions), while it may represent a challenge for non- conditional methods. In conclusion, apart from the computational aspects, the main difference between the two classes is in