The magnitude 9 nobody recorded: why simulated ground motions stay out of the codes

The probabilistic chain is where the limit lives. Fault models supply the set of possible ruptures and the magnitudes they can generate, and a ground-motion model turns each rupture into expected shaking at a site with probabilities attached ^[6]. That second step is a curve fitted to recordings rather than a calculation of rupture and wave path ^[7], so it speaks confidently only about combinations of magnitude and distance that appear in the recording set. The Physics World account puts the worst gap at higher magnitudes and shorter source distances, because only a handful of large earthquakes have occurred since instrumental recording began in the last century ^[8][9]. That corner of the space is also the design case for anything sited near a major fault.

Cascadia shows what the gap costs. The subduction zone runs for hundreds of kilometres along the coast of the US Pacific Northwest and southwestern Canada ^[10], and lies about 200 km west of the Portland venue where the workshop was held ^[11]. It has produced no magnitude 9, and no significant earthquake at all, during the century of recording ^[12], while the geological record holds several large events across thousands of years ^[13]. The scenario a designer has to survive therefore has zero entries in the dataset the design number is fitted to ^[20]. Simulating it, the author's PhD subject ^[14], is the only route to a shaking field for that case, and the same simulations carry through to the landslides and tsunamis that follow ^[19].

The adoption problem is not a computing problem, and a line from the workshop states it neatly. Greg Deierlein of Stanford said simulations are most useful when they provide different answers compared with conventional methods ^[15]. Taken as an adoption criterion, the useful case and the awkward case are the same case: a simulation that reproduces the empirical model adds a confirmation, while one that departs from it asks a committee to publish a number no recording supports ^[21]. Nothing in the present arrangement rewards that, which fits the observation that simulations sit outside most national hazard models and appear at negligible scale where they are used ^[16].

What closes the gap is unglamorous. It is documented evidence that simulations represent complex fault geometry and rupture mechanism, that velocity structure models are reliable, and that simulated motions reproduce empirical data ^[17], because trust among researchers and practitioners is built on testing and validation ^[18]. That is benchmark work with citable results, not a bigger machine, and it competes for time with running more simulations. One caveat on provenance: this is a first-person column by a researcher who works with physics-based ground motions, reporting one workshop ^[22], so it describes the community's stated obstacle rather than measuring it.