Computer simulation is a primary approach for studying complex processes in virtually every field of academic and industrial science and engineering. Communities in many disciplines are developing cyberinfrastructures to enable new scientific discovery on current distributed terascale resources, and to prepare for petascale computing.

The numerical relativity community has faced computational issues for years. Einstein's equations are among the most complex formulations in physics; solving them for realistic astrophysical scenarios requires a blend of theoretical physics, astronomy, nuclear and particle physics, computational mathematics, parallel computing, software engineering, and beyond. Single code modules, e.g., for locating a black hole horizon or extracting gravitational waves, may take several person- years to develop by one group, and are often either unavailable for other groups or incompatible with their code base, slowing progress. One critical area of numerical relativity research is the simulation of strongly gravitating systems, such as binary black holes. The merger of two black holes releases a tremendous amount of gravitational energy and is one of the most promising sources of gravitational waves in the universe. Such waves from stellar mass BHs are on the verge of being directly observed with newly operating ground-based laser interferometric detectors (LIGO, GEO600, VIRGO), while future space based observatories (LISA) are planned for observing galactic mass black holes. Numerical modeling of the waves produced by these binaries is crucial for these efforts.

However, black hole binaries deal only with vacuum; a deep understanding of other binary systems, includ- ing combinations of black holes, neutron stars, quark stars, etc., as well as core-collapse supernovae and gamma-ray bursts, is needed for a complete understanding of observable sources of gravitational waves. Modeling these systems is much more complex, requiring a complete treatment of GR hydrodynamics, complex equations of state, MHD, radiation transport, etc., with greater computational and collaboration challenges. Such problems will require petascale computations that scale, new algorithms, closer cooperation between scientists from different disciplines, and a software infrastructure framework that facilitates these activities.

XiRel will develop the basis for a next generation infrastructure for numerical relativity, building on existing efforts in Cactus and Carpet, and experiences in developing and using Cactus infrastructure for numerical relativity, and collaborations with new enabling technologies. XiRel has three core thrusts:

• Next Generation Infrastructure for Numerical Relativity: The central goal of XiRel is the development of a highly scalable, efficient and accurate adaptive mesh refinement layer based on the existing Carpet driver, fully integrated and sup- ported in Cactus and optimized for numerical relativity. XiRel will enhance current tech- nologies to enable current science needs, working towards longer term goals for petascale machines, with minimal changes needed to properly constructed physics modules. By researching new algorithms and implementations for automatic choice of required accuracy, dynamic load balancing, and task spawning to off-load selected tasks, we will: (i) dramatically improve the scaling of Carpet and numerical relativity codes for computation on at least 1000 processors or more, (ii) enable automatic grid hierarchy adaptation and dynamic load distribution on architectures with deep communication hierarchies, and finally (iii) collaborate with and provide expertise to other pro jects to incorporate and use new highly scalable mesh refinement drivers; all in anticipation of solving high resolution petascale sized problems deployed on 100,000 processors.
• Einstein Toolkit: XiRel will investigate the development of a community oriented toolkit for numerical relativity, learning from experiences with the CactusEinstein set of thorns for Cactus.