The Swiss High Performance and Productivity (HP2C) Initiative

Numerical simulations play a growing role in scientific research, including the area of materials simulations relevant for MaNEP. However, fermionic systems in particular are extremely difficult to tackle numerically due to an exponential scaling of the computational error with the size of the system (so called "sign problem"). In order to fight this exponential and obtain reliable results for reasonable system sizes or temperatures two complementary routes are followed: new algorithmic developments (see for example the article on the IUPAP prize awarded to Prof. P. Werner in this issue) enable more accurate and reliable simulations.

At the same time supercomputer performance continues to increase at an exponential rate, increasing at a factor of 1000 every 10 years, and no slowdown is in sight for the next decade. So one also benefits from this brute force increase in computer power. However, since the performance of individual CPUs are limited by physical issues (such as power dissipation), most of the performance increase in the past decades has come through increased parallelism. Modern petaflop machines have more than 100 000 CPUs and future machines will have millions of CPU cores.

This, and the other "brutal facts" of high performance computing (less and slower memory per core, slow growth of network and input/output performance, and higher failure rates as the number of components is increased) makes the development of simulation codes that can make use of such massively large number of CPUs a formidable challenge.  

Development of supercomputer performance according to the top-500 list (http://www.top500.org). The graph shows the performance of the fastest computer in the world (#1), the 500-th fastest computer in the world (#500) and the sum of performance of the 500 fastest computer in the world. Despite a stagnation of single-CPU performance massive parallelism accounts for continued exponential growth of total performance.

The Swiss High Performance and Productivity (HP2C) initiative (http://www.hp2c.ch) was started to help Swiss scientists tackle these challenges and to provide computational research groups at Swiss universities with the necessary resources to be the world leaders in the developments of high performance scientific simulation codes. In addition to a core group of scientific computing experts at the Swiss National Supercomputer Center (CSCS) and the Institute for Computational Sciences of the University of Lugano (USI) the HP2C initiative funds ten research projects with resources to hire software developers to assist in the development of simulation codes for future computer architectures.

One of the projects selected after a competitive call was our project "MAQUIS - Modern Algorithms for Quantum Interacting Systems", by MaNEP members T. Giamarchi of the University of Geneva, F. Mila of EPFL and M. Troyer of ETH Zürich. With funds to hire three to four software engineers MAQUIS will develop massive parallel implementations of a density matrix renormalization group, exact diagonalization, quantum Monte Carlo and series expansion algorithms that will help us tackle problems in frustrated quantum magnetism, ultracold atoms, strongly correlated fermions, and non-equilibrium dynamics of quantum systems. Besides this MaNEP project other HP2C projects of potential interest for MaNEP members are the BigDFT  (large scale density functional electronic structure calculations) by S. Goedecker at  the University of Basel and CP2K (New Frontiers in ab initio MolecularDynamics) by J. Hutter at the University of Zürich.

Example of the SzSz dynamical structure factor computed by the DMRG technique, for various values of the external magnetic field (top to bottom). The left are the symmetric excitations whose low-energy part can be described by a Luttinger liquid theory. The right are the antisymmetric excitations, that are specific to the ladder.