The accuracy and predictive power of materials’ simulations based on “first-principles calculations” allows nowadays to identify novel materials with improved or custom-designed properties and performance. The ever-increasing volume of computations and of data produced by them is triggering a scientific and technological revolution where behavior of ever-increasing complexity can be addressed in numerical experiments; where data and workflows’ sharing can greatly enhance the synergies between different communities and efforts; and where services can be provided in the form of data, codes, expertise, workflows, turnkey solutions, and a liquid market of computational resources. Sustaining the effort that is supporting such a revolution will require a paradigm shift in software design and in the day-by-day work of materials simulation practitioners, based on the integration of concepts, tools, and applications from high-performance computing (HPC) on novel computer architectures and platforms and on high-throughput computing (HTC) on heterogeneous and geographically distributed computational resources. The challenge is here to run and manage an ever-increasing number of concurrent simulations (HTC) of ever-increasing size at an ever-increasing speed (HPC).
QE is based on density-functional theory, pseudo-potentials, and plane-wave basis sets. Besides being extremely flexible, versatile, and highly optimized on a broad variety of computer architectures, QE owns its popularity to a a very successful series of international training courses, run under the auspices of the Quantum ESPRESSO Foundation (QEF), often in collaboration with the Abdus Salam International Centre for Theoretical Physics (ICTP). Traditionally these schools address the fundamentals of quantum simulation of materials based on density-functional theory, occasionally with emphasis on some advanced applications (such as, e.g. computational spectroscopy, quantum transport, magnetism, etc.).
The school being proposed here is the first specifically aimed at training simulation practitioners in the use of modern massively parallel HPC architectures (such as those currently available at the peta- scale level and foreseen to transition within few years to the exa- scale level) on one side, and on the use of advanced tools for generating, managing, storing, and sharing results from HTC simulations, on the other. HTC training within the school being proposed will be based on the AiiDA infrastructure.
AiiDA (Automated Interactive Infrastructure and DAtabase for computational science) is a Python materials’ informatics framework to manage, store, share, and disseminate the workload of high-throughput computational efforts, while providing an ecosystem for materials simulations where codes are automatically optimised on the relevant hardware platforms, and complex scientific workflows involving different codes and datasets can be seamlessly implemented and shared. Five active developers of the code are at EPFL (another main developer is at Bosch RTC in Cambridge, MA), and are complemented by two software engineers supported by the Swiss MARVEL project. AiiDA is designed around the four pillars of materials’ informatics, as follows. At the low-level, AiiDA takes care of (1) automation and (2) data storage for the management and safeguarding of calculations, data and workflows. At the user-level, it provides an advanced and intuitive research (3) environment for accelerating scientific discoveries, and (4) sharing capabilities to enable collaborative research.
Three main topics will be covered through a synergistic mixture of theoretical lectures (L),
technical lectures and demonstrations (T), and dedicated hands on sessions (H):
- Basic QE tutorial:
- Week 1, Day 1 (morning): DFT - the theory (L)
- Week 1, Day 1 (afternoon): basics of Unix, compilation, jobs submissions (T,H)
- Week 1, Day 2 (motninh): DFT - the practice (basis sets, pseudopotentials, FFT) (L,T)
- Week 1, Day 2 (afternoon): total energy and force calculations (T,H);
- Week 1, Day 3: structural optimization and ab-initio molecular dynamics; (L,T,H)
- Week 1, Day 4: linear response, DFPT, phonons; (L,T,H)
- Advanced QE tutorial:
- Week 1, Day 5: Vibrational and magnetic spectroscopies: IR, RAMAN, NMR. (L,T,H)
- Week 2, Day 1: Optical spectroscopies: time-dependent DFPT, many-body perturbation theory (L,T,H)
- Week 2, Day 2 (half day): Solvation and electrochemistry, PBC corrections (T,H)
- QE and AiiDA: the materials’ science ecosystem in action; Automating, storing, and sharing QE calculations using AiiDA (not only meant to people willing to run high-throughput calculations, but in general for results organization, and access to workflows and turn-key solutions).
- Week 2, Day 2 (half day):
- Setting up AiiDA: configuration of codes and computers (T,H);
- Running QE scf calculations using AiiDA and accessing the results (H);
- Week 2, Day 3:
- Querying AiiDA data to show/discover trends in the simulations (H);
- AiiDA workflows: compose automated sequences of calculations (e.g. relxations, band structures, phonons) (H);
- Writing a new plugin for AiiDA: supporting your favorite code (H).
Handling QE calculations on HPC systems:
- Week 2, Day 4:
- evolution of present and future HPC architectures (including a discussion of the
- roadmap towards exascale). (L)
- different levels of MPI parallelism of QE, features, pros & cons, including OpenMP
- support and hybrid MPI+OpenMP usage; (T,L)
- Week 2, Day 5:
- evolution of programming/parallelism paradigms (L).
- QE and GPU support: a prototype for scaling on heterogeneous arch (T,H);
- QE on Intel-Xeon-Phi as a prototype of scaling on many-core nodes (T,H);