Fellow: ESR1
Host institution: UNIFE
Ph.D. enrolment: UNIFE and BUW
Project Title: Optimized implementations of the Lattice Boltzmann Method in 2 and 3 dimensions on highly parallel computing devices
Objectives: Explore and evaluate opportunities for maximizing the utilization of future versions of highly parallel computing devices such as GPUs, for different implementations of the Lattice Boltzmann Method.
Expected Results: 1) An analysis of the performance of approaches to mapping different formulations of the Lattice Boltzmann Method onto future generations of GPUs and other highly parallel computing devices. 2) A description of performance limiting features of the architecture as well as an evaluation of strategies for overcoming these limitations. 3) Development of implementations, for selected cases, based on the hardware available at the given time to verify the theoretical findings.
Planned secondment(s):
  • JUELICH 3 PMs during M10-M12 training on optimization
  • NVIDIA 3 PMs during M22-M24 performance analysis of LBM
  • UTOV 3 PMs during M29-M31 interacting with experts in turbulence on LBM

 

Fellow: ESR2
Host institution: CyI
Ph.D. enrolment: CyI and BUW
Project Title: Exploiting memory hierarchies in future node architectures for lattice QCD applications
Objectives: Explore and evaluate opportunities for exploiting software managed memory hierarchies for lattice QCD applications and adapt lattice QCD kernels for new architectures as part of the co-design development.
Expected Results: 1) A systematic analysis of mapping different types of lattice QCD applications onto future node architectures, which comprise different memory types as well as compute performance of several TFlop/s. 2) An analysis of different algorithmic approaches with respect to memory capacity and bandwidth requirements and evaluation of suitable programming models to be evaluated. 3) Prototype implementations for selected kernels based on the hardware available at the given time to verify the theoretical findings.
Planned secondment(s):
  • JUELICH: 6 PMs, during M8-M13, to develop expertise in novel computer architectures and code adaptation
  • IBM: 3 PMs on site during M30-M32, interaction with computer designers and testing of software

 

Fellow: ESR3
Host institution: RWTH
Ph.D. enrolment: RWTH and CyI
Project Title: Massively parallel QM/MM
Objectives: The investigation of bio-chemical processes involved in complex pathways, such as the ones of medical relevance, requires a quantum mechanical description coupled to molecular dynamics approaches. If combined with a description based on classical force-fields (computationally less demanding and accurate), such methods (usually referred to as QM/MM) allow reaching the typical sizes of biological relevant systems. Several quantum chemistry codes (e.g. VASP, ABINIT, CPMD, CP2K) implement first principles based molecular dynamics schemes. However, their multiscale QM/MM interfaces show clear deficiencies in scalability and general efficiency. The aim is to develop new strategies to improve the parallel implementations of these hybrid QM/MM interfaces, in order to efficiently exploit the computational capabilities of massively parallel architectures such as IBM Blue Gene/Q ones.
This is a crucial step towards the porting of hybrid quantum molecular dynamics codes to exascale platforms. This would allow unprecedentedly large systems to be accessed by this powerful simulation tool. This will be done in collaboration with U. Roethlisberger’s and A. Curioni’s groups, adding additional human resources to the project.
Expected Results: 1) Development of new parallelization strategies for hybrid quantum molecular dynamics schemes. 2) Implementations in selected available quantum chemistry codes. 3) Scaling tests by benchmarks on IBM BlueGene architectures and validation of the codes by reproducing results available in the literature.
Planned secondment(s):
  • JUELICH: 5 PMs, M16-M20, development of new parallel strategies for hybrid QM/MM interfaces (4 PMs); benchmarks (1 PM)
  • IBM: 4 PMs, M21-M24, optimization of hybrid QM/MM molecular dynamics algorithms on massively parallel machines

 

Fellow: ESR4
Host institution: UTOV
Ph.D. enrolment: UTOV and TU/e
Project Title: Algorithms for Eulerian-Lagrangian approaches in Turbulence, micro- and nano-fluidics
Objectives: 1) Developing new algorithms to describe finite-size particles (with and without internal dynamics) in turbulent flows. 2) Development of novel Lattice Boltzmann approaches for thermal fluctuations at macro scale (boiling systems) and for fluctuating hydrodynamics of multi-component nano-fluids.
Expected Results: 1) Integration of Stokesian dynamics with and without lubrication forces in pseudo-spectral algorithms for homogeneous isotropic and anisotropic flows. 2) Development of the theoretical and numerical expertise to introduce fluctuating hydrodynamics forces in multi-component flows. Optimisation of Lattice Boltzmann methods for Boiling systems.
Planned secondment(s): BUW: 4 PMs, during M6-M9 for training in preconditioning methods

 

Fellow: ESR5
Host institution: UCY
Ph.D. enrolment: UCY and BUW
Project Title: Scalable algorithms for solvers and noise reduction techniques for disconnected quark loops in lattice QCD
Objectives: Two components contribute to the very high computational cost for the calculation of disconnect quark loops in lattice QCD: the numerical cost due to the repeated solves of ill-conditioned linear systems and the statistical cost, which requires many systems to be solved to keep the statistical error below the required threshold. The purpose of this project is to reduce this computational cost by providing an approximation to the operator, which can be inverted directly or very efficiently and at the same time reduce the statistical noise of the remaining part of the operator. We aim at having the contribution of this part to the inverse to be small, so that a relatively large statistical error contributes only little to the overall error, thereby reducing the work on the statistics part. Current approaches in this direction rely on the explicit deflation of small eigenmodes and therefore suffer from the fact that the numbers of eigenmodes to deflate increases proportionally to the volume, i.e. the approach does not scale with the volume.
Expected Results: 1) The replacement of explicit deflation of eigenmodes by deflation of approximate eigenmodes, given explicitly by a hierarchical representation like in multigrid methods. Here, we will build on recent results from aggregation based adaptive algebraic multigrid methods for the Wilson-Dirac operator. Instead of being proportional to the volume we expect the cost of the new approach to proportional to the logarithm of the volume. 2) The incorporation of novel, efficient and highly parallel techniques to approximate the (block) diagonal of the inverse of the operator, or its trace. The efficiency of the two approaches will be compared computationally and as far as possible, analytically.
Planned secondment(s):
  • TCD: 5 PMs, M6-M10, training in algorithms
  • EUROTECH: 4 PMs, M31-M34, scaling and performance test

 

Fellow: ESR6
Host institution: RWTH
Ph.D. enrolment: RWTH and CyI
Project Title: Protein-DNA interactions
Objectives: An enduring challenge in molecular biology is to understand the gene regulatory code; that is, how sequence-specific binding of proteins and other macromolecules to DNA and RNA alters the expression of genes under various cellular and developmental conditions. We aim to characterize the DNA sequence specificity of the REL1 transcription factor (TF) of the malaria-transmitting A. gambiae. Specifically, DNA sequence specificities of REL1, for which structural information is available, will be explored along with homology models of REL1 orthologs in 20 other mosquito species and also Rel, Dorsal and Dif from the fruitfly. Free energy calculations based on hybrid coarse grained/all atoms (CG/AA) molecular dynamics (MD) methods developed in the RWTH-Aachen group will allow a calculation of the protein binding affinity to DNA by using the so-called “alchemical transformation”. Homology models for further mosquito TFs based on fruitfly templates will be produced and their MD-derived DNA binding specificities will be reconciled with sequence-based motif predictions.
Expected Results: 1) Prediction of 20 REL1 homologs complex with DNA by homology modelling and docking. 2) Improvement of hybrid coarse grained/all atoms (CG/AA) molecular dynamics method introducing an adaptive resolution scheme. 3) Relative binding free energy of the complexes by using the so-called “alchemical transformation”. 4) Improved knowledge of mosquito regulatory networks.
Planned secondment(s):
  • ICL: 6 PMs, M15-M20 for bioinformatics analysis in identifying new TFs and binding sites
  • IBM: 3 PMs, M21-M23 for training on scaling of MD codes

 

Fellow: ESR7
Host institution: TU/e
Ph.D. enrolment: TU/e and UTOV
Project Title: HPC for Lagrangian/Eulerian Turbulence, micro- and nano-fluidic
Objectives: Implementation of novel and optimization of existing fully parallel and scalable algorithms to study: 1) finite-size particles (with and without internal dynamics) in turbulent flows. 2) Lattice Boltzmann Methods for thermal flows at macro-scales and fluctuating hydrodynamics at nano-scales.
Expected Results: 1) Development of architecture-optimized applications to handle the evolutions of millions to billions of Lagrangian point-like and/or extended objects moving in turbulent flows using pseudo-spectral or finite difference methods. 2) Development of optimized strategies to integrate Lattice Boltzmann Methods (LBM) including novel algorithms for thermal fluctuations and fluctuating hydrodynamics.
Planned secondment(s): EUROTECH: 4 PMs, M27-M30, scaling and performance tests of novel algorithms

 

Fellow: ESR8
Host institution: UTOV
Ph.D. enrolment: UTOV and CyI
Project Title: Hybrid Monte Carlo (HMC) algorithm for stochastic hydrodynamical systems
Objectives: 1) Path integral investigations of low-dimensional models in turbulence to address extreme and very rare fluctuations. 2) Development of HMC algorithm with constraints to enhance the occurrence of rare events. 3) Implement efficient, fully parallelized HMC code for the models considered, the one and two dimensional Burger's equation, cascade models, shell models and the 2D Navier-Stokes Equations.
Expected Results: 1) Precise evaluation of tails of velocity probability distribution for single point quantities and multi-point correlation functions in Burgers equations. 2) Determination of scaling exponents. 3) Proof of concept of feasibility of this very novel approach for eventual application for the 3D Navier Stokes equations.
Planned secondment(s):
  • DESY: 7 PMs, M12-M18, learn basics of path integral and HMC algorithm, start of simulations
  • NVIDIA: 3 PMs, M19-M21 optimization of HMC code on current GPUs

 

Fellow: ESR9
Host institution: BUW
Ph.D. enrolment: BUW and UTOV
Project Title: Modelling Turbulence in LBM
Objectives: Integrate turbulence models into a LBM framework using large eddy simulations (LES), an established multiscale approach for modelling turbulence, to resolve both the large scale motions and effects of the small scale motions (so-called eddies) using subgrid scale models.
Expected Results: 1) Develop a three-dimensional LBM code to use the large eddy simulations (LES) approach for modelling turbulent flow in the context of LBM and analyze its consistency to other LES for simulating turbulent flow. 2) Derive a perfectly matched layer technique for the 3D LBM code allowing the introduction of artificial boundaries in the simulation domain. This is useful for confining the simulation to a (possibly small) region of interest. We will investigate the usability of this approach and evaluate the accuracy and efficiency by a comparison to the results of a direct numerical simulation. Especially, in the direction of HPC, we want to clarify the computational efficiency of the LBM implementation on GPUs for massively parallel computation.
Planned secondment(s):
  • NVIDIA: 3 PMs, M9-M11, implementation of initial LBM code on current GPUs
  • RWTH: 5 PMs, M23-M27 to learn basics of multiscale algorithms for biosimulations

 

Fellow: ESR10
Host institution: CyI
Ph.D. enrolment: CyI and RWTH
Project Title: Integration of DNA microarray and next generation sequencing (NGS) gene expression data
Objectives: The development of tools for integrating gene expression data generated using two profoundly different technological platforms that are largely used today in fundamental and applied research for public and private health; oligonucleotide DNA microarrays and NGS. These tools will be developed through an academic-industrial partnership and will concern processing, normalization and visualization tools. Both the academic/public (e.g. VectorBase) and industry/private (e.g. OakLabs) sector are highly interested in and will utmost benefit from the project. The biological system will be A. gambiae infections with the deadly human malaria parasite Plasmodium falciparum. Knowledge derived from this research can be used to identify new targets to block malaria transmission. The specific objectives are: 1) To optimise protocols for compatible normalisation of DNA microarray and NGS expression data, including a thorough comparison of compute and data intensive sequence assembly and alignment methods. 2) To develop tools for parallel visualisation of DNA microarray and NGS expression data. 3) To integrate the aforementioned protocols and tools into appropriate user-friendly analysis package.
Expected Results: 1) Optimal protocols for compatible normalisation of data generated using the two platforms, including both legacy and contemporary data. These protocols, and procedures within those, will be published in open access scientific journals. 2) Tools and procedures to visualise simultaneously analysed data following normalisation. These procedures will be published in open access scientific journals. 3) Software that can be used by academia and industry to process, normalise and visualise expression data acquired by the two technological platforms. The software front end shall be user friendly and suitable for the most commonly used desktop and tablet operating systems.
Planned secondment(s):
  • OakLabs: 7 PMs, M14-M17 to design a combined A. gambiae and P. falciparum oligonucleotide DNA microarray platform and train on DirectArray software; M28-30 to integrate normalisation and visualisation tools into the DirectArrray software.
  • ICL: 3 PMs, M18-M20, sample preparation, DNA microarray experiments and NGS data generation and pre-processing.

 

Fellow: ESR11
Host institution: UCY
Ph.D. enrolment: UCY and BUW
Project Title: Observables probing nucleon structure and BSM physics
Objectives: Examples of quantities that can detect BSM physics are the nucleon sigma-term as well as time-like nucleon form factors, which are new observables for probing hadron structure. These and many others entering hadron structure involve disconnected quark loops. The main objective is to develop efficient algorithms in collaboration with ESR5 and use them to compute all disconnected loops for the observables of WP4 to sufficient accuracy thus obtaining a large number of observables such as the electric and magnetic form factors, average quark and gluon momentum in a nucleon and nucleon sigma-terms, neutron electric dipole moment as well as time-like form factors relevant for BSM searches.
Expected Results: 1) Benchmark results for axial charge and average momentum, direct comparison of form factors with experiment. 2) Precise calculation of nucleon sigma-terms to predict a Wimp-Higgs-nucleon cross section for dark matter search. 3) A quantitative estimate of the neutron electric dipole moment. 4) Exploratory study of time-like form factors for a direct comparison to parton distribution functions in collaboration with other members of the group at DESY.
Planned secondment(s):
  • DESY: 6 PMs, during M18-M23 to learn techniques for time-like form factors
  • JUELICH: 3 PMs during M31-M33 for implementation on GPUs

 

Fellow: ESR12
Host institution: BUW
Ph.D. enrolment: BUW and UCY
Project Title: Hadron spectrum and Resonances
Objectives: The fellow will be exposed to computation of hadron masses using clover and twisted mass fermion gauge configurations with physical values of the light quark masses. The study will include hyperons and charmed hadrons and methods for excited states and for determining resonances parameters. New algorithms for fast generation of quark propagators and noise reduction will be investigated. Input for the needed disconnected diagrams will be taken from ESR5 and ESR11.
Expected Results: 1) Hadron spectrum and comparison with experimental values. 2) New algorithms for speeding up quark inversions and noise reduction methods. 3) Determination of resonance parameters of unstable states. 4) Study of excited states and identification of exotics
Planned secondment(s):
  • TCD: 7 PMs, M16-M22, expertise in method for excited states and exotics
  • DESY: 3 PMs, M23-M25, complementary expertise in resonance studies

 

Fellow: ESR13
Host institution: UCY
Ph.D. enrolment: UCY and BUW
Project Title: Renormalization constants and semi-leptonic transition form factors
Objectives: Exposure of fellow to the renormalization aspects of quantum field theories and effective Lagrangians for treating b-quarks on the lattice with the aim to calculate observables that probe BSM physics. Specific examples are: 1) perturbative calculation of RCs for 4-fermi operators such as those appearing in an effective electroweak Hamiltonian description of physics within and beyond the Standard Model. 2) Operators of the electroweak effective Hamiltonian, such as the "chromomagnetic" one which appear in hadronic decays.3) Form factor computation of B to π and B to K decays.
Expected Results: 1) One-loop results for RCs of operators entering the computations of this project, using a large variety of lattice actions, which are currently being used in large-scale simulations. 2) For "ultralocal" fermion bilinears, results to two-loops. 3) Precise computation of B to π and B to K decay amplitudes.
Planned secondment(s): UC: 10 PMs, M18-M27, gain expertise in effective Lagrangians for b-quarks on the lattice

 

Fellow: ESR14
Host institution: CyI
Ph.D. enrolment: CyI and RWTH
Project Title: Targeting mosquito GPCRs and malaria transmission
Objectives: We will screen potential ligands binding to several G protein-coupled receptors (GPCRs) involved in regulating malaria mosquito infectivity. As neither the structure nor reliable structural templates are available for these receptors (sequence identity <24%), standard bioinformatics approaches alone may lead to inaccurate models. We will employ hybrid coarse grained/all atoms (CG/AA) molecular dynamics methods to refine the models. We will next estimate the affinities of non-toxic compounds obtained from ligands’ libraries by chemoinformatics. These affinities will be estimated by free-energy calculations using the “alchemical transformation” method. The compounds with the highest affinities will be acquired and when relevant de novo synthesized at Imperial College and tested in vitro for their ability to change the cellular function of the receptor and in vivo for changing the malaria parasite mosquito infectivity. This research may lead to the identification of new molecules capable of interfering with malaria transmission.
Expected Results: 1) Structural predictions of GPCR's involved in malaria including the gastrin/bombesin receptors GPGRP1 and GPGRP2, the putative neuropeptide Y receptor GPRNPY3, the gastrin/cholecystokinin receptor GPRCCK1 (Mendes et al., Infection and Immunity, 2011) and the gustatory receptor GR9 (Stathopoulos et al., PLoS Pathogens, 2014). 2) Potential ligand candidates for each GPCR by chemioinformatics structure-based ligand screening. 3) Small molecule inhibitors affecting malaria transmission. 4) Improvement of hybrid coarse grained/all atoms (CG/AA) molecular dynamics method by introducing an adaptive resolution scheme. This allows for rigorous free energy calculations. 5) Calculating the relative binding free energy of the complexes by using the so-called “alchemical transformation” method.
Planned secondment(s):
  • ICL: 5 PMs, M10-M14 to test drugs in mosquito infections with malaria parasites
  • NCSR: 3 PMs, M15-M17 to perform in vitro assay development and drug tests

 

Fellow: ESR15
Host institution: BUW
Ph.D. enrolment: BUW and CyI
Project Title: Communication efficient iterative linear solvers
Objectives: The cost for communication tends to be the limiting factor for linear solvers on massively parallel architectures. Communication avoiding numerical algorithms are therefore at the edge of current research in algorithms for scientific computing. Recently developed techniques in communication avoiding Krylov subspace methods will be extended to also comprise preconditioning, a situation which is most relevant for science applications. Multilevel hierarchical preconditioners and their multiplicative, additive and also asynchronous use as a preconditioner will be considered for two cases: 1) For linear operators coming from a regular discretization of a simple domain (lattice). 2) For more general sparse matrices coming from less regular discretization of less regular domains.
Expected Results: 1) Development of a communication avoiding preconditioner based on simple domain decomposition, i.e. without a coupling coarse system. 2) Development of an enhanced communication avoiding method which includes a small size coupling system between the domains to improve numerical properties. This will address implementations on architectures with hardware accelerators, where we distribute the computation efficiently between fat standard nodes and accelerators, using the additive and also the asynchronous concurrency models. 3) An implementation of these approaches for a model problem based on a regular grid discretization will be provided on state-of-the-art and also experimental novel architectures will be provided. 4) Demonstration of the generalization of these approaches to less regular discretizations.
Planned secondment(s):
  • JUELICH: 5 PMs, M16-M20 at exascale lab
  • NVIDIA: 3 PMs, M21-M23, implementation and scaling tests