PRACE Preparatory Access – 39th cut-off evaluation in December 2019

Find below the results of the 39th cut-off evaluation in December 2019 for the PRACE Preparatory Access.

Projects from the following access types:

 

Type A: Code scalability testing (23)

Simulation and modelization of pressurized gas flows in high temperature solar receiver

Project Name: Simulation and modelization of pressurized gas flows in high temperature solar receiver
Project leader: Mr Adrien Toutant
Research field: Engineering
Resource awarded: 50000 core hours on Joliot Curie (SKL) hosted by GENCI at CEA, France
Description

Concentrated solar power plant uses the concentrated solar radiation to heat a fluid. The solar receiver is very important because it transfers the heat toward the fluid. We limit the framework of our study to pressurized air systems. Temperature has an active role on the flow (thermal expansion). Our goal is to analyze and understand fluid behavior in solar modules. In particular, we take into account the coupling effect between temperature and turbulence solving Navier-Stokes equations under the low Mach number approximation. The flow comprehension will be useful for solar receiver optimization. Indeed, to ensure the technology competitiveness it is necessary to maximize heat transfers while minimizing pressure losses. For this purpose, we plan to perform flow simulations in the module for different temperature ranges. Our working conditions are representative of the THEMIS solar power plant experimental situation. The Reynolds number is around 100 000, and heat transfers between wall and fluid reach 280 kW/m². We modelize the solar receiver with a bi-periodic channel flow made of fixed wall temperatures.
top

Biodegrading plastic

Project Name: Biodegrading plastic
Project leader: Prof. Maria Ramos
Research field: Chemical Sciences and Materials
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

We want to be able to biodegrade polyethylene terephthalate (PET) plastic because it is one of the most abundantly produced plastics, widely used in packaging and textiles (where it is known as polyester), which is accumulating in the environment at a staggering rate, and for which no green, environment and economically sustainable recycling strategy exists. Over 60 million tons of PET plastic are produced every year, from which ~60% is used in non-recyclable textiles and ~30% in plastic bottles. Over 500 billion plastic bottles are produced every year, and more than half of them are never recycled. The bacterium Ideonella sakaiensis was found to have the fantastic ability of feeding on plastic. Very recently, the x-ray structures of the two bacterial enzymes responsible for this feat, PETase and MHTase, have been made available in the protein databank. These bacterial enzymes represent a very promising tool to solve the issue of PET plastic pollution because they exhibit a strong ability to biodegrade PET plastic at room temperature. This is the greenest way of biodegrading PET plastic used in medicinal purposes such as medical sutures, in which a small amount of the enzymes is necessary. However, PET plastic biodegradation on a larger scale e.g. for all plastic debris as referred above, needs upscaling engineering, where large amounts of enzyme have to be immobilized in solid surfaces to generate a PET-plastic degrading reactor. In that sense, detailed knowledge on the stability of the enzyme when adsorbed on specific solid surfaces, as done in industrial settings, is needed. Furthermore, the chemistry of these enzymes has to be understood, so that new enzyme mutants more resistant to immobilization and with faster PET plastic degradation rate can be rationally designed, in order to accelerate the process and therefore lower the cost of the production. For this purpose, computer molecular dynamics simulations of enzyme adsorption on solid surfaces, such as graphene, are needed to evaluate enzyme unfolding and to design immobilization-resistant mutants. Additionally, quantum mechanical/classical mechanical computer simulations can be used, to gain an atomic-level picture of the enzyme’s chemical mechanism, and based on it to design enzyme mutants that are highly active even when immobilized. As enzymes are fully biodegradable themselves, and operate at room temperature and pressure, the development of efficient enzymes through computer simulations will greatly facilitate and accelerate the development of new PETase and MHTase enzymes that can be used in industrial degradation of PET plastic waste, in a green, energy-saving and ecological way.
top

Combining Natural Language Processing and Deep Learning Techniques to predict bioactive peptides from natural sources.

Project Name: Combining Natural Language Processing and Deep Learning Techniques to predict bioactive peptides from natural sources.
Project leader: Mr. Hansel Gomez
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 100000 core hours on Piz Daint hosted by CSCS, Switzerland
Description

Bioactive peptides (BPs) are protein fragments that can have a positive impact on body functions through specific protein-protein interactions (PPIs). BPs are known to modulate the digestive, endocrine, cardiovascular, immune and nervous systems. Moreover, they can prevent oxidation and microbial degradation in foods. Unsurprisingly, BPs have drawn the attention of the scientific community, and indeed the pharmaceutical industry, which for decades had been focused primarily on small molecule drugs. There are many advantages of BPs over the former. Firstly, due to the nature of PPI interfaces, which are dominated by large contact areas (i.e. 1500-3000 A2), BPs are more efficient and specific than other drug molecules which typically target small, well-defined protein pockets. A typical PPI interface is constituted by multiple ‘hot spots’ or important interaction sites spread throughout the binding surface, which account for most of the interaction energy. In such cases, only molecules with big contact areas, such as peptides, can achieve nanomolar potency at the PPI interface. Moreover, the high selectivity of BPs translates into fewer off-target side effects than small molecule drugs. BPs degrade into amino acids thus minimizing the risk of toxicity. Furthermore, peptide therapeutics are typically associated with lower production complexity and therefore reduced costs. Finally, the chemical and physical stability, circulating plasma half-life and cell–penetrating properties of BPs can be appropriately improved through various engineering processes. Some BPs occur endogenously in nature; however, the vast majority are encrypted within the structure of parent proteins and so need to be unlocked via hydrolysis to release their activity. Founded in 2014, Nuritas is a biotechnology company that uses AI to predict bioactive peptides from natural food sources that can be used in therapeutic and preventative fields. The company’s unique, disruptive computational approach to therapeutic discovery uses artificial intelligence, deep learning and genomics to rapidly and efficiently predict and then unlock peptides, with peptide predictions validated in-house by our multidisciplinary team of laboratory scientists. The core of Nuritas relies on machine learning (ML); including Natural Language Processing (NLP) to extract information from scientific literature, state-of-the-art models to predict peptide bioactivity, in addition to techniques such as image processing and graph representation. Our main goal for using the PRACE facilities is to develop new NLP-based machine learning models that incorporate more advanced and complex embedding techniques to increase predictive power and therefore identify novel BPs from natural sources.
top

The Effects of Lipid Composition on the Spreading Human Tear Fluid Lipid Layer

Project Name: The Effects of Lipid Composition on the Spreading Human Tear Fluid Lipid Layer
Project leader: Dr. Giray Enkavi
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain, 50000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France, 50000 core hours on Joliot Curie (SKL) hosted by GENCI at CEA, France, 20000 core hours on JUWELS hosted by GCS at FZJ, Germany and 100000 core hours on Piz Daint hosted by CSCS, Switzerland
Description

The dry eye syndrome (DES) is common among people working with computers. It affects about one-third of the world’s population and results in large medical and societal costs. The major cause of DES is the impairment in the composition and the function of the tear fluid lipid layer (TFLL), which spreads over the surface of the cornea at the air-water interface. TFLL performs various essential functions regarding the health and optical properties of the human eye, such as maintaining water evaporation resistance and homeostasis. Proper spreading of TFLL during blinking cycle is crucial to avoid eye drying and the development of DES. The lipid and protein composition of TFLL is well characterized. Yet, the biophysical roles and phase behavior of individual lipid species remain unknown. Here, we will perform atomistic MD simulations of TFLL models composed of mixtures if the major lipid species. With these simulations, we will explore the effect of the non-equilibrium compression–expansion cycles on TFLL structure during the blinking process. We will also investigate how different lipid compositions change TFLL response to the compression–expansion cycles.
top

Scaling tests for the study of microphysical effects in binary neutron star mergers

Project Name: Scaling tests for the study of microphysical effects in binary neutron star mergers
Project leader: Dr. Albino Perego
Research field: Fundamental Physics
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain, 50000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France and 100000 core hours on Hazel Hen hosted by GCS at HLRS, Germany
Description

In this project we want to measure the performance of the WhiskyTHC code on a new HPC architecture in anticipation of a Prace application for the next 20th call. The WhiskyTHC is a mature Numerical Relativity code extensively used to model binary neutron star mergers. The code has been ported and used in intensive production phases on several machines allover the world, showing good parallelization performance within well-tested production setups. The future goal will be to perform a new set of high-fidelity, high-resolution simulations to study the impact of detailed microphysics on neutron star merger observables.
top

Studying competing and co-existing orders near quantum critical points using the conformal bootstrap

Project Name: Studying competing and co-existing orders near quantum critical points using the conformal bootstrap
Project leader: Dr. Chris Hooley
Research field: Fundamental Physics
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

The question of microscopic phase coexistence remains one of the most important in modern condensed matter theory. It occurs in many settings, from the relationship between charge-density-wave order and superconductivity in the cuprate superconductors to the question of supersolidity in liquid helium. Since this coexistence occurs near a quantum critical point, one might expect these questions to have a universal answer that depends only on the spatial dimensionality and symmetry properties of the system, independently of the microscopic details of the materials involved. This suggests a field-theoretic approach, and indeed there have been several works taking this approach over the past decade [1]. In many cases, the field-theoretic calculation is unambiguous, either ruling coexistence in or out on the basis of the interactions between the fluctuations of the two types of order under consideration. However, there are some combinations of symmetries, especially Z_2 x O(2) and O(2) x O(2), where the field-theory calculations do not completely answer the question. What should one do to understand these cases better? Our proposal is to concentrate not on the phases themselves, but on the multicritical points in the phase diagrams. These occur when one or more phase transition lines meet, and they should be described by a conformal field theory of the appropriate dimensionality and symmetry. If we were able to determine the operator spectrum of that conformal field theory, we could obtain unambiguous information about the number and nature of the phases that meet there, and thus potentially answer definitively the question of whether a microscopic coexistence phase is among them. Fortunately, exciting developments over the past decade have provided a tool that does just this: the conformal bootstrap [2]. It works by identifying when a possible combination of scaling dimensions for the operators in the theory is inconsistent, by showing in such cases that there can be no choice of coefficients in the theory’s operator product expansion that satisfies crossing symmetry, a requirement for a consistent conformal field theory. This is achieved by turning the crossing symmetry requirement into a semidefinite programme, which is solvable if and only if such a set of coefficients exists. [1] See, for example, A. Eichhorn, D. Mesterházy, and M. M. Scherer, Phys. Rev. E 90, 052129 (2014). [2] D. Simmons-Duffin, “The conformal bootstrap,” in J. Polchinski, P. Vieira, and O. DeWolfe (Eds.), “New Frontiers in Fields and Strings” (World Scientific, 2017).
top

Modelling the Mechanochemical Cycle of Cytoplasmic Dynein Machinery

Project Name: Modelling the Mechanochemical Cycle of Cytoplasmic Dynein Machinery
Project leader: Assist. Prof. Dr. Mert Gur
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

Dyneins are a family of AAA+ motors responsible for nearly all motility and force generation functions towards the minus-end of microtubules (MT). Because of their central roles in intracellular transport, cell division, and cilia; defects in dynein motility are linked to developmental and neurodegenerative disorders. Dynein forms a large (1.4 MDa) complex, the core of which consists of a catalytic ring of six AAA subunits. Conformational changes driven by ATP hydrolysis within the ring underlie dynein force generation and motion. Recent structural and biophysical studies have identified the major conformational states in distinct nucleotide states, but it remains unclear how key structural elements and biochemical states are synchronized in such a large molecule. This proposal aims to utilize all-atom Molecular Dynamics (MD) simulation of dynein under physiological conditions to reveal mechanochemical cycle of dynein. These simulations contain ~1 million atoms and need to run tens of µs; which require the usage of Tier-0 HPC systems to attain biologically relevant time scales. Specifically, simulations will reveal how nucleotide state of the AAA1 site reorganizes the AAA ring, controls the registry of the stalk coiled-coils that coordinate MT binding/unbinding, and triggers the powerstroke/recovery stroke of the linker that produces force and motion.
top

Scalability testing for atomistic molecular dynamics simulations of respiratory complex I

Project Name: Scalability testing for atomistic molecular dynamics simulations of respiratory complex I
Project leader: Dr. Vivek Sharma
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 50000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France, 50000 core hours on Joliot Curie (SKL) hosted by GENCI at CEA, France and 20000 core hours on JUWELS hosted by GCS at FZJ, Germany
Description

Energy production in cells takes place in mitochondria, which carry out a process called oxidative phosphorylation. In this process, electron transfer reactions are tightly coupled to proton pumping, which is responsible for the generation of energy in the form of ATP. One of the central enzymes that catalyze coupled electron/proton transfer is respiratory complex I. Recently, a number of high resolution X-ray and cryo EM structures of complex I have been solved, which provides us a platform to perform large-scale classical molecular dynamics simulations of complex I and understand its mechanism in full details, which would provide blueprints for futuristic drug design. In this project, we will perform scalability tests using GROMACS software, which is highly parallelized and highly efficient, on a large model system of mitochondrial complex I (1-1.5 million atoms).
top

Lipid droplet biogenesis: A molecular dynamics simulation study

Project Name: Lipid droplet biogenesis: A molecular dynamics simulation study
Project leader: Dr. Xavier Prasanna Anthony Raj
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain, 50000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France and 50000 core hours on Joliot Curie (SKL) hosted by GENCI at CEA, France
Description

Lipid droplets (LDs) are storehouses of neutral lipids in cells that serve as a primary source for energy production and provide the raw material for lipid biosynthesis. Lipid droplet biogenesis occurs at the endoplasmic reticulum (ER). The process begins with the spontaneous nucleation of neutral lipids/cholesterol esters into a “lens-shaped” aggregate which eventually develops into a mature LD. Several proteins are involved in regulating LD formation and growth. Although the function of most of these proteins is well known, their mechanism in LD biogenesis is not well established. In this study, we will examine the process of LD formation and growth using multi-scale molecular dynamics simulations and provide molecular details of the role of proteins in facilitating this process.
top

ai-GRALIB

Project Name: ai-GRALIB
Project leader: Dr. Marios Zacharias
Research field: Chemical Sciences and Materials
Resource awarded: 100000 core hours on MARCONI (KNL) hosted by CINECA, Italy
Description

In this project we aim to study and design novel graphene-based structures, and to elucidate the physical mechanisms that make them excellent candidates to serve as metal-host materials in rechargeable metal ion batteries. To achieve this task, we will develop a first-principles technique that relies on the newest ideas of density functional theory, and ab-initio molecular dynamics as implemented in state-of-the-art computational packages for materials science. In particular, ab-initio molecular dynamics will be employed to study thermodynamic stability of graphene-based architectures and metal-ion diffusion. Furthermore, temperature and quantum nuclear effects on the storage properties of these systems will be studied using large simulation cells in conjunction with the “Special Displacement Method” that involves the deterministic displacement of the atoms away from their crystallographic positions.
top

Scalability tests

Project Name: Scalability tests
Project leader: Prof. Alfredo Soldati
Research field: Engineering
Resource awarded: 50000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France
Description

The current project is intended to run weak and strong scalability analysis of our proprietary code. These analysis will be used for the 20th Prace call. This code performs Direct Numerical Simulation of multiphase flows in a rectangular channel with the eventual presence of a surface-active agent and has already been tested, benchmarked and run on several different architectures like Intel Xeon E5 (Marconi A1 at CINECA), Intel KNL (Marconi A2 at CINECA) and IBM BG/Q (Vesta at Argonne National Laboratory). For a complete scalability analysis, we expect that 200000 core hours will be needed.
top

FLUPS, a Fourier-based Library of Unbounded Poisson Solvers

Project Name: FLUPS, a Fourier-based Library of Unbounded Poisson Solvers
Project leader: Prof Philippe Chatelain
Research field: Engineering
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain and 20000 core hours on JUWELS hosted by GCS at FZJ, Germany
Description

The solution of unbounded 3D Poisson problems is a ubiquitous problem in computational physics as it concerns problems ranging from fluid dynamics to particle physics. Consequently, there is a widespread need for adapted numerical solvers. To date, numerous implementations exist to tackle (part of) the problem. Yet, none of them offer an integrated solution that combines the desired properties of versatility, performance, scalability, and portability. In order to fill that gap, we have developed FLUPS, a Fourier-based Library of Unbounded Solvers on uniform grids. The library is targeted for large computational architectures and addresses a great variety of 2D and 3D Poisson problems, with the capability of handling unbounded, semi-unbounded, symmetric and periodic boundary conditions. The 3D solver relies on the FFTW library to perform 1-dimensional Fourier transforms and it provides the needed interface to easily solve a 3D Poisson equation using any combination of boundary conditions among the 1000 possibilities (e.g. semi-unbounded in X, periodic in Y and even-odd in Z). Moreover, several Green’s functions are proposed, offering various convergence orders. We claim that our implementation is as fast as the renowned P3DFFT on classical FFT’s, yet much more optimized for the unbounded cases as we explicitly do not compute the useless FFT’s, dividing the computational cost by two with respect to an approach relying on the P3DFFT package. Moreover, we highlight that FLUPS is designed to be used as an external library with a simple interface oriented on the problem. The library takes care of memory management and parallel efficiency, and this is hidden to the user. Eventually, we intend to use FLUPS for large scale external aerodynamics and fluid dynamics problems, such as the study of turbine wakes and their interactions in wind farms. However, the library could be used in many other disciplines.
top

ESCS – Electromagnetic Simulations of Complex Structures

Project Name: ESCS – Electromagnetic Simulations of Complex Structures
Project leader: Prof. Franco Moglie
Research field: Engineering
Resource awarded: 50000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France
Description

Researchers in electromagnetic and stochastic computational techniques are involved in this project. Many researchers of the team are involved in the solution of electromagnetic compatibility problems, where incoherent radiators, semi coherent emitters and complex devices are quantified. Usually, the involved geometry is large and may have highly complexity in field pattern. Moreover, chaotic structures are investigated and the results can be obtained only as an ensemble average of simulations by changing geometrical parts or sources. Three dimension simulations of complex sources in complex environments require Tier-0 machines. All the participants of this team have a background in the parallel computation. The group of Ancona developed an FDTD code for the simulation of reverberation chambers; the group of Granada developed “UGRFDTD”, a general purpose (EMC-oriented) state-of-the-art FDTD solver. We participated to previous PRACE projects and domestic calls. The previous projects used the FERMI and Marconi clusters based at CINECA. Now CINECA is going to replacing Marconi cluster, therefore by the end of November the entire cluster will be shut down. The new machine will have a different architecture, based on GPUs. The porting of our codes on the new machine will not be end for the deadline of PRACE 20th Call. We would like to continue on a cluster with the same architecture of Marconi-KNL where our codes are optimized. Our codes are capable to simulate different geometries as set of stirrer angles in the reverberation chamber and complex sources for the propagation of the stochastic noise emissions. The code of Ancona was optimized for the FERMI and Marconi-KNL architectures during all the previous PRACE projects and the code of Granada was optimized for Marconi-KNL architecture during the PRACE 13th and 17th Regular project. The Ancona code is mainly divided in three modules: 1) an electromagnetic time domain solver; 2) a fast Fourier transform; 3) a statistical module to obtain the cumulative results. All the modules were previous optimized for high-performance parallel computers using hybrid method (MPI and OpenMP) and they was used successfully in the previous PRACE projects. The availability of a code, that solves the previous three steps in a unique job, makes the simulations very appealing. Moreover, the availability of an optimized simulation code will give the results in short time avoiding long measurement campaigns. The team was a part of the group of the COST Action IC1407, that began in April 2015 and ended in April 2019.
top

BITCART Scaling Study for WMLES of Complex Automotive & Aerospace Flows

Project Name: BITCART Scaling Study for WMLES of Complex Automotive & Aerospace Flows
Project leader: Dr. Neil ashton
Research field: Engineering
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

This project will focus on assessing the scalability of a high-order immersed boundary method code on PRACE machines. This will be used to then access for full project access to conduct large-scale (~1 billion cell on >20,000 cores) of wall-modelled Large-Eddy Simulation of complex automotive and aerospace flows. There is real need for WMLES approaches to bridge the gap between DNS (academia) and RANS (industry) methods and decrease the use of costly and time-consuming wind-tunnel and road/flight tests. These methods still rely on large-scale HPC resources which are only available on the PRACE network of systems.
top

HA0RES – Mechanisms of mutation resistance in hemagglutinin: Is the uncleaved form HA0 the right target?

Project Name: HA0RES – Mechanisms of mutation resistance in hemagglutinin: Is the uncleaved form HA0 the right target?
Project leader: Dr. Tiziana Ginex
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

Each year, influenza A and B viruses are responsible for 3-5 million severe cases and 290,000-650,000 deaths worldwide. Available pharmaceutical strategies consist on (i) vaccination, which is characterized by some limitations as suboptimal effectiveness, the need for annual injection and possibility of antigen mismatch, and (ii) neuraminidase inhibitors. This poor scenario highlights the need for new solutions that would complement and possibly enhance the efficacy of antiviral therapy. Embedded to the viral membrane, the trimeric hemagglutinin (HA) glycoprotein is an attractive drug target due to its critical and dual role in virus entry: i) virus attachment to host by recognition of sialylated cell surface glycans located at the HA globular head, and ii) host-guest membrane fusion thought pH-induced (~5) refolding of HA. Fusion pore formation then allows guest-to-host relocation of the viral material for replication. Several HA inhibitors, impeding virus attachment or membrane fusion, have been proposed so far. In this latter category, two major binding sites localized in the HA1-HA2 stem region have been identified and validated by structural data. However, the efficiency of these novel compounds is severely affected by the occurrence of resistance-associated mutations, which increase the virulence of the mutated strains. Understanding the molecular mechanisms for drug resistance is of utmost importance to prevent the burden of influenza-associated illness and death. In this context, attention has been focused on the effect of these mutations on the pH-promoted release of the fusion peptide in the active, fusion-competent (cleaved) form of HA. However, the insensitivity of certain mutated strains to sensible changes in the pH of acidification raises doubts about the specific molecular events implicated in virulence enhancement. We hypothesize that resistant mutations, at least in specific cases, may act at the level of the inactive, fusion-incompetent (‘uncleaved’) form of HA, denoted as HA0. In particular, our preliminary data suggest that some mutations may enhance instability at the level of the cleavage site of the HA0 precursor, enhancing protease susceptibility and hence HA activation. So, the main aim of this project is to properly tune and optimize parameters necessary to carry out an efficient mapping of the conformational and energetic costs that would lead to the hydrolytic step prior to the refolding of the fusion peptide. Enhanced sampling MD (as REMD and PT-MetaD) have been thus planned to properly explore the feasibility of this mechanism.
top

Testing scalability for intra- and inter- membrane interactions in health and disease

Project Name: Testing scalability for intra- and inter- membrane interactions in health and disease
Project leader: Dr. Anna Duncan
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

Biological membranes surround and define cells, separating then cell from its environment, and creating functionally important compartments with cells. Experimental methods reveal the structures of membrane components whilst computer simulations play a key role in allowing us to understand how these components are assembled to form a complex and dynamic functional membrane. Molecular dynamics simulations of biological membranes have now reached a scale and complexity which provides direct insights into the molecular basis of disease. We will examine dynamic membrane organization for two systems of biomedical relevance: host cell-viral pathogen interactions; and mitochondrial membrane defects. Interactions between and within complex membranes are central to these processes but in many such cases are difficult to address experimentally. Thus, molecular dynamics simulations provide a ‘computational microscope’ to gain molecular-level understanding and drive and guide further experimental and computational investigations. We wish to perform scalability tests with the intention to apply for full project access, focussing on two areas: inter-membrane interactions, as seen in host-pathogen interactions, by simulating association of host cell and viral membrane associations; and intra-membrane interactions, in defective mitochondrial membranes, by simulating the formation of respiratory supercomplexes in membranes containing cardiolipin. These simulations are of exceptionally large systems (several millions of particles). Even when simulated using the coarse-grained methods, the processes under investigation require several tens of microseconds per simulation. Therefore the use of Tier-0 is essential. The outcomes will be an unprecedented level of understanding, in molecular detail, of disease-causing processes, in their full physiological complexity.
top

FLUPS, a Fourier-based Library of Unbounded Poisson Solvers

Project Name: FLUPS, a Fourier-based Library of Unbounded Poisson Solvers
Project leader: Prof Philippe Chatelain
Research field: Engineering
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

The solution of unbounded 3D Poisson problems is a ubiquitous problem in computational physics as it concerns problems ranging from fluid dynamics to particle physics. Consequently, there is a widespread need for adapted numerical solvers. To date, numerous implementations exist to tackle (part of) the problem. Yet, none of them offer an integrated solution that combines the desired properties of versatility, performance, scalability, and portability. To fill that gap, we have developed FLUPS, a Fourier-based Library of Unbounded Solvers on uniform grids. The library is targeted for large computational architectures and addresses a great variety of 2D and 3D Poisson problems, with the capability of handling unbounded, semi-unbounded, symmetric and periodic boundary conditions. The 3D solver relies on the FFTW library to perform 1-dimensional Fourier transforms and it provides the needed interface to easily solve a 3D Poisson equation using any combination of boundary conditions among the 1000 possibilities (e.g. semi-unbounded in X, periodic in Y and even-odd in Z). Moreover, several of Green’s functions are proposed, offering various convergence orders. We claim that our implementation is as fast as the renowned P3DFFT on classical FFT’s, yet much more optimized for the unbounded cases as we explicitly do not compute the useless FFT’s, dividing the computational cost by two with respect to an approach relying on the P3DFFT package. Moreover, we highlight that FLUPS is designed to be used as an external library with a simple interface oriented on the problem. The library takes care of memory management and parallel efficiency, and this is hidden to the user. Eventually, we intend to use FLUPS for large scale external aerodynamics and fluid dynamics problems, such as the study of turbine wakes and their interactions in wind farms. However, the library could be used in many other disciplines.
top

RSNSE

Project Name: RSNSE
Project leader: Prof Pierluigi Maponi
Research field: Mathematics and Computer Sciences
Resource awarded: 50000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France and 50000 core hours on Joliot Curie (SKL) hosted by GENCI at CEA, France
Description

The present project is the continuation of a research which was partly supported by the European PRACE Project n. 2015133169, and is related to the question of the existence of finite-time singularities (blow-up) for the solutions of the incompressible Navier-Stokes (NS) in R^3 in absence of boundary conditions and external forces. The starting point of the research is a paper of Li and Sinai [LI, D. \& SINAI, YA. G. (2008) Blowups of complex solutions of the 3D Navier-Stokes system…. ( J. Eur. Math. Soc., 10, 267-313]. They write the NS system as an integral equation in Fourier k-space and introduce a class of solutions for which the support in k-space is contained inside a thin cone around a fixed axis. The energy transfer to the high |k| modes is controlled by a fixed point (depending on the initial data) of a map related to the flow of energy along the cone. For a particular choice of the fixed point Li and Sinai proved the existence of complex solutions that become singular (blow-up) at a finite time. For a better understanding of the circumstances of the blow-up we resorted to computer simulations, with the help of a special computation program, created for simulating the solutions of the integral equation in k-space, which takes into account the particular structure of the support of the Li-Sinai solutions. The program worked very well, and it allowed, for some choices of the parameters, to follow the solutions up to times close to the blow-up. The purpose of the present project is to simulate real flows, related to the complex solutions of Li and Sinai (they are obtained by antisymmetrizing the initial data of the complex solutions that blow up). The theoretical analysis of the real solutions faces some difficulties, and the computer simulations can provide a better insight. The structure in k-space of such solutions is also concentrated around a fixed axis, and the program that we prepared also works well, with small modifications. Results of simulations of a real solution related to the fixed point for which the blow-up is proved was recently published (arhiv). The solution behaves for some time very much “tornado-like” (except that there is no rotation): it exhibits an increase of the enstrophy, and also of the maximal values of the vorticity and velocity, with concentration in a finite region, but finally decreases. It appears that, due to the fixed point, which has axial symmetry, the solution develops an approximate axial symmetry, and for axial symmetric flows there is n blow up. We plan therefore to simulate real flows associated to a few fixed points which have no axial symmetry, and for which a complex blow-up is not proven, but is expected. We apply for this preparatory call to have data for the next regular call
top

Scaling test of TurboRVB on MareNostrum for Call 20 Proposal SHIVA-QMC, 2019215068

Project Name: Scaling test of TurboRVB on MareNostrum for Call 20 Proposal SHIVA-QMC, 2019215068
Project leader: Prof. Philip Hoggan
Research field: Chemical Sciences and Materials
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

The Technical review of PRACE proposal 2019215068 at BCN has requested scaling information on the software TurboRVB on MareNostrum. As PI of proposal 2019215068, I understand that, without this data, the use of TurboRVB on MareNostrum will not obtain approval in the technical review.
top

Opto-Electronic properties of epitaxial Ge-Si-C superlattices

Project Name: Opto-Electronic properties of epitaxial Ge-Si-C superlattices
Project leader: Dr. Carlos Loia Reis
Research field: Fundamental Physics
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

Computational materials science and engineering is of crucial importance to a number of industrial applications, including one of the most important industries: CMOS-based technologies. CMOS technology is based on Si, an indirect band gap material with low photo absorption and emission efficiencies, which are the key barriers to increase performance and functionality of technologies such as: photodiodes in high-efficiency photovoltaic cells and CMOS image sensors, LEDs and Lasers for intra- and inter-chip optical interconnects. The goal of this project is the simulation of CMOS-compatible band structure engineered materials (i.e., electronic metamaterials) whose optoelectronic properties are not found in naturally-occurring materials, and which can be achieved with semiconductor superlattices. With self-limiting atomic layer epitaxy methods in conjunction with suitable chemical precursors, the number of atomic planes and their composition can be extremely well-controlled. Intense R&D on the monolithic integration of III/V compound semiconductors in leading edge CMOS has not yet been successful. Only Group-IV elements and their alloys are currently in use, or are projected to be used in Nanosheet MOSFETs (DOI:10.1109/MSPEC.2019.8784120). Group-IV superlattices that can be grown pseudomorphically on top of a variety of silicon crystallographic surface orientations, are the most promising solutions for a straightforward integration of optoelectronic metamaterials with CMOS. The proposers’ software was used to discover a first set of Si-Ge-C superlattice compositions possessing direct band gaps (DOI:10.1016/j.sse.2015.01.019), and since then several superlattice compositions were discovered to be Topological Materials (unpublished results). Topological Materials can have strong non-linear effects for light-matter interactions and can also be used as building blocks for Quantum Sensing and Topological Quantum Computing. Thickness and lateral dimensions of the materials used in advanced CMOS technologies are extremely reduced, and cannot be modelled by macroscopic models, thus requiring ab-initio quantum mechanical modelling and the tools of modern electronic structure theory. The proposers will improve their electronic structure codes and methods and will try to bridge the “scale gap” between the microscopic scale of quantum mechanical calculations and the mesoscopic scale of device simulations. The proposers developed a chain of simulation tools, starting from the construction of a periodic unit cell model of a given superlattice, followed by a first principles optimization of the cell geometry in order to obtain relaxed atomic positions and lattice parameters under the constraint that it should be pseudomorphic on a given surface. For this relaxed cell geometry, an accurate band structure is calculated, the band edges identified, and the optical matrix elements are calculated to distinguish the cases where these elements are large and therefore the material could have interesting opto-electronic properties. This project will allow a systematic study of Group-IV superlattices, to identify trends with layer composition, number and thickness of layers, or substrate orientation, seeking peculiar features of the band structure that indicate interesting opto-electronic and topological properties, and that would make a superlattice desirable for integration in devices such as photo-diodes, LEDs, Laser advanced transistor concepts, topological qubits and quantum sensors.
top

Scalability TEst MultiNest

Project Name: Scalability TEst MultiNest
Project leader: Dr. Yuxiang Qin
Research field: Universe Sciences
Resource awarded: 100000 core hours on MARCONI (KNL) hosted by CINECA, Italy
Description

To take advantage of the physical bounty of the cosmic 21-cm signal, we must simulate how the unknown properties of the first galaxies imprint themselves on cosmological scales. The computational challenges of this enormous range of relevant scales and astrophysical uncertainties motivated the development of more approximate, fast, so-called semi-numerical simulations. Our public simulation codes, 21cmFAST (Mesinger et al. 2007; 2011) and associated MCMC sampler 21CMMC (Greig et al. 2015; 2017), have become the world standard in this field, and are being used by all 21-cm interferometers for parameter inference. Our MCMC employs the EMCEE sampler (Foreman-Mackey et al. 2013), which supports only shared memory multiprocessing and has difficulties reaching convergence in high dimensional parameter space (e.g. Binnie et al. 2019). With more physics being constantly added to our simulation (Park et al. 2019; Qin et al. in prep.), the parameter space of unknowns has become very large: 15 astrophysical parameters + 6 cosmological parameters. It is necessary to implement an alternative sampler, which has the advantage of fast convergence and also supports MPI, allowing it to be run on large CPU clusters. To this end we recently implemented the MULTINEST (Feroz et al. 2007; 2008; 2013) sampler in 21CMMC. MULTINEST is based on nested sampling and can directly compute the Bayesian evidence we need for model selection. This preparatory project is to test the scalability of our sampler using MULTINEST. We will be varying the number of cores as well as the number of parameters. After exploring the scalability, we will apply for a Tier-0 project to explore the full parameter of our 21-cm simulations, quantifying constraints on the first galaxies achievable with the Square Kilometre Array (SKA).
top

Demonstrating the scalability of probabilistic seismic tomography to study the interior structure of volcanoes

Project Name: Demonstrating the scalability of probabilistic seismic tomography to study the interior structure of volcanoes
Project leader: Dr. Corinna Roy
Research field: Earth System Sciences
Resource awarded: 50000 core hours on MareNostrum hosted by BSC, Spain
Description

Signals from earthquakes contain a wealth of information that enables seismologists to determine structures and processes at all depths in the planetary interior. Knowing the subsurface seismic velocity structure, the earthquake locations and their tradeoffs is important for various research questions. For example, earthquakes induced through hydro-fracturing need to be precisely located in order to calculate their local magnitude and decide whether to stop operations or proceed. Natural seismicity can be used to track fluid migration in volcanoes and image their subsurface structure, however, earthquake locations depend on the seismic velocity model used, which is rarely known. This project is part of a larger project to develop methods to estimate jointly the seismic velocities of the subsurface and the earthquake location using Bayesian inference. The fully non-linearised approach allows us to calculate true tradeoffs between earthquake parameters and subsurface properties, yielding probabilities that an event occurs in a certain location, and above a certain magnitude and their uncertainties. Crucially, with our approach we can for the first time place rigorous error bounds on the seismic and hence physical structure of such settings, and truly test which physical processes can give rise to the images we retrieve. The Markov chain Monte Carlo (McMC) code has been used successfully for smaller-scale problems with around 100 earthquakes recorded at a handful of stations. However, to reveal more about the tectonic, volcanic and microseismic processes at play in the Earth, more ambitious datasets must be tackled. We aim to use thousands of earthquakes to image the subsurface under an active volcano in the central Main Ethiopian Rift where rigorous debate has yet to settle whether the source of ground uplift and subsidence is related to thermal, fluid of magmatic effects. Therefore, we need to establish the scalability of the code over a short period of time before running at scale. This project will establish the scaling of the code and permit us to make small algorithmic optimisations to ensure maximal use is made of larger computational resources in the future. The main objective of the project is to establish the scalability to Tier-0 computational resources of a non-linear transdimensional seismic traveltime tomography algorithm which has been developed by us.
top

Polymer material physics.

Project Name: Polymer material physics.
Project leader: Dr. Carsten Svaneborg
Research field: Fundamental Physics
Resource awarded: 20000 core hours on JUWELS hosted by GCS at FZJ, Germany
Description

The Kremer-Grest (KG) model is the standard Molecular Dynamics model for studying static and dynamic properties of generic polymer materials. Having recently learned how to match the KG model to specific real polymers and secondly convert simulation units into polymer specific SI units. Our aim is to design and run highly entangled computational polymer model melts (and networks) matching specific systems for which experimental results for instance macroscopic and mesoscopic properties such as shear-relaxation moduli, small angle scattering form factors, NMR and so on has been performed and reported. We expect the simulations to produce reference configurations that matching all experimental results, and from which we can predict the result of any experiment not yet performed, and furthermore data from which we can debug existing theories of polymer physics.
top

Type B: Code development and optimization by the applicant (without PRACE support) (3)

DAFNE project: A Decision-Analytic Framework to explore the water-energy-food NExus in complex and transboundary water resources systems of fast growing developing countries.

Project Name: DAFNE project: A Decision-Analytic Framework to explore the water-energy-food NExus in complex and transboundary water resources systems of fast growing developing countries.
Project leader: Dr. Jazmin Zatarain Salazar
Research field: Engineering
Resource awarded: 200000 core hours on Joliot Curie (SKL) hosted by GENCI at CEA, France
Description

The DAFNE project is a H2020 EU project targeted to establish a decision-analytic framework for the exploration of the Water-Energy-Food (WEF) nexus in transboundary river basins. DAFNE integrates multiple and diverse international and local academic expertise from natural sciences, water engineering, environmental economics, as well as water governance and law experts in order to facilitate social understanding of the impact and support comparative analysis of alternative development pathways. The framework will be tested in two pilot case studies in rapidly growing African regions, the Zambezi and the Omo Rivers basins. The DAFNE project integrates a high level integrated water resources management approach, which addresses the WEF nexus explicitly and from a novel perspective, and aims at promoting a green economy in regions where infrastructure development and expanding agriculture have to be balanced with the local social, economic, and environmental dimensions. DAFNE quantify and analyze the WEF nexus with respect to the trade-off between conflicting objectives, such as hydropower production vs. irrigation, and exploitation vs. conservation. The nexus will also be translated into use and non-use economic values and impact on terrestrial and aquatic ecosystems, also taking into consideration ecosystem services. For more information about the project please refer to http://dafne-project.eu/.
top

Development and optimization of algebraic multigrid domain decomposition (AMG-DD) on GPU clusters

Project Name: Development and optimization of algebraic multigrid domain decomposition (AMG-DD) on GPU clusters
Project leader: Dr. Wayne Mitchell
Research field: Mathematics and Computer Sciences
Resource awarded: 100000 core hours on Piz Daint hosted by CSCS, Switzerland
Description

Algebraic multigrid (AMG) is a widely used scalable solver and preconditioner for large-scale linear systems resulting from the discretization of a wide class of elliptic PDEs. While AMG has optimal computational complexity, the cost of communication has become a significant bottleneck that limits its scalability as processor counts continue to grow and as GPU acceleration becomes more prevalent on modern machines. This project will continue the development of a novel, communication-avoiding algorithm, Algebraic Multigrid Domain Decomposition (AMG-DD). The goal of AMG-DD is to provide a low-communication alternative to standard AMG V-cycles by trading some additional computational overhead for a significant reduction in communication cost. Previous testing of AMG-DD through PRACE Type A access to the Piz Daint cluster showed that this tradeoff can result in faster runtimes for AMG-DD compared to AMG. Particularly on large clusters of GPUs, such as Piz Daint, local computation is very cheap compared with communicating between GPUs on different nodes, leading to superior performance for AMG-DD. As such, future development and optimization of AMG-DD will target these architectures. This project aims to further develop and optimize the AMG-DD code with two major goals in mind. First, the setup phase of the code will undergo major optimization efforts and be ported to the GPU, as the current unoptimized, CPU-only version of the code was shown to perform very poorly during the PRACE Type A access scaling studies. Second, algorithm design and parameters will be further explored in order to enhance performance and convergence behavior of AMG-DD on a wider range of problems.
top

Monitoring proteins conformational changes through multiscale simulations

Project Name: Monitoring proteins conformational changes through multiscale simulations
Project leader: Dr. PeDr.o Ojeda May
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 100000 core hours on Piz Daint hosted by CSCS, Switzerland
Description

Resistance to certain types of drugs can be linked to large conformational changes in protein kinases where flexible regions interconvert from an active to an inactive conformation upon signaling, for instance phosphorylation. Understanding the mechanism of interconversion at the molecular level can assist the process of discovering better drug candidates. We plan to perform string method (SM) molecular dynamics (MD) simulations to cover the whole landscape between active and inactive conformations in order to have a better understanding of the interconversion mechanism.
top

Type C: Code development with support from experts from PRACE (2)

Load Balancing of Molecular Properties Calculations In VeloxChem Program

Project Name: Load Balancing of Molecular Properties Calculations In VeloxChem Program
Project leader: Dr. Zilvinas Rinkevicius
Research field: Chemical Sciences and Materials
Resource awarded: 100000 core hours on SuperMUC hosted by GCS at LRZ, Germany
Description

The quantum chemical computations of molecular properties are one of the key components of various spectroscopies. They are extensively applied to study various molecular materials to gain a microscopic understanding of physical processes driving material response and to develop “structure-to-property” relationships used in the search for novel molecular materials. Among the quantum chemistry methods, the density functional response theory (DFRT) approaches are most frequently employed to investigate various linear and non-linear properties of molecular materials due to good ration between accuracy and computational cost. Unfortunately, DFRT approaches become computationally expensive for extended molecular systems, which contain over few hundreds of second-row atoms. To overcome this limitation of DFRT approaches, we recently developed a new quantum chemistry program VeloxChem (https://veloxchem.org) [1], which is designed to effectively perform DFRT computations of various molecular properties in HPC environments. Currently, the VeloxChem program can effectively perform DFRT computations on systems up to 500 atoms using up to 16 000 CPU cores. In this project, we aim to enhance the scalability of DFTR computations in the VeloxChem program up to 100 000 CPU cores, and in this way, enable routine investigations of molecular properties in systems consisting of 800+ second-row atoms. This improvement will allow us to carry out investigations of the optical properties of realistic metal and metal oxide nanoparticles. References: [1] Zilvinas Rinkevicius, Xin Li, Olav Vahtras, Karan Ahmadzadeh, Manuel Brand, Magnus Ringholm, Nanna Holmgaard List, Maximilian Scheurer, Mikael Scott, Andreas Dreuw, Patrick Norman. VeloxChem: a Python-driven density-functional theory program for spectroscopy simulations in high-performance computing environments.”, WIREs Computational Molecular Science, 2019, p. e1457.
top

Intelligent HTC for Committor Analysis

Project Name: Intelligent HTC for Committor Analysis
Project leader: Dr. Alan O’Cais
Research field: Biochemistry, Bioinformatics and Life sciences
Resource awarded: 200000 core hours on Joliot Curie (KNL) hosted by GENCI at CEA, France, 200000 core hours on Joliot Curie (SKL) hosted by GENCI at CEA, France and 50000 core hours on JUWELS hosted by GCS at FZJ, Germany
Description

Committor analysis is a powerful, but computationally expensive, tool to study reaction mechanisms in complex systems. The main goal of this project is to facilitate an implementation of committor analysis in the software application OpenPathSampling (OPS, http://openpathsampling.org/) that is performance portable across a range of HPC hardware and hosting sites. As an example system, we will look at an oncogenic mutation of the protein KRas, which plays an important role in cell growth and proliferation. In its active (GTP-bound) state, KRas activity is mediated by a dynamic equilibrium between a more structured state that can activate the relevant signaling pathways, and a less structured state that cannot. Mutations of KRas can shift that equilibrium toward the structured state, causing the mutant to be more active than the wild type. This in turn leads to too much cell growth and proliferation: that is, it leads to cancer. Committor analysis is essentially an ensemble calculation that maps straightforwardly to an HTC workflow, where typical individual tasks have moderate scalability and indefinite duration. Since this workflow requires dynamic and resilient scalability within the HTC framework, OPS is coupled to a custom HTC library (jobqueue_features, https://github.com/E-CAM/jobqueue_features) that leverages the Dask (https://github.com/dask) data analytics framework and can manage MPI-aware tasks. Across scientific fields, High Throughput Computing (or HTC) is becoming a necessary approach in order to fully utilize next-generation computer hardware. As an example, consider molecular dynamics: excellent work over the years has developed software that can simulate a single trajectory very efficiently using massive parallelization. Unfortunately, for a fixed number of atoms, the extent of possible parallelization is limited. However, many methods, including semi-classical approaches to quantum dynamics and some approaches to rare events, require running thousands of independent molecular dynamics trajectories. Intelligent HTC, which can treat each trajectory as a task and manage data dependencies between tasks, provides a way to run these simulations on hardware up to the exascale, thus opening the possibility of studying previously intractable systems. The project targets porting the custom HTC library to a wider variety of HPC platforms, specifically additional resource managers and MPI runtimes. Both OPS and jobqueue_features are Python-based and we will also develop more sophisticated support for Python-based tasks, including direct access to the memory space of executed tasks to avoid the use of filesystems for data transfer. We will also investigate the use of UCX protocol for communication in the HTC framework to reduce the overall overhead of the HTC framework.
top