ICHEC Annual Review 2009

Annual Review

2009

Annual Review

2009

GalwayInformation Technology BuildingNational University of Ireland, GalwayIrelandTel: +353 91 495305Fax: +353 91 495573Email: [email protected]: www.ichec.ie

DublinTower BuildingTrinity Technology & Enterprise CampusGrand Canal QuayDublin 2IrelandTel: +353 1 5241608Fax: +353 1 7645845Email: [email protected]: www.ichec.ie

Ireland’s EU Structural FundsProgrammes 2007 - 2013

Co-funded by the Irish Government and the European Union

ANNUAL REVIEW 2009 PAGE 1

Table of Contents

1 Executive Summary 2

Chairman’s Introduction 2

Director’s Introduction 3

2 Service Provision and Research Enablement 4

2.1 Main Achievements for 2009 4

2.1.1 Overview 4

2.1.2 National HPC Infrastructure 4

2.1.3 National Service 6

2.1.4 Visualisation Facilities at ICHEC 8

2.1.5 International Initiatives 11

2.1.6 Education and Outreach 12

2.1.7 e-INIS 14

2.1.8 Met Éireann Collaboration 14

2.1.9 Research Activities 16

2.1.10 Other Developments 17

2.2 Looking Forward 18

3 Financial Summary 19

4 ICHEC Support for Projects 21

4.1 Class A Projects 23

4.2 Class B Projects 41

4.3 Class C Projects 87

5 Appendices 90

5.1 Appendix 1 - People and Partnerships 90

5.2 Appendix 2 - Outputs 91

5.3 Appendix 3 - The Condominium Cluster Model 95

ANNUAL REVIEW 2009 ANNUAL REVIEW 2009 PAGE 3

Director’s Introduction

The last year, 2009, was a very significant one for ICHEC. Funding to sustain our operations to end-December, 2011 was secured in late 2008 and a new state-of-the-art capacity cluster, Stokes, was commissioned. The new National HPC Service officially opened on 1st January 2009, bringing low latency computing to researchers in all of the institutions. Additional staff were recruited to accelerate community development and broaden the range of scientific disciplines we support. In January, 2009, ICHEC moved to new office space in Dublin (the National Hub for high-end computing), where its open plan structure has provided a particularly attractive environment for visiting researchers to engage with ICHEC staff in code development and optimization. Finally, a 3-D immersive visualization service with full staff support was opened in the Hub.

In this Annual Review we highlight a number of significant developments:

We engaged with other HPC centres in the EU through participation in projects such as DEISA and PRACE, in the case of the latter helping to secure 8.7 million core-hours for ICHEC users on some of the EU’s most advanced compute systemsWe initiated a programme in GPGPU computing and in this regard began a significant engagement with the leading manufacturer of graphics chips, NVidia as well as with HPC vendor, SGI in this areaWe established a significant presence in the area of climate modeling, working with researchers in two HEIs (UCD and NUIM) along with Met ÉireannWe developed a strategy for technology transfer and innovation enablement particularly with the SME sector in collaboration with other EU centres that have well established credentials in this areaOur partnership with Met Éireann continues to grow and develop leading to a successful joint application for funding to the Environmental Protection Agency for a Research Fellowship in Climate modeling

It is now clear that ICHEC can justifiably claim to have achieved a standard of excellence in the provision of high-end computing resources to the research community that is comparable with its peers in the EU and beyond.

James SlevinDirector

Chairman’s Introduction I am very pleased to introduce the 2009 ICHEC Annual Review.

ICHEC made great progress in 2009, and ICHEC is now widely seen as a stable, well-run, highly competent, and user-friendly component of the Irish shared services research infrastructure.

From an Oversight Board perspective, the development of a formal Vision and Mission statement for ICHEC, along with an Overall Strategy and a 3-year Strategic Plan, were the most important milestones achieved in 2009. Following from these, the Director and his staff have developed a 3-year rolling operational plan for ICHEC.

The ICHEC Vision is that the Irish research community will be among the leaders in Europe in the use of high-performance computational resources in support of research, and that ICHEC will be an internationally recognised, independent1 , centre for high-end computing, providing a first class service to Irish researchers, and reaching out to and partnering with other similar centres in the EU and beyond, and making a significant contribution to economic development.

The Mission statement is that ICHEC will provide education, expert training, and support to enable the Irish research community to best exploit local, national and international high-performance computational (HPC) resources, and make appropriate levels of national HPC resources available to Irish researchers, and through technology transfer to best contribute to the Irish national Smart Economy initiatives. Research enablement – enabling Irish researchers to undertake world competitive research using high performance computing tools – is what drives ICHEC.

ICHEC’s Overall Strategy to deliver on its Mission is to focus on:

support,

world-class research across all the main disciplines and institutions, 2 ICHEC Science Council peer review, to a

balanced set of HPC resources providing general purpose, scalable mid-level, and high-end “Grand Challenge” HPC resources3 ,

organisations, Irish industry and IDA supported foreign direct investment.

ICHEC’s strategy is driven by and responsive to the needs of Irish research users who have access to the ICHEC Oversight Board through the ICHEC Users Council.

I would like to pay tribute to my colleagues on the ICHEC Oversight Board for their input in helping shape the development of the organisation, to the ICHEC Science Council and ICHEC Users Council for their support, to the ICHEC staff, and, especially, to Dr. J-C Desplat, Associate Director, and to Professor Jim Slevin, Director, for their tireless work to ensure the continuing success of the ICHEC services to Irish researchers. I would also like to thank NUI, Galway for its continuing support in hosting ICHEC on behalf of the research community, and to thank Science Foundation Ireland and the Higher Education Authority for their continuing support.

Dennis JenningsChairmanICHEC Oversight Board

1 Executive Summary

1. Independent, in this context, means independent of any individual university or research institution.

2. Independent, in this context, means independent of the Oversight Board, management and staff of ICHEC.

3. General Purpose, scalable Mid-Level, and high-end “Grand Challenge” are relative terms that change by approximately a factor of 10

every 3 years as the computing and communications technologies improve. Scalable mid-level systems are those that support codes

that scale to thousands of processors.


System Schrödinger Lanczos Stokes Stoney

Service Capability Capability National National

Manufacture / Model IBM Blue Gene/P

IBM Blue Gene/L

SGI Altix ICE 8200X

Bull Novascale R422-E2

CPU PowerPC 450 PowerPC 440 Intel Xeon E5462

Intel Xeon X5560

CPU Clock 850 MHz 700MHz 2.8 GHz 2.8 GHz

CPU Cores 4096 2048 2560 512

Memory 2048 GB 1024 GB 5120 GB 3072 GB

Peak Performance 13.93 TFlop 5.73 TFlop 28.67 TFlop 5.73 TFlop

Linpack Performance 11.11 TFlop 4.74 TFlop 25.11 TFlop 5.14 TFlop

Interconnect Blue Gene Tree/ Torus

Blue Gene Tree / Torus

ConnectX Infiniband (DDR)

ConnectX Infiniband (DDR)

Storage (Formatted Capacity)

33 TB (shared) 33 TB (shared) 84 TB 21 TB

Launched Jan 2008 Jan 2008 Jan 2008 Jul 2009

2.1 Main achievements for 20092.1.1 Overview

The year 2009 saw the transition from the original SFI grant awarded in 2005, to the three-year SFI award to December 2011. In 2009, this grant provided funding for a total of 14 staff, distributed as

3 systems administrators, 6 computational scientists , a business development post,2 management positions (Director and Associate Director), and 2 administrative positions (Head of Admin and PA to the Director)

Most of these positions were filled by extending the contract of staff employed under the previous SFI grant, but it also allowed us to appoint a PA to the Centre Director, two additional Computational Scientists, and a Business Development post . This funding came in addition to the existing funding awarded through the HEA-funded PRTLI4 project, e-INIS which funded both staff (6 FTEs) and equipment.

The transition between the two grants proved remarkably seamless both for ICHEC personnel and for users of the National Service.

2 Service Provision & Research Enablement

2.1.2 National HPC Infrastructure

The installation of new equipment, financed under the HEA PRTLI4 e-INIS project, was completed in Q4-2008 with SGI winning the contract to provide a 2,560 core cluster, the SGI Altix ICE 8200EX. Both UCD and NUIM contributed additional funds to increase the capacity of this cluster, in exchange of which they are receiving commensurate levels of resources on the cluster. This scheme, known as the Condominium Cluster (or “Condo” in short), is described in Appendix 3.

The National Service was officially opened on this system named “Stokes” on 1st January 2009, and we are pleased to report unprecedented levels of availability (98.7%) and serviceability (99.7%). Note that as the Met Éireann national weather forecast is also run on Stokes, users of the National Service benefit from the 365/365, 6am-midnight cover, which effectively is a near mission critical service.

1

2

5

9

14

20

4

8

13

19

7

12

18

11

1716

3

6

10

15

21

ICHEC Sta!

1. James Slevin

2. J-C Desplat

3. Gráinne MacNamara

4. Niall Wilson

5. Kashif Iqbal

6. Ivan Girotto

7. Bruno Voisin

8. Marc Doumayrou

9. Simon Wong

10. Alin Elena

11. Alistair McKinstry

12. Adam Ralph

13. Asier Roa Mendiguchia

14. Rosemarie Lalor

15. Shiyu Wang

16. Honore Tapamo

17. Michael Browne

18. Gilles Civario

19. Eoin McHugh

20. David De La Harpe Golden

21. Christos Kartsaklis

Niall WilsonICHEC Infrastructure Manager

A second cluster, named Stoney, was added to the National HPC Infrastructure later in the year. Dr. Andy Shearer (NUIG) was responsible for the procurement of this new and important facility and has arranged for it to be hosted in the School of Physics’ machine room in NUI, Galway. ICHEC has agreed with Dr. Shearer to administer and manage the system and make it available to the National Service. A National Service has been provided on this system from 1st August 2009.

This system is a Bull Novascale R422-E2 InfiniBand cluster with 512 cores of the (then) new Intel Nehalem processor. Each of the 64 nodes comprises eight cores and 48GB of memory. Due to the much improved memory bandwidth capabilities of the Nehalem architecture and the 48GB of memory on each node, this system will offer a more suitable platform than Stokes for projects which have particularly high memory requirements. Another novel feature of the cluster is that it uses water cooled doors at the rear of the racks to more efficiently dissipate the heat produced by the densely packed high power equipment: these doors also reduce the running costs by 30%. The purchase of Stoney was funded by the Higher Education Authority through the PRTLI-4 e-INIS project. This grant was awarded to NUI Galway who in turn provides the system for national use.

Since its introduction to the National Service in August 2009, the Stoney system has been very stable with 100% system availability.

Finally on the HPC side, ICHEC continues to operate the National Capability Computing service on the IBM Blue Gene /L (“Lanczos”) and /P (“Schrödinger”), with access to large-scale overseas IBM Blue Gene facilities at the IBM Rochester and Watson laboratories.

The BlueGene /L and BlueGene /P were targeted at a small number of capability computing projects. The main system – Stokes – satisfied the requirements of the majority of HPC users. Despite its smaller size, Stoney’s Nehalem processors and larger per node memory means this system fulfills a complementary role for memory intensive applications. Full technical details for all four systems are included in the table to the right.

Our hardware infrastructure also includes a 3D immersive visualization environment and access to the distributed PB national storage infrastructure deployed as part of the e-INIS project (See section 2.1.7).

The Stokes cluster (L-R: front of the cluster, APC Cube hosting environment at UCD, compute blade with 2 Intel Harpertown quad-core processors)

The Stoney cluster (L-R: front, radiator on the water-chilled door, and compute node)

Figure 1: HPC Infrastructure 2009


2,531,255

250,394

1,144,191

55,176

372,152

519,092 1,681,938

135,265

166,300

2,320

419,021

802,391

350,613

167,201

8,588,739

DCU DIAS NUIG NUIM TCD

Tyndall

NUIMCondo

UCC

UCDCondo

UCD

EC Earth

Others

Met Eireann

ICHEC

Teagasc

production jobs through the queues). A total of 108 projects from 10 institutions have been given access to the infrastructure. The actual breakdown of usage per scientific area and institutions can be seen in the figures below.

As of 31st December 2009, the distribution of projects per category (Class A, B, or C) was of 7, 45 and 56 respectively (see Figure 5).

The projects supported as part of the National Service are looked at in more detail in Section 4 of the Review, “ICHEC Support for Projects”.

The popularity of the National HPC Service (as confirmed by the National Survey of HPC Users in Ireland, 2009 Edition) can be attributed in large part to the quality of the human component of the infrastructure. For instance, although the ICHEC helpdesk had to cater for 550 issues in 2009 (up from 387 in 2008), the average time to resolution has decreased to only 15 hours in 2009 (down from 22.9 hours in 2008), see Figure 6 below.

The recruitment of additional Computational Scientists has allowed ICHEC to implement more fully the specialised support model advocated since its creation in 2005, whereby ICHEC staff engage in collaborative research (i.e. take part in the scientific exploitation of some of the work he/she will be facilitating). This model enables them to them to have a better understanding of the specific domain and has lead to joint publications in the past. The continuity of interaction between the research group and the Computational Scientist fosters trust and long term planning, key assets in software development. The model has consistently proven to be particularly effective for high impact research.

Bringing expertise to the community has a number of additional of benefits. The most immediate is to facilitate and accelerate the development of major multi-disciplinary initiatives (e.g. Systems Biology), or allow the Irish community to join international initiatives on a much stronger footing (e.g. the Irish involvement in CECAM or EC-Earth), as the type of expertise developed by ICHEC is highly valued in these communities. As ICHEC is a multidisciplinary organisation by nature, its involvement in such initiatives facilitates effective communication across various disciplinary areas of the project.

Chemistry 22

2

2

1

1

1

1

5

5

5

3

510

0

0

0

6

8

13

13

13

14

ComputerScience 6

EarthScience 6

Engineering 11

Life Sciences 18

Physics 29

Others 16

Class A (7) Class B (45) Class C (56)

Figure 2: Utilisation profile (Stokes cluster, 2009)

2.1.3 National Service

The official opening of the National HPC Service on Stokes took place on 1st January 2009. As can be seen from the graph below, utilisation quickly reached high levels, to reach 78.3% over the year, including the ramping up period in Q1-09. We expect utilisation to settle at 85% and contention to build up progressively in subsequent years.

The number of users registered reached 410 by year end, with 157 of those users being active (e.g., having submitted

Figure 3: Utilisation per PI affiliation (Stokes, in core-hours)

Figure 5: Distribution of project categories per scientific area

Figure 6: Number of issues answered per calendar-month, and

response time

>48 Hours

<48 Hours

<24 Hours

Figure 4: Utilization per scientific discipline (Stokes, in core-hours)

Astrophysics Chemistry Computer Science

Course Testing NUIM Condo

Earth Sciences Engineering Life Science

UCD Condo Met Eireann Hirlam

Mathematics Physics Others

EC-Earth Oasis

5,941,800

55,176382 (Course)371,251519,092

1,681,938

104,46161,839

135,2651,193,227

5,681,859

482,73428,813 (Mathematics)369,303

103,796

455,112

0 (Oasis)


Principle Investigator Scientific area Institution Description

Dr. Mary Hearne NLP DCU Code development and parallelisation

Dr. Jiri Vala QMC NUIM Code development, parallelisation, optimisation for terascaling

Prof. Chris Bean Geology & Geophysics UCD Code development, parallelisation and optimisation for terascaling

Prof. Paul Mc Keigue Bioinformatics UCD Code development, parallelisation,

Prof. Stefano Sanvito Life Sciences. & Nanotech TCD Code development and optimisation, support for terascaling

Dr. Kingshuk Roy Chowdhury Applied Mathematics UCC Code development and parallelisation

Dr. Indiana Olbert Marine Science NUIG Code development, parallelisation and optimisation

Dr. Nathan Quinlan Biomedical Engineering

NUIG Code development, parallelisation and optimisation

Prof. Heather Ruskin Life Sciences DCU Debugging and deployment

Dr. Turlough Downes Astrophysics DCU Debugging and terascaling

Dr. Marcus Greferath Mathematics UCD Parallelisation and optimisation

Dr. Damien Thompson Nanoelectronics Tyndall Institute Programme of work to be defined by PI. So far, installation of applications packages

Dr. Niall English Nanoparticles UCD assistance in code installation, script development, job creation and adaptation, data preparation and analysis

Prof. Gary McGuire Mathematics UCD Parallelisation and optimization on multiple platforms

Prof. Giovanni Ciccotti Protein Molecular Dynamics UCD Explore and revisit parallel algorithms to speed up protein molecular dynamics

Dr. Giorgos Fagas Nanoelectronics Tyndall Institute Code parallelisation

Dr. Charles Patterson Materials TCD Code parallelisation

Dr. Ivan Rungger Ab Initio Simulation TCD Code development and optimisation

Dr. Gareth O’Brien Seismology UCD Portability, parallelization and optimisation

Dr. James McInerney Bioinformatics NUIM Scripting and debugging

Dr. Charles Spillane Bioinformatics UCC Package modification and configuration

Prof. Ann Burnell Bioinformatics NUIM Workflow configuration and data analysis

Dr. Michael Hartnett Marine Science NUIG Coupling of atmospheric and ocean models

Dr. Rodriguo Caballero Climate Modelling UCD Data management

Prof. John Sweeney Climate Modeling NUIM Porting of WRF on GPGPU

Dr. Noel Harrison Biomedical Engineering NUIG Optimisation and tuning.

Table 1: Projects supported as part of ICHEC’s Consortium support activities. (See http://www.ichec.ie/support/consortium for an up to date list of projects supported as part of the Consortium Support initiative.)

One of the key measures of the impact of the National HPC Service, and of the effectiveness of its support programme is through the monitoring of the refereed published output, both in terms of quantity, and quality (as assessed by the Impact Factor). A total of 66 refereed publications have been reported as published in 2009, up from 49 in 2008. See Appendix 2 for a list of these publications.

2.1.4 Visualisation Facilities at ICHEC

The ICHEC HPC Hub houses a 3D immersive visualisation environment allowing researchers to gain further insights into large and complex data sets, as well as prepare material for public outreach activities. In the spirit of the National Service, ICHEC provides the expertise and resources to generate the visualisation software to relieve researchers from having to learn complex new visualisation languages or environments.

This visualisation facility enables full active 3D stereoscopic visualisation of complex data and simulations. It allows researchers to view virtual objects or environments in 3D through the use of special glasses. But unlike a passive movie, the motion tracking system allows for user position awareness (e.g. where virtual objects move with the user) and for manipulation

of the environment (e.g. rotating a virtual object/landscape).The facility consists of a range of high-end visualisation hardware. Two powerful workstations - each equipped with two quad-core Xeon processors, 24GB of memory, a NVIDIA FX5800 graphics card with 4GB of graphics memory, fast 15,000 rpm SAS hard drives - run the visualisation software and ensure that large volumes of data can be processed in real-time. Visual output in 2D or 3D is then sent to the 100-inch transparent diffusion screen via a high performance projector. 3D stereoscopic visualisation is enabled by the use of active glasses, and the 6-camera motion tracking system provides awareness of user position and allows manipulation of virtual objects/environments.

Life Sciences / Bioinformatics

National Service

PRACE

DEISA

HECToR

Alin Elena

Gilles Civario

Christos Kartlakis

Materials / Condensed Matter

Environmental Sciences

Engineering

Astrophysics

Earth Sciences

Computer Science / Mathematics

PRACE

DEISA

HECToR

Gilles Civario

Christos Kartlakis

Ivan Girotto

Figure 7: The real life implementation of the support model presented in the joint proposal in 2008.

Prof. Chris Bean, UCD using ICHEC’s visualisation suite.


of roughly 200,000 times. That is to say when a real problem of scientific interest is spread across more and more of the system it is able to use the additional resources with a very high level of efficiency. This is achieved by carefully tailoring the code to exploit the network topology of the machine. This screenshot below shows a graphical representation of jobs using the 72 rack Blue Gene system. Here we can see, shown in pink, one of Dr. Downes’ jobs occupying the entire system. Perhaps the most significant result of these tests is that it is now possible to prepare a programme of simulations which will look at the underlying scientific problems in greater detail secure in the knowledge that the code’s ability to scale up will not be a limiting factor.

2.1.5.2 Distributed European Infrastructure for Supercomputing Applications (DEISA)Another major Supercomputing project is DEISA, the Distributed European Infrastructure for Supercomputing Applications . DEISA is a consortium of leading national Supercomputing Centres in Europe to advance computational sciences in the area of supercomputing. The consortium has deployed and operates a complex and heterogeneous HPC infrastructure at a continental scale with an aggregated peak performance of about two PetaFlop/s. More than 180 European research institutes and universities have already benefited from the DEISA Extreme Computing Initiative (DECI), involving 25 European countries and four more continents. The DEISA Consortium is continuing its services through EU FP7 support for the DEISA2 project with the goal to provide a turnkey operational solution for a persistent European HPC service.

We are glad to report that seven Irish research groups made an application to the 2010 DEISA call. This is very significant as this is the first time in the history of this programme that Irish groups have attempted to secure resources on this infrastructure. Though the programme itself is very competitive (the number of DECI applications has jumped

ICHEC provide computational facilities to run the finite volume fluid dynamics code which created the data for the study and it provided visualisation facilities to generate the visualisations of the data showing the formation of the stellar elephant trunks. ICHEC provide support and instruction where required to support this process, in particular the use of the Visit software package to generate the final results.

The end products for this project produced a set of visualisations in movie format, as well as images contained in Mackey and Lim (2010), a publication which credited ICHEC’s support.

Mackey J. and Lim A. J., Dynamical models for the formation of elephant trunks in HII regions, Monthly Notices of the Royal Astronomical Society, Volume 403, Issue 2, pp. 714-730, May 2010.

2.1.5 International Initiatives

2.1.5.1 Partnership for Advanced Computing in Europe (PRACE) Building a world-class pan-European High Performance Computing (HPC) Service is a highly ambitious undertaking that involves governments, funding agencies, centres capable of hosting and managing the supercomputers, and the scientific and industrial user communities with leading edge applications. In contrast to Research Infrastructures that focus on a single scientific instrument HPC Infrastructure has two unique characteristics: supercomputers serve all scientific disciplines and tier-0 supercomputers have a three year depreciation cycle as tier-0 implies leading edge services. This requires a periodic renewal of the systems and a continuous upgrade of the infrastructure. Furthermore, novel architectures and system designs will be created by the vendors for leadership systems. At any given time there will be between three and five different systems each of them best serving a particular application spectrum. This fact mandates a distributed Research Infrastructure, since no single site can host all the necessary systems because of floor space, power, and cooling demands. PRACE, the Partnership for Advanced Computing in Europe , has the goal of creating the prerequisites for just such a pan-European HPC service satisfying the objectives mentioned and aims to move into its implementation phase in summer 2010.

With ICHEC’s support, five Irish researchers have been successful in their applications for access to the PRACE prototype systems. The total amount of resources secured by these groups was approximately 96% of all allocated resources from the two rounds of prototype applications, and represents a nominal value of 473k. The largest of these systems, JUGENE, the IBM Blue Gene/P system at the Juelich Supercomuting Centre in Germany has an enormous 294,912 cores. On this system Dr. Downes (DIAS/DCU) who is working in collaboration with ICHEC Computational Scientists has been able to demonstrate so called “hard scaling” of a factor

In addition to making the visualisation facility available to users, ICHEC also provides technical support in preparation of the data to be visualised. Hence the entire process - from getting the data to visualising it - is facilitated by ICHEC. One simply has to make an appointment through the ICHEC Helpdesk, supply some data specification and visualisation requirements, then visit the hub to work with ICHEC staff before finally examining the results. The results can then also be readily used in traditional 2D mediums such as journal publications and slides.

Several groups have already availed of the ICHEC visualisation service. The following highlights just some of the work carried out.

TURBULENCE IN MOLECULAR CLOUDS (DR. TURLOUGH DOWNES) Dr. Downes’ work involves simulations of complex astrophysical phenomena in three dimensions. Effective visualisation of the resulting data is critical because the physical phenomena, more specifically the fluid density and surrounding magnetic fields, can be observed graphically thus allowing correlations to be established and scientific conclusions to be drawn. In the visualisation of the 3D data produced by these simulations, there was a difficulty in observing the correlation between the fluid density and magnetic fields. Using the Visit visualisation software package, ICHEC staff have assisted Dr. Downes in producing a more effective visualisation, in which the interaction of the fluid density and magnetic field was more clearly observed. Furthermore, the 3D visualisation system installed at ICHEC was used for immersive visualisation of the data, thereby allowing important physical effects to be easily identified.

NUMERICAL STUDIES OF THE EVOLUTION OF STRUCTURES IN H II REGIONS (DR. JONATHAN MACKEY)

The formation of elephant trunks, pillars and fingers or long columns of neutral gas pointing towards the central star in H II regions was the subject of this study. It used volume rendering to explore various linear and triangular clump configurations with different types of cooling over time periods of 50, 150, 250, and 350 kyr.

Dr. Turlough Downes

Fig. 9: Volume rendering of a 3D simulation of isolated pillar formation with three massive clumps initially in a triangle configuration.

Fig. 8: Volume rendering of a 3D simulation of the photoionization of a random clump distribution.

Figure 10: Dr. Downes’ run utilizing all 294,912 cores on the PRACE JUGENE Supercomputer, (the most powerful supercomputer in Europe in 2009). A further user (Prof. McGuire, UCD) has run his code on 262,144 cores again on the Juelich resource.


by 62% this year to a record 122 applications, and was ten times oversubscribed), we expect the strong partnerships between these groups and ICHEC to lead to acceptance rates well above average. Successful applicants will be given access to Europe’s most powerful supercomputers at one or more of the 13 DEISA partner sites which operate fifteen of the Top 100 supercomputers in the world.

2.1.5.3 Innovative and Novel Computational Impact on Theory Experiment (INCITE)Finally, ICHEC intends to provide assistance to Irish research groups to help promote their applications to the fastest supercomputer in the world at Oak Ridge National Lab, through the US DoE INCITE (Innovative and Novel Computational Impact on Theory and Experiment) programme.

Oak Ridge National Lab “Jaguar” - the world’s fastest supercomputer (Image courtesy of the National Center for Computational Sciences, Oak Ridge National

Laboratory.)

Dr. Simon Wong, Training Co-ordinator.

2.1.6 Education and Outreach

(2.1.6.1) ObjectivesDevelopment of the HPC user community in Ireland has always been a primary focus of ICHEC. From the very beginning, ICHEC has been committed to providing training courses for users and the wider academic community. Demand for our courses has consistently been high and flagged as a top priority among past, current and potential users according to various sources (e.g. surveys, meetings). ICHEC regularly delivers a “core” set of introductory courses that covers HPC, MPI and OpenMP, at various institutions around the country. In 2009, our main objectives with respect to training courses were as follows:

interested research groups or institutions all around the country.

developing more specialised, advanced courses in addition to our core set of courses. Coverage of additional topics is dependent upon user demand.

Although training courses represent a primary focus, the portfolio of ICHEC’s education and outreach activities has also expanded during the course of 2009, which led to further activities:

up or participation in various graduate programmes in collaboration with third-level institutions.

funding for undergraduate students to carry out summer projects on topics related to HPC or computational science.

consortium with a view to developing training resources and to avail of PRACE training opportunities for Irish researchers.

2.1.6.2 AccomplishmentsThe following represent the main accomplishments and key figures for the year 2009:

The core set of HPC courses was delivered to the following institutions:

the year spanned a total of 28 course-days attended by a total of 102 students. Since each course typically lasts 4 days, this amounts to 408 student-days.

to the core HPC courses:

Carpentry for Scientists”.

the first of which were awarded to two undergraduate students, Tim Hayes (TCD) and Colin MacSweeney (UL). Each student, supervised by ICHEC computational scientists, successfully completed a respective 10-week research project, producing a final report. The project outcomes are published on the ICHEC website (http://www.ichec.ie/research/summer_scholarships).

10 online lectures (via video conferencing technology) for the Nanoscale Simulators of Ireland (NSI) graduate programme. The lectures make up a module in the programme entitled “Programming Concepts”, the topics of which range from basic computer architecture to software engineering and sequential/parallel programming.

Date Venue

Feb Armagh Observatory

Mar University College Dublin

May Dublin Institute for Advanced Studies

Jun University College Cork

Oct Tyndall National Institute

Oct-Nov University College Dublin

Nov NUI Galway

2.1.6.3 Plans for 2010In 2010, ICHEC will maintain its education and outreach activities as well as engaging in new initiatives. For existing programmes, the focus will be on enhancing their quality. New activities will have an emphasis on ensuring that maximum benefits are realised at minimal cost. The following outlines ICHEC’s plans for 2010 in the area:

TRAINING COURSES

to institutions around the country on an on-demand basis. We expect that the level of demand will remain constant or grow slightly in 2010.

courses such as the 2-day Advanced MPI course. The target is that each new course will be delivered at least once in 2010.

on feedback from users, via Users Council meetings, etc. As course development involves significant effort from ICHEC staff external material will be adapted where possible.

WORKSHOPS

to arrange topical workshops for the community. At least two such workshops are to be organised in 2010.

GRADUATE PROGRAMMES

programme for the third consecutive year. This will likely involve the delivery of 10 online lectures preceded by a review of the material.

setting up or participate in graduate programmes together with other third level institutions.

STUDENT MENTORING

awarded. We aim to host three students in 2010.

Attendees at the UCD workshop in October Ivan Girotto helping attendees at the UCC workshop in June


2.1.7 e-INIS

Significant advances in developing the shared national ICT research infrastructure were made by the e-INIS partners during 2009. Following on from the considerable investment in High Performance Computing during 2008 which saw the procurement of the Stokes and Stoney compute clusters, 2009 saw the focus shift to support users on these platforms while developing the remaining components of the infrastructure such as data storage and advanced networking.

Great progress was made in the deployment of the federated data store, which aims to provide increased capacity for data-centric and data-intensive research. The data store is also intended to facilitate the sharing data among research collaborations and indeed to disseminate information to the general public through appropriate portals.

Almost one Petabyte (1048576 Gigabytes) of storage has so far been made available to the Irish research community. These storage resources have been installed at DIAS, UCC, and TCD and a number of high profile collaborations are now making use of the data management resources. These include researchers from areas including climate modelling and atmospheric chemistry, neonatal brain research, bioinformatics, solar physics, particle physics and digital humanities.

Underpinning all aspects of the e-Infrastructure is the network infrastructure put in place by HEAnet along with the computer services departments of the institutions hosting the shared resources. 2009 saw the deployment of the 10Gbps Lambda-Switched backbone and the connection of a number of resources. This network will provide closer integration of the distributed equipment making up the e-Infrastructure and, through the dedicated HEAnet router, provide improved connectivity with European resources and researchers across the GÉANT infrastructure.

Of course the development of an integrated e-Infrastructure is not limited to the commissioning and maintaining the physical resources. Considerable effort continues to be invested in user training and support. The advances in the federated identity management activity, Edugate, are expected to simplify the authentication and authorisation of users from Irish HEIs who wish to gain access to these shared resources.

Further exciting advances are planned for 2010 with additional HPC and storage resources being put in place along with further connections to the 10Gbps optical network. It is hoped that new user communities will be encouraged to make use of the infrastructure as a whole and the data management resources in particular.

2.1.8 Met Éireann Collaboration

ICHEC provides compute resources for Met Éireann’s operational forecast 4 times each day. In only 2 instances out of more than 1,500 runs on Stokes over the last 18 months has the forecast had to be transferred to our back-up resources. In essence, therefore, the quality and reliability of our hardware and the management of it by ICHEC’s very experienced systems administrators is such that we can with justification claim to provide for all of our users a mission critical resource. In addition to this, the direct involvement of ICHEC’s staff with a Met Éireann computational scientist has allowed Met Éireann to have a much more significant role in the development of EU-led new-generation forecasting software than would have been possible otherwise. Ireland will soon benefit from these significant developments in forecasts when significantly higher spatial resolutions become part of the daily forecast.

The Met Éireann collaboration has provided a suitable framework for ICHEC to bring its expertise in HPC to the climate modelling and weather forecasting communities. ICHEC has taken an active part in the development of faster and more accurate simulation codes that are of interest to Met Éireann and other national weather forecast agencies as part of the Harmonie initiative. Alastair McKinstry, ICHEC Software Architect, has been testing HARMONIE since April 2007, liaising with members of Hirlam project in the development of the model. HARMONIE runs on a much finer grid compared with the current model and will provide a more realistic description of local weather. It will also enable Met Éireann to improve its forecasting capabilities in areas such as the prediction of extreme precipitation events and flooding, road ice and the spread of weather-sensitive diseases such as the foot-and-mouth virus.

From ICHEC’s perspective, the Met Éireann collaboration constitutes a clear endorsement of our ability to deliver a mission-critical service. ICHEC has renewed its collaboration with Met Éireann for 2009 and expects to remain collaboratively engaged with Met Éireann well into the future. Other benefits to Met Éireann include the opportunity to grow their forecasting services by securing additional resources “on demand”, and also the ability to evaluate the performance of their codes on new HPC systems hosted by ICHEC.

Alastair McKinstry

EC-EARTH, a project that is part of a Europe-wide initiative to improve climate modelling, will be used to simulate Ireland’s future climate, looking ahead 10-20 years with high resolution simulations as well as simulations to the end of this century (for the 5th IPCC assessment report). EC-Earth will seek to develop a global Earth System model consisting of a state-of-the-art atmospheric general circulation model, a state-of-the-art ocean general circulation model, a sea-ice model, a land model, and an atmospheric chemistry model. ICHEC helped to roll out the project, building on the work done in the area of weather forecasting but incorporating advanced new components to provide a more realistic description of the climate than current models allow. This partnership has enabled Ireland to play a much more prominent role in the EC-Earth initiative, by pooling combined expertise.

Further information about EC-Earth can be found at http://ecearth.knmi.nl/index.php?n=PmWiki.AboutEC-EARTH, including a discussion of the specific goals of the project, the ECMWF software stack relies on for modelling.

“The Government is committed to sustaining and developing a climate-modelling framework within Met Éireann, building on the C4I project, with links to national and international research in this area, to ensure that Ireland has an advanced capability for prediction of future climate conditions.”

– National Climate Change Strategy 2007-2012.

to support and benefit from these projects (including EC-EARTH) in the future”.

ICHEC also held a number of exploratory talks within the semi-state sector to explore the possibilities of collaboration with other organisations such as Teagasc.

Screen shot of the portal for accessing IPCC data. ICHEC is an ESG data node. Part of the European Distributed climate data archive. EC-Earth data for the UN IPCC report will be managed by ICHEC and accessed via the Earth System Grid.

EC-Earth spin-up runs: Met Eireann and ICHEC ran “spinup” runs, testing the model against pre-industrial climate in 2009. This is the output of the climate in the 17th century.


GPGPU OPTIMISATION OF THE WEATHER RESEARCH AND FORECASTING (WRF) MODELAs part of a collaboration with the Irish Climate Analysis and Research Units (ICARUS) at NUI Maynooth, Nicola McDonnell (on secondment from EPCC) is involved in the implementation of the physics kernel of WRF on the Stoney system where a number of NVIDIA Tesla cards have been installed. Previous work by Michalakes et al. at NCAR (National Center for Atmospheric Research) has utilised NVIDIA systems to achieve a 100-fold speed-up in the physics kernel of WRF, or a 30% speed-up of the overall model.

QUANTUM ESPRESSO GPU PORTINGICHEC is collaborating with the CNR DEMOCRITOS Group (Trieste, Italy) to study the feasibility of porting Quantum Espresso to GPGPU architectures. Quantum Espresso is a very flexible code with high scalability making it ideal for users who wish to access the EU’s new PRACE petaflop systems for Grand Challenges in materials science. Two parts of this integrated suite of codes for electronic-structure calculations and materials modelling at the nanoscale are being undertaken in parallel by two different groups. ICHEC computational scientist Ivan Girotto is carrying out the PWscf part which performs electronic and ionic structure calculations while a group from Washington University is doing the section related the basic Car-Parrinello simulations. Two development approaches are being evaluated in parallel: HMPP and CUDA.

2.1.9 Research Activities

ICHEC both facilitates and participates in many research activities. The active involvement of ICHEC staff in scientific research is particularly important in bridging the gap between increasingly-powerful HPC facilities and researchers from different communities. We interact with many groups varying in size from individuals to large international collaborations. Information on some of the research projects and collaborations ICHEC has been associated with are provided below.

RESEARCH IN CLIMATE MODELLINGHarmonieHarmonie is the next-generation weather model for Met Éireann, in development as part of the Met Éireann collaboration. Operational testing of the CY33 version has been ongoing, showing some meteorological issues with non-physical storms. To fix this, ALARO physics from Météo France and a parameterisation for the “gray area” between 1km and 10km resolution sizes was implemented in HARMONIE in the CY35 cycle. Much work was done for the CY35h1.3 version by ICHEC; this is now being tested for operational use.

WRFA collaboration was set up in 2009 with NUI Maynooth to take advantage of the GPGPU processors in Stoney, acquired as part of the NUI Galway e-INIS procurement. This involves re-writing part of the WRF weather model used at NUIM for GPGPU processors. As a first step, the work of Michalakes has been ported to the Tesla S1070 GPGPUs in Stoney, to gain experience in WRF and CUDA programming. Work on implementing the radiative transfer physics of the WRF model will take place in 2010.

OASISThe OASIS coupler , used by many climate models to connect atmosphere, ocean and other components, was seen as an issue for several parties such as the EC-EARTH consortium and the users of the COSMOS model in the NIUG marine modeling group. In particular there were issues with mathematical interpolation errors and a lack of scalability of OASIS3. While OASIS4 is parallelised, it is still in ”beta” testing and has not been accepted yet by any of the climate groups to date. To this end a joint project was set up and a postdoctoral fellow (Dr. Adam Ralph) hired with SGI CAS funding to test OASIS4 and work on interpolation issues.

CMIP5The IPCC AR5 runs of EC-Earth are expected to generate approximately 130-200 TB of data. A project was set up to store this data on the e-INIS storage system. This involved setting up Grid access on Stokes, and credentials to enable metadata servers to access the e-INIS storage. This is now functioning, and work is presently underway to add metadata servers for this data.

Other WorkWork was carried out for the NUI Galway marine modelling group to implement OpenMP parallelised versions of the REMOTE20 and ECOMSED models. This has enabled a 4-6x speedup over existing work.

RESEARCH IN GPGPU COMPUTINGGPGPU (General Purpose computing on Graphics Processing Units) has become increasingly popular in the HPC community in recent years, in addition to their traditional role in visualisation work. While it is not the only path towards widespread availability of peta-scaling computing, it is currently the most promising one. With this in mind ICHEC has pursued the following goals in 2009:

porting some codes that have a major interest for the Irish scientific community to GPGPUs.building a strong expertise in the field, permitting us to effectively advise and support users who want to evaluate GPU computing themselves.building partnerships with manufacturers, developers and users communities to allow us access to new hardware, codes and tools.

At this particular time, GPGPU technology is not ready for general purpose production computation. However, the next version of NVIDIA’s architecture named FERMI looks extremely promising. This could lead to major changes in the HPC landscape by providing some features that true HPC hardware must provide: full IEEE 754 norm support and ECC memory correction for example. It is therefore very important to be able to utilise this technology when it becomes available. In this regard, the development and expertise being gained on today’s hardware will be beneficial for tomorrow’s. One may even expect some spectacular performance gains compared to today’s chips, due to the very promising features FERMI will provide.

The following paragraphs provide an overview of our GPGPU projects:

A JAVA LIBRARY FOR THE GENERATION AND SCHEDULING OF PTX ASSEMBLY ICHEC computational scientists Christos Kartsaklis and Gilles Civario, have developed a tool called JASM, introduced at the prestigious NVIDIA GPU Technology Conference 2009 where Christos presented a talk which discussed ongoing progress regarding the development of a Java-based library for rapid kernel prototyping in NVIDIA PTX and PTX instruction scheduling. It is aimed at developers seeking total control of emitted PTX, highly parametric emission of, and tunable instruction reordering. It is primarily used for code development at ICHEC but is also hoped that the NVIDIA GPU community will also find it beneficial.

DL_POLY CUDA PORTINGDL_POLY is a well known molecular dynamics simulation code developed by STFC Daresbury Laboratory in the UK. As part of a collaborative research programme with Daresbury Labs, ICHEC has undertaken to port DL_POLY version 3.10 to CUDA. The port has been done by Christos Kartsaklis in close collaboration with Dr. Ilian Todorov and Prof. Bill Smith form Daresbury. The code has now been successfully ported and run efficiently with a speed-up of a factor around 4 so far. Furthermore, the code is parallelised with a mix of MPI, OpenMP and CUDA, allowing an efficient usage of both the CPUs and the GPUs in this new hybrid architecture. The ported code should be integrated into the official distribution by Daresbury. We expect it to be integrated into the forthcoming version 4.

2.1.10 Other DevelopmentsOther significant developments in 2009 include:

Ireland (2009 Edition).

Performance Computing which provides a shared space for researchers to work in close contact with ICHEC staff: a shared space to exchange knowledge and foster interaction.

outreach activities. The first such event involved our participation in the IRCSET-sponsored event “Innovation fuelling the smart economy”, organised at the RDS Dublin in September 2009. This event, which saw ICHEC provide a closed environment for the display of 3D interactive scientific demos, proved to be a great success and motivated us to ramp up this type of activity.

Gilles Civario

Nicola McDonnell

Ivan Girotto

Christos Kartlakis

3D interactive demos at the RDS Dublin:


2.2 Looking ForwardWhile 2009 was a successful year in consolidating ICHEC’s role as the national HPC provider, the year 2010 will see ICHEC, pursuing a forward-looking strategy, developing from a “resource centre” for high-end computing services into an internationally recognised centre of excellence following a number of key high-level strategic goals:

capacity and services.

the academic research community to encompass the broader industry sector and building strategic alliances with HPC providers

consortia and programmes.

In line with this strategy, ICHEC will develop and broaden its key strategic relationship with HPC vendor SGI. From the establishment of the Centralised Application Support group (CAS) in 2008, which provided an investment vehicle to fund joint research on Climate Modelling and GPGPU consultancy work, 2010 will see ICHEC positioned as an exclusive partner for SGI in Europe for consultancy on GPGPU-related projects, including code design and porting, optimisation, benchmarking etc. This partnership will clearly position ICHEC as a significant presence in the field of GPGPU programming. In this regard ICHEC will continue to develop expertise in GPGPU computing and expects to establish close links with the high profile US company NVidia, the leading manufacturer of GPGPUs.

We also look forward to rolling out our technology transfer programme with new staff appointments and a strategy to engage with a number of key agencies and bodies such as Enterprise Ireland and IBEC to identify and engage with Irish companies where our expertise and compute resources can be beneficial.

The year 2010 will see a step change in terms of Irish involvement in international initiatives as a result of ICHEC’s pro-active programme to procure access to large-scale world-class capability computing infrastructures for Irish research groups. Following many years planning, the EU initiative PRACE (Partnership in Advanced Computing in Europe) to create a sustainable HPC ecosystem for researchers will finally be launched in mid-2010, making petaflop computing available for Grand Challenge projects. We expect to receive funding for 2.5 staff as part of our involvement in this programme. Further opportunities are expected over the next two years, principally in the field of GPGPU programming and community code development (e.g., through the CECAM/ACAM link).

We are also developing plans to ensure adequate resources are available to our users despite the current economic downturn and tightening of public funding that the capital funding required for the replacement of the Stokes

national cluster may not be made available on time. We expect to upgrade Stokes in late-2010 to provide enhanced performance and additional resources for our users. We are also planning for the deployment of a shared national cluster called a “condo cluster” where the institutions buy “shares” in the National Condo Cluster, in return for which they are issued a commensurate level of resources “at cost”. The institutions become liable only for running costs pro-rata to their share of ownership. As ICHEC takes responsibility for the procurement, installation, and operation of the HPC infrastructure, it means that benefits to stakeholders are numerous. The additional benefit is to provide hesitant users with a friendly entry point to the National Cluster through the institutions’ IT services. This has already been shown to increase the subsequent uptake of the National Service among grass root researchers. This shared services model which we will actively seek to roll out across all of the institutions will lead to significant savings. It is a best-in-breed model of a centralised, co-ordinated shared services agenda for HPC computer resources. We believe our National Condominium Model will deliver unprecedented levels of effectiveness of public funding for HPC resources.

Another important strand for ICHEC is that of e-Infrastructures, and more specifically data management. We foresee that our involvement with the Humanities and Social Sciences, with the bioinformatics community and the climate modelling community will all continue in 2010, and in the case of the latter two at least until late 2011. In bioinformatics, ICHEC’s “BioPortal” will be extended to support a wider range of bioinformatics applications, and additional services such as basic data management and visualisation tools. In climate modelling, our work will continue to integrate the CMIP5 and C4I data in the e-INIS National Storage Infrastructure, and develop suitable metadata to make it discoverable by climate sciences tools such as ESG. The addition of metadata for GeoNetwork will enable the general public to access information contained in these climate data via popular interfaces such as Google Earth. All three activities will provide significant international exposure to Irish data (Humanities), Irish application software (BioPortal) and Irish Data Management expertise (climate).

If suitable funding is forthcoming, ICHEC would be enthusiastic in leveraging on its combined expertise in software engineering, portal development, data management and data mining to develop solutions for other scientific disciplines, e.g., such as for high-throughput sequencing.

Budget 01/01/09 – 31/12/09

Labour 1,326,000

Equipment 49,000

Hosting Costs 355,000

Other 337,000

Total 2,067,000

Actual 01/01/09 – 31/12/09

Labour 1,121,000

Equipment 80,000

Hosting Costs 179,000

Other 443,000

Total 1,823,000

3 Financial Summary


4 ICHEC Support for ProjectsStarting from a ‘green-field site’ in late 2005, ICHEC has in a relatively short space of time developed into a very competent national facility, delivering capability and capacity computing to researchers in the universities and other third level institutions as well as engaging with and supporting SMEs to help build competitive advantage, profitable sales, exports and sustainable employment. The number of researchers using the service has increased steadily, rising from ~140 after one year of operation to 300+ users at the end of 2009. Capability computing power is now available locally through IBM Blue Gene /L and /P as well as from a number of other supercomputer centres in the EU through partnership agreements and access to EU projects such as DEISA and PRACE. ICHEC staff mentor and train Irish researchers in developing and tuning parallel code. They advise and support PIs in formulating applications for support to various funding agencies; these include initiatives in areas as diverse as Language Processing (SFI, lead DCU), Climate Modelling (PRTLI5, lead NUIM), Systems Biology (SFI, lead UCD). The availability of ICHEC’s clusters is now virtually 100%.

As a national facility, ICHEC has to cater for the needs of a diverse community of users and requirements. The application procedure for access to the national compute clusters takes this diversity into account by providing three possible routes to apply for resources, ensuring the best compromise between a fast response time, fairness of the application process, and an efficient usage of the resources while complying with the strategic objectives of the national centre (which is primarily to provide the capability to address “Grand Challenge” applications).

Class A applications (“Grand Challenge”)Class A applications are typically submitted by consortia concerned with “Grand Challenge” problems. These groups require resources representing a substantial fraction of the centre’s resources over a long period. This type of application is expected to yield high-impact scientific publications.

Class B applications (“Regular”)Class B applications meets the needs of the bulk of our user community, typically consisting of small research groups or individual researchers.

Class C applications (“Discovery”)Class C applications support users will little or no prior experience of HPC, as well as more experienced users who would wish to gain a better understanding of their requirements before committing the resources to prepare an application for a Class A or B project. Their typical use consists in small scale runs for the former, and code porting, optimising, and benchmarking for the latter.

The metrics used to categorise applications is based on the concept of notional value, which include the aggregate cost for compute and storage over the duration of the project.

Supporting users, whether they are new to HPC or seasoned customers of the service, is at the core of what ICHEC is about. We take this part of our mission very seriously and are constantly trying to improve what we do in this area. We have engaged with this challenge through different approaches:

comprehensive range of training material and courses in various geographical centres

groups to discuss and assess their objectives and requirements

staff engage with the community in a collaborative way, assisting key groups with the development, porting and optimisation of their parallel applications and become members of the team

been logged and dealt with by ICHEC staff.

The feedback we get from our user community, whether formally through the Users’ Council or informally on a day-to-day basis, indicates to us that our users are appreciative and satisfied with our service.

Our commitment to provide a “user friendly” service to the Community, and particularly considering that a large fraction of our user base is still relatively inexperienced in HPC, ICHEC has a policy of supporting software applications which are of interest to small groups of users. Understandably, focus is placed on programming environments, and on software applications for emerging communities particularly within the Life Sciences. Our policy of having our computational scientists “specialised” in certain research areas allows us to provide in-depth support, not only in terms of providing a centralised installation, but also in helping with the use of these applications on our infrastructure.

Details of our software environment can be found on our Web site at http://www.ichec.ie/infrastructure/software/

ICHEC has now been built into a very competent national facility, delivering capability

and capacity computing to all of the universities and other third level institutions.


4.1 Class A Projects

Dr. Turlough Downes, Dublin City University

Turbulence in Molecular Clouds: Beyond Magnetic Flux Freezing

Dr. Niall English University College Dublin

Effects of Electric and Electromagnetic Fields on Nanoparticle-Protein Systems

Dr. Marcus Greferath, University College Dublin

Performance Analysis for Low Density Parity Check Codes

Dr. Michael Nolan, Tyndall National Institute

Engineering Metal Oxide Interfaces

Dr. Stefano Sanvito, Trinity College Dublin

Transport Properties of Organic Macromolecules: Modelling at the Boundaries between Biology and Nano-Electronics

Dr. Damien Thompson, Tyndall National Institute

Faster, Bigger, Better: Supra-Nanoscale Molecular Dynamics for Nanoelectronics

Dr. Jiri Vala, National University of Ireland, Maynooth

Topological Phases in Quantum Lattice Models


Turbulence in Molecular Clouds: Beyond Magnetic Flux Freezing

Dr. Turlough Downes Dublin City University and Dublin Institute for Advanced Studies.

PROJECT SUMMARY We know that stars like our own sun form from clouds of gas and dust as they collapse under their own gravity. These clouds, called molecular clouds, are cold (with temperatures of only around 10-100K) and relatively dense. They are also threaded by magnetic fields which are often assumed to be “frozen in” to the cloud: if the gas of the cloud moves then the magnetic field must move with it and vice versa. The magnetic fields are important in the dynamics of the molecular cloud - they resist being compressed and the magnetic field lines tend to straighten out if at all possible.

Observations of molecular clouds show strong indications of the presence of supersonic turbulence. It seems intuitive to say that turbulence, or indeed any motion of the gas making up the molecular cloud, could disrupt the process

Figure 2: Number of computational zones updated per second as a function of the numbers of cores. This scaling data is for the JUGENE system (solid line) and shows extremely good scaling up to the full 294,912 cores of this system. Also shown is ideal scaling for comparison (dashed line).

of star formation. If the turbulence is strong then it will continually “stir up” the molecular cloud, preventing it from forming stars and actually prolonging the life of the cloud itself. In short, turbulence is thought to play a crucial role both in star formation and in the overall dynamics of molecular clouds themselves.

It seems that to understand how stars form we need to understand more about this turbulence. Two obvious questions are where does this turbulence come from, and how quickly does it die away?

In order to address these issues we need to look again at the assumptions which are normally made. The frozen-in assumption for how magnetic fields and the gas of the molecular clouds interact is known to be a risky one. We know that the field does not perfectly follow the gas in molecular clouds but that the two move almost, but not quite, in concert. At a very fundamental level there must be at least a slight drift between how the field moves and how the gas moves as, without it, stars could never actually form: the pressure associated with the magnetic field would stop the gravitational collapse before nuclear fusion could start at the centre of the forming star.

Figure 1: Snapshot of a multi fluid magnetohydrodynamic turbulence simulation. The simulated volume is a cube and three slices through the cube showing iso-density contours and magnetic field directions are displayed.

Until now, it has been impractical to incorporate the real physics (so-called “non-ideal” effects) associated with the magnetic field dynamics, and its effects on the overall cloud dynamics, into simulations. However, using a new world leading code incorporating novel algorithms developed in-house by Dr. Turlough Downes and Dr. Stephen O’Sullivan it is now possible to study the behaviour of molecular clouds using simulations which incorporate the effects of this extra physics. This state-of-the-art code is being used in this project to study the behaviour of turbulence in molecular clouds. The complexity of the physics involved means that the computational requirement for these simulations is enormous: a minimal resolution required for turbulence simulations is 5123 points (512 points in each of the 3 spatial dimensions) for which around 4 - 6 days is required for a standard simulation on 2048 cores of the BG/P machine. A snapshot of one of these simulations is shown in Figure 1 where contours of equal density are shown along with magnetic fields lines.

The changes in the behaviour of the magnetic field resulting from the non-ideal effects lead to significant changes in the

distribution of material in turbulent clouds. They cause the turbulence to die away more quickly and significantly reduce the amount of structure created on small scales. This has implications for the origin of the observed turbulence: if it dies away more quickly then more energy is required to sustain it. It also has implications for star formation which is believed to be strongly influenced by the presence of structures on small scales.

During 2009 Dr. Downes, working in association with Dr. Michael Browne of ICHEC, was awarded 4.3 million core hours on PRACE systems (see http://www.prace-project.eu/ ) under their prototype testing programme. This time was used to test the performance of HYDRA on prototype petascale systems. The results of this work were extremely successful with HYDRA displaying strong scaling from 8192 cores up to 294,912 cores on the peta op JUGENE system based in Julich, Germany (see Figure 2). This result allows Dr. Downes’ research team to prepare applications for access to very large scale PRACE and international compute resources secure in the knowledge that HYDRA is a proven code on one of the most advanced systems in existence.

2e+08

4e+08

6e+08

8e+08

1e+09

1.2e+09

1.4e+09

1.6e+09

1.8e+09

2e+09

50000 100000 150000 200000 250000

Zone

upd

ates

/sec

ond

Number of cores

1024^3 sim with HYDRAIdeal scaling


E!ects of Electric and Electromagnetic Fields on Nanoparticle-Protein Systems

Dr. Niall EnglishUniversity College Dublin

The dominant focus of this research project is concerned with the non-equilibrium molecular dynamics (MD) simulation of both thermal and non-thermal effects of static electric and electromagnetic (e/m) fields on nanoscale systems of interest to biology and nanomedicine. This serves to extend our understanding of important potential applications for nanoparticle-protein interactions and binding. In particular, molecular simulation will offer a unique insight at the atomistic level of how static electric and e/m radiation affect nanobiological structures and systems.

The massively large molecular simulation of large systems is required in this project, which has also led to PRACE and CINES activities in 2009/10 on a variety of platforms.

Water-self diffusion through single-walled carbon nanotubes (SWCNTs) inserted normal to a phospholipid membrane has been studied using equilibrium and non-equilibrium molecular dynamics simulations in the presence of static and electromagnetic (e/m) fields1, 2. Four different SWCNTs were investigated: (5,5), (6,6), (8,8) and (11,11) and also three arrays of four (6,6) SWCNTs separated by 15, 20 and 25 Å respectively. The (5,5) system shows interesting behavior, where an increase in the applied field frequency in the z direction decreases the water permeation rates, reaching values at higher frequencies similar to zero field conditions. The (6,6) arrays simulations demonstrated that there is a friction effect, when the nanotubes are closely packed, which retards the movement of the individual water files.

The steady-state unidirectional permeation rates for water molecules crossing through the SWCNT (in molecules per ns), in the absence and presence of static and e/m fields, are provided in Fig. 1. These results show no appreciable trend when the e/m field frequencies are incremented for the (6,6), (8,8) and (11,11) systems, in both y and z directions, and the (5,5) when the field is applied perpendicular to the SWCNT (y direction).

When the (5,5) systems are subjected to fields applied in the z direction (see Fig. 1b), we observed a monotonic decrease of water permeation rates with the field’s frequencies. Furthermore, these rates reached values similar to the zero-field conditions at higher frequencies. Statistically, with the remarkable exception of the (5,5) system under the influence of e/m fields applied along the principal axis of the tube (cf. Fig. 1b), we observe that there is no correlation between the field frequencies and the computed water permeation rates. Because of these intriguing effects, we decided to investigate further the water dynamics inside (5,5) SWCNTs under axially-applied e/m fields.

It was shown that static EFs applied along the axis of (5,5) SWCNT enhanced water self-diffusion across them due to a decrease in the fluctuations of the number of water molecules inside the SWCNT, which was a direct consequence of the water molecules’ dipole orientation with the field1. This was found to take place because of a reduction of clusters of misaligned dipoles that occurred with high frequency in the zero field case. In Fig. 2, we present results for these features under the effects of axially-applied e/m fields. Interestingly, both the dipole orientations and pore loadings correlate very well to increments in the field frequency.

In the case of water dipole orientations inside the pore (see Fig. 1a) the influence of the external field on the probability distributions of the cosine of the angle of the dipole with the positive z axis, , is illustrated. Although the distributions are shifted almost entirely to the left side of the diagrams in the static field case and for lower frequencies, at higher frequencies the distribution becomes more symmetric with peaks at -1 and 1, in very good agreement to the zero-field case.

As mentioned previously, the pore loading depends strongly on the water molecules’ orientation within it; in Fig. 2b, we have plotted the distribution of the intrapore loadings for different field frequencies. At lower frequencies, we observe that the fluctuations in particle numbers within the pore are similar to the static field case; however, at higher frequencies, the distributions are equivalent to the zero field case.

From the features considered in previous sections it is clearly evident that the results of low-frequency e/m field simulations are more similar to the static EF case, whereas high-frequency fields resemble zero field conditions. These results support the finding that the effects of low-frequency e/m fields should be closer to those of static EFs (if any), because the slower time variation of the field (longer period) relative to permeation and loading timescales allows the field to influence the dynamics of these events. On the other hand, high-frequency e/m fields do not readily permit enhancements in permeation or alterations in dipole alignment vis-à-vis the zero-field case, due to their faster oscillation and shorter period relative to permeation timescales in (5,5) NTs.

REFERENCES1A. Garate, N.J. English and J.M.D. MacElroy, Molec. Sim., 35, 3 (2009) 2J.-A. Garate, N.J. English and J.M.D. MacElroy, J. Chem. Phys., 131,

114508 (2009)

Figure 1: Steady-state permeation rates for all of the systems studied under the influence of static and time-varying electrical fields. (a) (5,5) under y-field conditions, (b) (5,5) under z-field conditions, (c) (6,6) under y-field conditions, (d) (6,6) under z-field conditions, (e) (8,8) under y-field conditions, (f ) (8,8) under z-field conditions, (g) (11,11) under y-field conditions, (h) (11,11) under z-field conditions. The dashed gray lines represent the zero-field conditions and the zero frequency points are the permeation rates in the presence of the static fields. The black error bars represent the standard errors for the zero field conditions. Note that in (b) there is no overlapping of error bars at lower frequencies.

Figure 2: (a) Normalized probability distributions of the cosine of the angle which the dipole moment of water molecules makes with the positive z-axis, for the (5,5) systems. (b) Normalized histograms of the number of water molecules within the (5,5) SWCNTs.


Performance Analysis for Low Density Parity Check Codes

Dr. Marcus Greferath University College Dublin

BACKGROUNDCoding Theory is the science of finding efficient schemes by which information can be prepared for (more) reliable transmission through a noisy channel. Its basic idea is to add redundant data with each transmission so that, even if errors occur, sufficient protection exists to recover the original message.

During the recent decade a class of codes that exhibits performances close to a theoretical limit (Shannon limit) for noisy channels was rediscovered . These are what are called Low Density Parity Check (LDPC) codes. Originally introduced by Gallager 1 in 1963, LDPC codes can yield high performances on the additive white Gaussian noise (AWGN) channel, and have been shown to outperform previous code constructions in many applications.

The algorithm used for removing the errors from the distorted word received at the end of the communication channel, is known under the term message passing algorithm. It uses a graphical representation of the code, the Tanner Graph, and it is highly efficient. Obviously it is desirable to have the encoding process of such codes most efficient as well. To achieve this goal one searches for systematic constructions of LDPC codes. Many good constructions are known nowadays, but only a few of them (if at all) outperform random constructions of LDPC codes. For this reason there is further demand for the systematic construction of LDPC codes with excellent performance.

WATERFALL DIAGRAMS AND ERROR FLOORSThe performance of the LDPC codes is generally illustrated by means of what are called called waterfall diagrams.

In such a diagram (see Figure 1) three curves are set in relation to each other. One is a vertical line that is usually referred to as the Shannon limit. It marks the point on the x - axis right of which the benefit of coding should be expected. A further curve of very moderate slope marks the behaviour of a communication system in which no coding is performed at all. The signal-to-noise ratio (SNR) of the channel is essentially mapped in a one-to-one manner to the bit-error-ratio (BER) in the received word. This curve might be considered as a gauging curve. The third curve, the actual waterfall, is measuring the performance of the given code on an AWGN channel. The steeper this curve, and the closer it is to the vertical Shannon limit, the better.

Many LDPC codes exhibit what is called an error floor. This is a region in which the plotted waterfall curve flattens out. Its slope approaches that of a horizontal line, and hence, an improvement of the channel quality does not yield any further improvement in bit error ratio that results from using the code. Regarding the applications it is considered to be ineffective to use a given code with a too high error floor. Developers therefore strive for the construction of codes exhibiting error floors only at a very low bit error ratio.

LDPC CODES FROM FINITE TRIANGLE-FREE GEOMETRIESSystematic constructions of LDPC codes have been given in various ways. A few years ago, it turned out that such codes can be obtained from projective and affine spaces over finite fields (cf. 2).

Later, finite generalized polygons were used to construct good LDPC codes (cf. 3). As a rule of thumb it can be stated, that geometric structures containing only a few or no triangles have a particular potential to yield LDPC codes of good performance.

In the project at hand we came up with a class of LDPC codes derived from incidence structures that are known as inversive spaces (cf. Figure 2). These spaces consist of points and circles, and a maximal set of circles mutually tangent in one and the same point is referred to as a pencil. It was our conjecture based on strong evidence, that under quite general assumptions, a derived incidence structure consisting of these pencils as points, and the original circles as lines is a triangle-free partial linear space, also known as a (0; 1) -geometry.

A THEORETICAL RESULTOn the theoretical side, we have now been able to obtain a mathematical proof for our abovementioned evidence. This proof was pointed out to us by Aart Blokhuis 4 at the recent conference ALCOMA 2010 in Thurnau, Germany. We will not present the proof here, but a theorem that garantees the properties of the (0; 1) -geometry. This result will be published soon (cf. 5).

Theorem: Let M be a miquelian inversive space of dimension r and order q . If q is even, or q and r are odd, then the partial linear space induced by the pencil-circle incidence structure in M does not contain triangles.

PERFORMANCE SIMULATIONSGiven an LDPC code there is generally no way to determine its performance except by doing an extensive simulation of its behavior in a communication channel. The simulation

procedure works as follows: a message vector is randomly generated, the message is encoded using the generator matrix to produce a codeword, which is modulated to create the desired signal strength. Noise is generated randomly and added to the codeword to simulate the channel. The received corrupted codeword is then decoded using a decoding algorithm. This procedure is repeated a (large) number of times, and the results, namely the BER as a function of the SNR is then recorded in a waterfall diagram.

We did this based on vast support from ICHEC who were also very helpful in adapting our software packages to the requirements of the Blue Gene cluster and particularly coming up with a pure C version of the programmes. Figure 3 shows the waterfall diagram produced by an LDPC code of quite large length and very high rate. It can readily be seen that this code performs closely to the Shannon limit and furthermore does not show any error floor above the BER 10-9 .

To prove the appropriateness of our codes for the applications we need to evaluate their performance in further simulations down to a biterror probability of less than 10 13 . The goalof the project is hence, to run a very high number of parallel simulation tasks for times ranging from several weeks to several months.

References1 R.G.Gallager, “Low density parity check codes,” 1963. PhD Thesis, MIT press, Cambridge ,MA, 1963.2 Y. Kou, S. Lin, and M. P. C. Fossorier, “Low density parity check codes based on finite geometries: A rediscovery and new results.” Preprint, 2001.3 P. O. Vontobel and R. M. Tanner, “Construction of codes based on finite generalized quadrangles for iterative decoding,” in Proc. IEEE Intern. Symp. on Inform. Theory, Washington, D.C., USA, p. 223, 2001.4 A. Blokhuis, private communication at ALCOMA-2010, Thurnau, Germany.5 M. Greferath, C. Roessing, and L. Storme, “Galois geometries and LDPC Codes,” in Leo Storme: Current Topics in Galois Geometry, 2010,

to appear.

1e-10

1e-09

1e-08

1e-07

1e-06

1e-05

0.0001

0.001

0.01

0.1

1 2 3 4 5 6 7 8

Bit E

rror R

ate

Eb/No (dB)

LDPC performance - [43680,39603] inversive space LDPC code (2,6,6)

LDPC CodeUncoded BPSK

Shannon Limit

1e-08

1e-07

1e-06

1e-05

0.0001

0.001

0.01

0.1

1 2 3 4 5 6 7 8

Bit E

rror R

ate

Eb/No (dB)

LDPC CodeUncoded BPSKShannon Limit

Figure 1: Typical shape of a waterfall diagram. Figure 3: Performance of a long high-rate code obtained from an inversive space of dimension 6 and order 2 .

Figure 2: The smallest example of an inversive space.


Engineering Metal Oxide Interfaces

Dr. Michael NolanTyndall National Institute

The EMOIN project commenced in late 2009 with the aim of studying the deposition of metal oxide clusters at metal oxide surfaces as potential materials for hydrogen production, via photocatalysis or chemical reactions, such as water gas shift. In the field of renewable energy first principles computer modelling has the potential to be widely used for the rational design of novel materials to realize the aim of a sustainable energy economy. To date, the work has focused on the following tasks:

initially TiO2, both undoped and doped; with cluster

diameters up to 4nm.

2 clusters adsorbed at the rutile TiO2

(110) surface, with cluster diameters up to 1nm using density functional theory (DFT).

clusters of CeO2 and FeOx at the TiO2 (110) surface,

inspired by recent experimental work on such systems

metal oxides, assessing a number of DFT approaches; in particular using hybrid DFT.

Future work includes:

the Blue Gene P is being used) and over 4nm using density functional tight binding,

2 surfaces (anatase (101), (001), rutile (001))

and improved interatomic potentials to simulate the deposition of clusters under realistic conditions of temperature and impact energy,

clusters for hydrogen production, including water gas shift.

From the initial work on bare clusters we have obtained minimum energy structures for clusters of TiO

2 up to 2nm diameter and investigated how the “band gap” of these systems depends on cluster size and the choice of DFT method; part of this work is in collaboration with University College London. There is a very strong dependence of the band gap on the cluster diameter and DFT method used. Work is underway to use the correlated configuration interaction (CI) approach for the smaller clusters to establish a benchmark for the energies obtained from DFT. In addition, for larger clusters, the minimum energy structure found also depends on the DFT method applied.

We study simple substitutional doping of the clusters at the oxygen site, since experimental work appears to show that such dopants provide good photocatalytic properties,

and find changes to the band gap that indicate band gap modulation via both cluster size and cluster doping (which can be easily achieved during synthesis). This will be further explored as a potentially fruitful avenue to design systems with desired band gaps. Work to investigate co-doping, which has seen much attention recently, will be undertaken.

For clusters deposition at the rutile TiO2 (110) surface, we

have started by adsorbing clusters up to 1nm diameter, giving Ti8O16, in different configurations and allowing them to relax. This will enable us to investigate what features are necessary for strong binding of the cluster to the surface. The most stable cluster adsorption sites appear to maximize the formation of Ti-O bonds between the surface and the cluster. We have also found that the formation of cluster-surface Ti-O bonds is maximized for sub-1nm clusters, which is due to these clusters having exposed, undercoordinated Ti species, whereas larger clusters are terminated by oxygen and effectively hide the reactive, undercoordinated Ti, reducing the interactions between the cluster and the surface. In addition, the rutile (110) surface presents rows of protruding oxygen atoms, whereas other rutile and anatase surfaces are more flat, presenting potentially more surface sites for the cluster to bind with. Work to investigate this aspect of the cluster surface interaction is ongoing. The application of these supported clusters ultimately depends on two of their properties – the band gap and the reactivity. We find that the presence of the smallest clusters at the surface reduces the band gap of the overall system relative to the bare surface, due to the interaction of the cluster electronic states with the surface electronic states, which is a potentially useful result. Initial work on doping of the supported clusters with N and C also indicates formation of band gap states, although we still need to investigate if these dopants will be compensated by formation of other defects.

In terms of reactivity, we are examining two reactions. The first is the formation of oxygen vacancies in the cluster-support assembly and comparing this with oxygen vacancy formation in bulk TiO

2. From this we can see if the cluster is more reactive to vacancy formation and which oxygens are most susceptible; this requires the application of the DFT+U approach to describe the localised Ti3+ states that will form; these systems are presently too large for routine hybrid DFT calculations to be carried out. The second reaction is the oxidation of CO on the clustersupport assembly, which is a good probe of surface reactivity, as well as being important in its own right for CO removal and as a key step in the water gas shift reaction.

With regard to oxygen vacancy formation, we find that this process is energetically more favourable at the supported cluster compared to bulk TiO

2 , with a significant reduction in the oxygen vacancy formation energy and formation of localised reduced Ti3+ species. For the reaction between

CO and the supported cluster, we find that this reaction is also much more favourable at the supported cluster than at the surface of TiO2 and these supported clusters could be potentially useful materials for CO oxidation.

The figure shows a Ti2O4 cluster adsorbed at the rutile TiO2 (110) surface and the formation of an oxygen vacancy whereby an oxygen is removed from the cluster; the formation energy of this oxygen vacancy is much reduced over the bulk material, giving an initial clue as to the improved reactivity of supported clusters compared to bulk TiO

2. Finally, the figure shows that CO reacts at the cluster abstracting oxygen from the cluster and forming a loosely bound CO2 species, which is easily removed. This indicates that these sub-nm supported clusters could be useful for enhancing the reactivity of TiO

2 towards chemical reactions. We are presently working at studying these reactions for different sized clusters to

establish a pattern for cluster reactivity as a function of cluster size; as the cluster size increases, we expect that bulk-like properties including reactivity will be recovered, so that there will be an optimum cluster size for improved reactivity. Work has also begun on studying sub-nm clusters of other metal oxides, including ceria, copper(II) oxide and different iron oxides, all of which are driven partly by recent experimental results and new collaborations with Argonne National Laboratory in the USA. These supported species all appear to show good chemical reactivity for the water gas shift and methanol conversion to formaldehyde and there are suggestions that the latter two species also show enhanced photocatalytic activity as a consequence of the interaction between the different metal oxides. Our work in this field will be able to provide a complete understanding of these processes.


Transport Properties of Organic Macromolecules: Modelling at the Boundaries between Biology and Nano-Electronics

Dr.. Stefano Sanvito Trinity College Dublin

Collaborators: C.D. Pemmaraju, I. Rungger, X. Chen, T. Archer, M. Stamenova, N. Baadji, D. S. Kesanakurthi, N. Caffrey, A. Droghetti, S. Sanvito

Organic electronics or bio-electronics is currently a very active area of research with immense technological potential. The field is mainly concerned with understanding the properties and behaviour of hybrid devices at the boundary between physics and biology. Current research in the area is expected to lead a wide variety of applications in the not too distant future, ranging from revolutionary computer architectures and disposable electronics, to diagnostic tools for genetically driven medicine.

Theoretical modelling of electron transport in organic and biological systems is an extreme computational challenge since the accuracy of sophisticated electronic structure methods must be combined both with large system sizes typically containing thousands of atoms and with quantum transport. However recent advances of order-N methods and the availability of large-scale computational facilities make these calculations possible. The work carried out in this project is based on the electronic transport code smeagol1

which has been developed within our group. The main aim of this project was to study the transport properties of various macro-molecules ranging from magnetic molecules to DNA. In the following sections we present a brief summary of the studies carried out and their outcomes.

TRANSPORT PROPERTIES OF Mn12 SINGLE MOLECULE MAGNETSSingle-molecule magnets (SMMs) represent a unique playground for fundamental quantum physics and exhibit exotic phenomena such as magnetic hysteresis as well as magnetization reversal through quantum tunnelling. Inspired by a recent series of experimental transport measurements on Mn

12 based molecular magnets in single-molecule-transistor devices2, we have conducted pioneering theoretical calculations into the transport properties of a single-molecule magnet based on the Mn12 system. In its ground state, this molecule has a total spin of S = 10 and several low lying spin excitations of the molecule have also been probed experimentally. Simulation of the electron transport through this system is a demanding task requiring over 500 atoms in the simulation cell and approximately 2500 orbitals. In addition the calculations must be performed as spin-polarized. Typical runs involve 64 to 128 CPUs in parallel and a complete I -V up to 0.5 Volt require about 20 bias points. The basic question we have addressed is: can an electrical measurement establish the spin-state of the molecule? Towards this goal we have carried out Smeagol simulations for Mn

12 contacted by gold electrodes via thiol groups (see figure 1). We have investigated two possible configurations. In the first one the molecule is in its ground state, with 8 Mn3+ ions anti-aligned to the 4 Mn4+ in the Mn12O12 molecular core. In the second configuration, the local spin moments on one Mn3+ and one of the Mn4+ ions are flipped in such a way as to create a S = 9 spin state. We find the transport properties of the SMM in both spin states to be dominated by tunnelling type behaviour across the organic functional groups and asymmetric coupling to the leads. As a result, we also observe asymmetric I-V curves under positive and negative bias(see figure 2). Our calculations further show that while the zero-bias conductance can hardly distinguish between the two spin-states, the whole I-V is more informative and

quantitative differences in the differential conductance between the S=10 and S=9 states cen be identified. The most notable difference between the I-V curves obtained for the different spin states is the much higher low-bias current of the ground state S=10 configuration and the presence of negative differential resistances (NDRs), which are specific to the spin state. These are a consequence of orbital re-hybridization under bias, which causes a highly non-linear bias-dependent coupling of the molecular levels to the electrodes(see figure 3). We predict that this is

a general feature of molecular junctions characterized by closely spaced orbital multiplets, such as those appearing in magnetic molecules. Importantly, since both the orbital symmetry and their localization depend on the molecule’s spin state, we expect different I-V fingerprints for different magnetic configurations. This means that the overall molecular magnetic configuration is readable entirely from a single non-spin-polarized current readout. This work was published in Physical Review3.

Figure 1: Transport simulation cell used in this work. Color code: Blue=Mn, Red=O, Green=C, Light Blue=H, Yellow=Au, Dark Yellow=S, Purple=Mn (flipped).

Figure 2: Transport properties of a Mn12 two-probe device. The I-V curves for both the GS and the SF configuration are presented in (a), while

(b) and (c) show the I-V spin-decomposition and the differential conductance respectively.

Figure 3: Spin-resolved transmission coefficient, T(E) as a function of bias for the GS configuration. Red, solid (green, dashed) curves show

the transmission of the majority (minority) spins. The isosurfaces display the wave-functions of a number of molecular levels at different V.


ELECTRON TRANSPORT IN DNA

A clear understanding of the charge transport mechanism in deoxyribonucleic acid (DNA) is of significance to a number of scientific research areas such as Nano-Electronics, biology and gene-sequencing. Based on the insights from experimental and theoretical works over the past decade4, it is now believed that charge transfer in DNA is an extremely complex phenomenon that involves both coherent as well as incoherent transport processes across different length and time scales. Furthermore it depends strongly on several extrinsic factors such as salvation conditions, device geometry etc. As such a quantitative theoretical description of electron transport through DNA remains a very challenging problem.

In this work we carried out first principles electronic transport calculations investigating both the zero-bias and finite bias conductance properties of short poly(dG)-poly(dC) A-DNA strands attached to gold electrodes. By using the non equilibrium Green’s function approach, combined with self-interaction corrected density functional theory (DFT), we calculated the fully self-consistent coherent I-V curves for both break-junction and STM-type device geometries. Firstly, by calculating electron removal energies using accurate Hybrid-DFT functionals, we established the correct level alignment of the molecular levels of the A-DNA strand with respect to the electrodes Fermi level. The zero bias and finite

oligonucleotides (see figure 4). The I-V curves calculated for 6 base-pair pGpC DNA oligomers (see figure 5) show that the currents from coherent electron tunnelling at voltages of around 1 V are several orders of magnitude smaller than those observed experimentally through single DNA transport devices. In order to calculate the parameters needed to estimate the magnitude of incoherent contributions to the electron transport in DNA strands, we carried out classical molecular dynamics and electronic structure calculations. In particular the static as well as dynamic reorganization of molecular energy levels brought about by the presence of a solvent and counter ion system was studied and the magnitude of the energy disorder induced as a result of the dynamical interactions with the polar solvent molecules was calculated (see figure 6). Finally diffusive/inelastic effects were treated at a semi quantitative level via a model employing parameters obtained from DFT calculations and we show that such effects dominate the electronic transport behaviour even in short A-DNA strands attached to metallic electrodes. Within this study, simulation cells of various sizes ranging from 2500 to 5000 atomic orbitals were employed. Computation were typically carried over 128 to 256 CPUs in parallel and I -V curves up to 1.5 Volts with 20 bias points were calculated. A manuscript summarizing this work is currently in preparation.

References1 A.R. Rocha, V.M. Garcia Suarez, S.W. Bailey, C.J. Lambert, J. Ferrer and S. Sanvito, Phys. Rev. B 73, 085414, (2006), Spin and Molecular Electronics in Atomically-Generated Orbital Landscapes.2 H. B. Heersche et. al, Phys. Rev. Lett, 96, 206801 (2006)3 C. D. Pemmaraju, I. Rungger and S. Sanvito, Phys. Rev. B 80, 104422 (2009), Ab initio calculation of the bias dependent transport properties of Mn12 molecules4 R. G. Endres, D. L. Cox and R. R. P. Singh, Rev. Mod. Phys. 76, 195 (2004)

Figure 4: (top left) Projected density of states calculated for the scattering region in the reference G3’C3’G5’C5’ geometry. The highest occupied and lowest unoccupied molecular orbitals on the DNA are indicated by arrows and labelled 1 and 2 respectively. (bottom left) Zero-bias transmission coefficient for the same system. (top right) Charge denisty corresponding to the highest occupied orbital on the DNA strand. This is localized near the 5’ end on the Guanine stack. (bottom right) Charge denisty corresponding to the lowest unoccupied orbital on the DNA strand. This is localized near the 5’ end of the Cytosine stack.

Figure 5: I-V curve for a 6 base pair A-DNA strand attached to Au electrodes.

Figure 6: Spread induced in the electronic energy level positions of a 6 base pair A DNA strand surrounded by a solvent and counterion system. N represnets the number of times a particular energy level is located in the energy range specified by the bin width of the histogram. The top panel shows the data for the HOMO and LUMO levels of the DNA strand while the bottom three panles show data for levels localized on individual bases.

bias transport properties are then calculated within the Smeagol framework by incorporating an approximate self interaction correction so as to reproduce the correct level alignments. We showed that the localization of the electronic wave functions induced by disorder strongly suppresses the magnitude of the elastic conductance in A-DNA Dr. Stefano Sanvito


Faster, Bigger, Better: Supra-Nanoscale Molecular Dynamics for Nanoelectronics

Damien Thompson Tyndall National Institute

Conventional nanoscale molecular dynamics, consisting of nanosecond sampling of nanometre size systems, has seen enormous success in recent years. It has provided crucial high-level mechanistic detail to complement and guide experimental efforts to understand and rationally modify both synthetic nanomaterials and biological systems. Indeed, such is the experimental expertise and computational sophistication that the line between synthetic and biological systems is becoming ever more blurred. In many cases however either significantly larger system sizes or significantly longer sampling times or both are needed to probe eg collective behaviour of large nanoassemblies (on the order of 50-100 nm), conformational change in proteins (large-scale structural rearrangements can be in the microsecond regime). The supra-nanoscale simulations which ICHEC HPC makes possible has provided new information about the mechanisms by which self-assembly and molecular recognition operates, incorporating heretofore inaccessible data on long-range and long-time behaviour, crucial for the design of materials for next-generation (bio)nanoelectronics devices.

NANOSCALE MECHANISMS FOR MONOLAYER SELF-HEALING VIA EXCESS INK TRAPPING IN MICROCONTACT PRINTINGThe control and fine-tuning of surface diffusion is of considerable interest in the nanotechnology community and is a key parameter in the further development of nanopatterning techniques including dip pen nanolithography, high-speed microcontact printing (µCP), edge spreading lithography and microdisplacement printing. We focus on characterizing the influence of surface defects on the diffusion, or “spreading”, of hexadecanethiol “ink” molecules in µCP using hexadecanethiol self-assembled monolayers (SAMs) on gold.

A detailed picture of the specific excess ink/SAM interactions is difficult to obtain and requires a careful combination of experimental and simulation data. Experiments can be used to estimate overall diffusion rates but the underlying nano- and atom-scale diffusion mechanisms are much more difficult to infer from experiments alone. In the present work we take advantage of the capability provided by highly parallel computing to simulate SAM on Au(111) structures with over 3,000 square nanometers surface area (Figure 1). Simulations of this scale are readily compared to scanning tunneling microscopy (STM) and atomic force microscopy (AFM) images, with atomic-scale resolution retained in the simulations. The simulation cells are large enough to allow explicit treatment of defects in the gold layer and to relate these defects to the structure and dynamics in the SAM.

We find a pronounced dependence of the ink/SAM interaction on local ink concentration, which is known to be dependent on the ink type, concentration and printing temperatures used in µ-CP experiments, which serves to explain several features of experimental imaging studies that highlighted the ubiquity of SAM defects at the nanoscale. Our molecular dynamics simulations provide atom-scale details of the competition between spreading and trapping of excess ink molecules in SAM formation, which dictates the ultimate pattern resolution attainable from µ-CP. Local excess ink concentration and the atom-scale features of the SAM film also dictate the orientation of trapped molecules, crucial for SAM on gold applications in, for example, molecular electronics.

Our principal finding is that naturally-occurring SAM film defects can act as barriers to molecular diffusion, providing new data on how excess ink molecules can be integrated into the self-assembling monolayer. Calculated ink spreading and trapping rates (Figure 2) exhibit a pronounced dependence on both excess ink concentration and the atom-scale features of the SAM film, paving the way for the directed molecular assembly of more complex pattern geometries using alkanethiols on gold which could potentially provide feature sizes down to 1-2 nm, approaching the regime of truly writing with molecules on surfaces. In the shorter term, this new data on the nanoscale mechanisms of SAM self-healing and self-limiting excess ink spreading will aid the identification of optimum processing conditions for µ-CP and shows the power of large-scale molecular simulations to complement and deepen experimental knowledge for directed “bottom up” molecular assembly, specifically the optimization of nanopatterning using self-assembly and molecular recognition.

In summary, our massively-parallelized molecular simulations permit multiple, long simulations of system sizes on the order of 106 atoms and reveal how the molecular assembly prerequisite for monolayer healing and growth can occur via “trapping” of excess ink molecules by interfacial domain boundary regions in the self-assembling monolayer. These naturally-occurring domain boundaries act as diffusion barriers. This new data on the nanoscale mechanisms of SAM self-healing and self-limiting ink spreading will aid the identification of optimum processing conditions for µCP and serves as a further illustration of the power of large-scale molecular simulations to fill in gaps in experimental knowledge for directed “bottom up” molecular assembly, specifically the optimization of nanopatterning using self-assembly and molecular recognition1,2,3.

ATOM-SCALE MODEL OF A MOLECULAR TUNNEL JUNCTION FOR NANOELECTRONICSThe aim of this work was computer-aided design in support of the development of synthesis and assembly processes for novel nanoelectronic structures and architectures comprising highly ordered networks of inorganic nanocrystals (NC)

bridged by single functional organic molecules assembled at top-down contact electrodes on SiO

2 substrates. The modelling results support the development of novel processes that will represent the first example of designed single molecule nanostructures and architectures.

A combination of atom-scale quantum mechanical (QM) and molecular dynamics (MD) computer simulations were used to describe the structure, dynamics and energetics of NC-molecule-NC and interactions, and provided the first working atomic-resolution model for the molecular tunnel junction formed between NCs in the formation of a nanoelectronics source-NC-molecule-NC-drain circuit (Figure 3)4.

REFERENCES1 Gannon, G.; Larsson, J.A.; Greer, J.C.; Thompson, D. (2010). Molecular dynamics study of naturally occurring defects in self-assembled monolayer formation. ACS Nano, 4, 921-932.2 Thompson, D. et al. (2010) in preparation3 Thompson, D. et al. (2010) submitted4 Thompson, D. et al. (2010) in preparation

Figure 1 –Plan view for a high excess ink concentration system, showing (A) the initial conformation of excess inks on the relaxed SAM, and (B) the final conformation following 10 ns of free dynamics. Panels (C) and (D) give side views of these initial and final conformations; panel (E) is the same as (D) but with the underlying SAM removed to highlight the trapping.

Figure 2 - The number of excess ink molecules trapped at two different types of domain boundary the (A) and (B) at three different excess ink concentrations (low: 196 molecules, medium: 784 molecules, and high: 3136

molecules) as a function of time.

Figure 3 – The atom-scale features of the molecular dumbbell structure that links gold (Au) NCs. The Au NCs are shown as gold spheres, citrate carbons are grey, oxygens red and sodium counterions purple. The rhenium-based linker molecule has grey carbons, dark blue nitrogens, red oxygens, yellow sulphurs and light blue rhenium atoms. Hydrogen atoms and water molecules are omitted for clarity.

A

C

D

E

B


Topological Phases in Quantum Lattice Models

Dr. Jiri ValaNational University of Ireland, Maynooth

Collaborators: Niall Moran, Ahmet Tuna Bolukabasi, and Dr. Graham Kells, Department of Mathematical Physics, National University of Ireland, Maynooth

Quantum computation as a new computing paradigm has the potential to revolutionise our lives in much the same way as the laser or personal computer. This results from the immense computing power that derives from exotic properties of quantum systems and that reaches beyond the capabilities of any conventional computer. Quantum computation is expected to have a profound impact on information and communication technologies, cryptography, nanotechnology, drug design and material science.

The most significant challenge in building a quantum computer is suppressing the computational errors caused by undesirable interactions between the computer and its environment. Engineering approaches to fault tolerant quantum computation exist but are unrealistically demanding.

Topological quantum computing provides fault tolerance built into the computing hardware. Quantum information here is stored and protected in massively correlated states of certain many body quantum systems. These topological quantum materials form exotic topological phases which are insensitive to local deformations or errors and whose spectral gap suppresses possible non-local error processes.

Evidence for topological phases which are the essential ingredient of topological quantum computing 1, 2 exists both in theory and experiment. They are formed in certain two-dimensional quantum lattice models which provide a setting for both their theoretical investigation and experimental realization in atomic and molecular systems. They are also believed to exist in realistic materials, including fractional quantum Hall systems 3, certain classes of superconductors 4, graphene 5 and others. In addition, a topologically protected quantum bit has recently been experimentally realized using a lattice of super conducting electronic elements 6.

Numerical investigation of topological phases in quantum lattice models provides important insights into the structure, properties and stability of topological quantum materials. The data obtained is of great value for testing and formulating appropriate analytical theories. Using numerical tools it is possible to observe signatures of topological phases which include for example topologically dependent ground state degeneracy, robust spectral gap, and topological entanglement entropy. We employ two complementary categories of numerical tools: exact methods and approximative variational approaches. We also consider stochastic methods like quantum Monte Carlo techniques but their use is limited due to a sign problem.

Exact methods provide numerically exact results and can give access to all the properties of the system under study. These methods account for all degrees of freedom in the given system. However the number of degrees of freedom in a quantum system grows rapidly with the system size. Though this naturally limits exact computations to small system sizes, it provides important insights into finite size effects and also acts as an essential benchmark for approximative variational techniques.

The exact diagonalisation code we have developed, named DoQO7 for Diagonalisation of Quantum Observables constructs and iteratively diagonalises the Hamiltonian observable for spin ! and spinless fermionic systems to retrieve the low energy spectrum and states of these systems. Being able to handle interactions and observables involving more than two sites and efficiently exploiting distributed memory architectures makes DoQO a unique tool for quantum lattice models with topological order. The recent addition of the ability to exploit the symmetries which conserve parity, filling and momentum allows for larger systems to be treated, while at the same time providing access to the values of additional quantum numbers The Blue Gene/P continues to be the platform of choice for running DoQO. The multiple dedicated communications networks provide unrivalled scaling performance and allows us to keep pushing the limits of the calculations that are possible.

Last year we performed calculations on the Blue Gene/P as part of our project on the investigation of topological phases in quantum lattice models 8, 9, 10, 11, 12. Carrying on from the work on the finite size effects in the Kitaev model 9 we were able to perform calculations for a system of 36 spin ! particles and observe that the ground state for this system is indeed four fold degenerate up to the sixth order in perturbation theory. This 36 spin ! system has over 64 billion degrees of freedom The fact that parity is conserved along each row and that the model is translationally invariant in the horizontal direction means that the system can be divided into 192 blocks each with a third of a billion degrees of freedom. Each of these blocks can then be treated individually.

Over the last year an international collaboration with researches at the University of Amsterdam has been initiated. The aim of this collaboration is to investigate the possibility of the existence of so called `super-topological’ phases in supersymmetric lattice models 13 on novel lattice geometries. Figure 1 shows a plot of some preliminary results of calculations performed using DoQO for these models. In addition, we have continued collaboration with researchers from the University of Vienna on lattice models relevant to superconductors with high transition temperature 16.

The exact solutions to the Kitaev honeycomb lattice 10 and Yao-Kivelson star lattice 14 models which were initially verified using ICHEC resources 12, have been now employed in a number of new computational projects. At production stage is the effort to calculate the Berry phase associated with anyonic braiding 15. These calculation will provide explicit veri cation for the existence of a non abelian topological phase in the Kitaev model. The Octave package on Stokes is being used for these calculations. Coming online in the coming months will be the new project to simulate large scale circuit structures using the edge states observed in these models, which are Bogoliubov de-Gennes type topological insulators 11.

Figure 1: One point functions for the ground state space of the supersymmetric lattice model on an open 36 site square octagon lattice with both horizontal and vertical defects on the centre plaquette. There is a three fold degenerate ground state space in this case. The plots labelled (a) show the one point functions in the basis chosen such that the tails from the vertical defect can be clearly seen. In the plots labelled (b) the basis was chosen such that the tails from the horizontal defect can be clearly seen.

With the unique tools we have developed it is possible to get accurate numerical results for a very large range of interesting lattice systems which are relevant to topological quantum materials. With continued refinements, improvements and innovations to these approaches and the continual improvements in the available computational resources we can continue to improve the accuracy, applicability and efficiency of these methods. This will facilitate a better understanding of topological quantum materials and their applications in topological quantum computation.

References1 G. P. Collins, Computing with quantum knots. Sci. Amer. April (2006).2 C. Nayak, S. H. Simon, A Stern, M. Freedman, S. Das Sarma , Non-Abelian Anyons and Topological Quantum Computation. Rev. Mod. Phys. 80, 1083 (2008).3 R. L. Willett, L. N. Pfeiffer and K. W. West, Measurement of filling factor 5/2 quasiparticle interference: observation of charge e/4 and e/2 period oscillations PNAS 106 22 8853 (2008);4 C. J. Bolech and Eugene Demler, Observing Majorana Bound States in p-wave Superconductors Using Noise Measurements in Tunneling Experiments Phys. Rev. Lett. 98, 237002 (2007).5 K. S. Novoselov, E. McCann, S. V. Morozov, V. I. Fal’ko, M. I. Katsnelson U. Zeitler, D. Jiang, F. Schedin and A. K. Geim, Unconventional quantum Hall effect and Berry’s phase of 2pi in bilayer graphene. Nature Physics 2, 177 (2006).6 S. Gladchenko, D. Olaya, E. Dupont-Ferrier, B. Douot, L. B. Io e and M. E. Gershenson, Superconducting nanocircuits for topologically protected qubits. Nature Physics 5, 48 (2008).7 N. Moran, G. Kells, J. Vala, Diagonalisation of Quantum Observables on Regular Lattices and General Graphs, In preparation.8 G. Kells, A. Bolukbasi, V. Lahtinen, J. K. Slingerland, J. K. Pachos, J. Vala, Topological degeneracy and vortex manipulation in Kitaev’s honeycomb model. Phys. Rev. Lett. 101, 240404 (2008).9 G. Kells, N. Moran and J. Vala, Finite size effects in the Kitaev honeycomb lattice model on a torus. J. Stat. Mech. Theory Exp. P03006 (2009).10 G. Kells, J. K. Slingerland and J. Vala, A Description of Kitaev’s Honeycomb model with Toric-Code Stabilizers. Phys. Rev. B 80 125415 (2009).11 G. Kells and J. Vala, Chiral edge modes on a p-wave magnetic spin model, In preparation.12 V. Lahtinen, G. Kells, A. Carollo, T. Stitt, J. Vala and J. K. Pachos, Spectrum of the non-abelian phase in Kitaev’s honeycomb lattice model. Ann. Phys. 323, 2286 (2007). 13 P. Fendley, K. Schoutens, and J. de Boer, Lattice Models with N=2 Super- symmetry. Phys. Rev. Lett. 90, 120402 (2003).14 G. Kells, D. Mehta, J. K. Slingerland and J. Vala, Exact results for the star lattice chiral spin liquid. Phys. Rev. B 81 104429 (2010).15 A. Bolukbasi, N. Moran, G. Kells and J. Vala, Berry phase calculations in non-abelian phase of Kitaev honeycomb lattice model, In preparation.16 E. Rico, R. Hbener, S. Montangero, N. Moran, B. Pirvu, J. Vala and H.J. Briegel, Valence Bond States: Link models. Annals Phys 324 1875 (2009).


4.2 Class B Projects

CHEMISTRY

COMPUTING

EARTH SCIENCE

ENGINEERING

LIFE SCIENCE

PHYSICS

ASTROPHYSICS


ICHEC CLASS B PROJECTS

ASTROPHYSICSDr. Turlough Downes, DCU and DIASTurbulence in Molecular Clouds: Beyond Magnetic Flux Freezing

Dr. Mark Hannam, UCCGravitational-Wave Astronomy from Black-Hole Collisions

Dr. Andrew Lim, DIASNumerical Studies of the Evolution of Structures in H II Regions

Dr. Gareth Murphy, DIAS3D PIC Simulations of Near and Mildly Relativistic Shocks

Dr. Stephen O’Sullivan, DCUStochastic Acceleration of Cosmic Rays in Giant Radio Lobes

CHEMISTRYDr. Jeremy Allen, TCDDefect Properties of Tin Monoxide

Dr. Simon Elliott, Tyndall InstituteGrowth of Rare Earth Metal Oxides as High-k Dielectrics

Dr. Niall English, UCDEffects of Electromagnetic Fields on Proteins

Dr. Natasha Galea, TCDQM/MM Study of Isolated Surface Defects on Ceria

Dr. Andreas Larsson, Tyndall InstituteDoped Fullerenes and Peapod Nanotubes – a First-Principles Investigation

Dr. Benjamin Morgan, TCDElectronic Structure Studies of the Defect Chemistry of TiO2

Dr. Michael Nolan, Tyndall InstituteFirst Principles Simulations of Shape Memory Alloys based on NiTi

Dr. Michael Nolan, Tyndall InstituteMetal Oxide Surfaces and Interfaces: at the Frontier of First Principles Simulations

Dr. Michael Nolan, Tyndall InstituteOxidation of Ternary Shape Memory Alloys based on NiTi using First Principles Simualation

Prof. John Simmie, NUI GalwayBurnQuest: Towards a World Class Combustion Chemistry Centre

Dr. Damien Thompson, Tyndall InstituteAtom-Scale Structure, Dynamics and Energetics of Alkanethiol Bilayers as Nanostructure-Directing Agents

Dr. Damien Thompson, Tyndall InstituteThe Interaction of Room Temperature Ionic Liquids with Phospholipid Bilayers: a Computational Investigation

Prof. Graeme Watson, TCDDeNOx Catalysis on Ceria Surfaces

Prof. Graeme Watson, TCDOptimisation of Ionic Conductivity in Intermediate Temperature Solid Oxide Fuel Cells

Prof. Graeme Watson, TCDOptimising the Oxygen Storage Capacity of Ce1 –xMxO2 (M = Noble Metal)

Prof. Graeme Watson, TCDThe Behaviour of Hydrogen Defects in p-type Oxides

COMPUTINGDr. Patrik Lambert, DCUOptimising Hybrid Machine Translation

EARTH SCIENCESDr. Rodrigo Caballero, UCDPresent and Future Statistics of Extreme European Storms

Dr. Estelle Roux, DIASJoint Inversion of Magnetotelluric and Surface Wave Measurements: Application to a Real Dataset from Northern Canada

Dr. Michael Hartnett, NUI GalwayCoupled Ocean Atmosphere Climate Modelling

Prof. Peter Lynch, UCDClimate Change and the Future Wind Energy Resource of Ireland

Dr. Gareth O’Brien, UCDSeismic Source Modelling and Propagation in Complex 3D Earth Models

ENGINEERINGDr. Marco Grimaldi, UCDEvaluation of Novel Features

Dr. Noel Harrison, NUI GalwayA Computational Analysis of the Microscale Forces that Drive Cell-Derived Tissue Formation: A Tissue Engineering Solution

Dr. Patrick McGarry, NUI GalwayModelling the Active Evolution of Focal Adhesions and Stress Fibres in Cells as a Result of Mechanical Stimuli

Prof. Peter McHugh, NUI GalwayDevelopment of Discrete Finite Element Cell Model

LIFE SCIENCESDr. David Fitzpatrick, NUI MaynoothSearching for Evidence of Interkingdom Horizontal Gene Transfer between Bacteria and CTG Species

Dr. Gemma Kinsella, NUI MaynoothThe Retinol Binding Protein System and Type 2 Diabetes

Prof. Peter McHugh, NUI GalwayHigh Resolution Modelling of Bone Loss in Vertebral Trabecular Bone

Dr. James McInerney, NUI MaynoothThe Tree of One Hundred Percent

Prof. Denis Shields, UCDFunctional Variation in Evolution and Populations

Dr. Charles Spillane, UCCIdentifying Taxon Restricted Genes in Higher Plants and Epigenomics of Genome Dosage in Arabidopsis Thaliana

PHYSICSDr. Thomas Archer, TCDMultiferroic Tunneling Junctions

Dr. Nadjib Baadji, TCDScanning Tunneling Microscopy for Magnetic Molecules

Dr. Sri Chaitanya Das Pemmaraju, TCDDefects and Impurities Induced Magnetism in Oxide Semiconductors and Insulators

Dr. Claude Ederer, TCDFunctional Properties of Multiferroic Heterostructures

Dr. Maria Elena Grillo, Tyndall InstituteAtomic-Scale Characterisation of Charge Traps in Sonos Memory Devices, and of the Effect of Surface Reactivity on Growth Rates in AL2O3 ALD

Dr. Niall English, UCDComputational Modelling of Materials for Artificial Photosynthesis

Dr. Giorgos Fagas, Tyndall InstituteSemiconductor and Molecular Nanowire Simulation for Technology Design

Dr. Giorgos Fagas, Tyndall InstituteElectronic Properties of Oxidised Silicon Nanowires

Prof. Stephen Fahy, UCCNanoscale Simulators in Ireland (NSI) Graduate Programme Practicals

Dr. Zoltan Neufeld, UCDAggregation and Collective Motion of Microorganisms in Fluid Flows

Dr. Jim Greer, Tyndall InstituteAtomic Scale Model Interfaces between High-K Silicates and Germanium

Dr. Charles Patterson, TCDSpin Polarons in p-type Magnetite

Prof. Stefano Sanvito, TCDSimulating Electronic Transports in Scanning Tunneling Microscopy

Dr. Jonivar Skullerud, NUI MaynoothHot QCD Simulations on Anisotropic Lattices

Dr. Maria Stamenova, TCDAb Initio Simulations of Spin-Dynamics


ASTROPHYSICS

GRAVITATIONAL-WAVE ASTRONOMY FROM BLACK HOLE COLLISIONS

Mark Hannam, Sascha Husa, Niall O MurchadhaUniversity College Cork

Following the first detection of gravitational waves (GWs) in the next few years, the new field of gravitational-wave astronomy will begin in earnest. One of the strongest and most interesting sources of GWs is the inspiral and merger of black holes. To detect black-hole-binary GW signals, and to use them to characterize their sources, we need accurate predictions of what the waveforms look like. The only way to calculate the GW signal predicted by Einstein’s full general theory of relativity for black-hole collisions is to solve Einstein’s equations numerically. Such simulations only became possible in 2005, and only now are we beginning to perform simulations suitable for GW astronomy applications In this project we will perform a large survey of the parameter space of the most complex black-hole binaries, where each black hole is spinning, and the spin direction (and the binary’s orbital plane) precesses during the inspiral.

ASTROPHYSICS

NUMERICAL STUDIES OF THE EVOLUTION OF STRUCTURES IN H II REGIONS

Dr Andrew LimDublin Institute for Advanced Studies

We have developed a radiation-magneto-hydrodynamics grid-based code to study the effects of ionisingradiation on the interstellar medium. We propose to use this code to model the physics of the formation and evolution of structures observed in H II regions around young stars, a good example being the star forming pillars in M16 (the Eagle Nebula). Our code is efficiently parallelised using MPI, enabling us to perform very high spatial and temporal resolution simulations on the parallel cluster Walton.

ASTROPHYSICS

TURBULENCE IN MOLECULAR CLOUDS: BEYOND MAGNETIC FLUX FREEZING

Ms. Aoife Jones & Dr. Turlough DownesDublin City University and Dublin Institute for Advanced Studies

Stars are born in molecular clouds, when a dense clump of gas in that cloud undergoes gravitational collapse. In Giant Molecular Clouds, many stars can be formed in the cloud before the cloud itself collapses under gravitational pressure. Astronomical observations suggest that this process is delayed by some internal turbulent energy that supports that cloud against collapse. One likely candidate for such turbulence are newly formed stars within the cloud, which inject energy back into the cloud through jets and outflows.

As a protostellar jet propagates into the surrounding cloud, there is a velocity difference between the material within its bowshock and that outside of it. This region is thus prone to shear instabilities. One such instability is the Kelvin-Helmholtz instability. Our project focuses on this particular instability as it can lead to the turbulent mixing of the two layers of material involved.

The Kelvin-Helmholtz instability has been widely studied in astrophysical applications using a simplifying approximation in which the magnetic field is assumed to be “frozen-in” to the plasma. In molecular clouds, a high fraction of neutral hydrogen atoms means that this idealised approximation is invalid. A numerical code which does not require this approximation has been developed by Dr. Turlough Downes and Dr. Stephen O’Sullivan. This allows more realistic models to be used. Incorporating non-ideal effects is computationally expensive however, as we wish to simulate regions of billions of kilometres in length, while still capturing non-ideal dynamics that occur much smaller length scales.

The Kelvin-Helmholtz instability causes a distinct effect at the interface of the two layers between which it occurs, known as the “Kelvins cat’s eye” vortex. In the figure shown, this vortex can be seen in the gas as the instability causes the two layers on either side of the grid to wind up together. The effect that this has on the magnetic field will depend entirely on how well the magnetic field is tied to the plasma. This is where the non-ideal effects come into play, and our results show that the different non-ideal effects that are expected to occur in molecular clouds can cause physically significant differences in the evolution of the magnetic field during the development of the instability.

Figure 1: Plot of the magnitude and vector field of the velocities in the system. The winding-up of the fluid on the left-hand and right-

hand sides of the grid gives the “Kelvin’s cat’s eye” vortex.

ASTROPHYSICS

3D PIC SIMULATIONS OF NEAR AND MILDLY RELATIVISTIC SHOCKS

Dr. Gareth MurphyDublin Institute for Advanced Studies

The aims of the proposal are (i) to explore at kinetic level with particle-in-cell simulations the formation of mildly relativistic shocks for a range of magnetic field strengths, orientations, plasma temperatures and density ratios, (ii) to search for the onset of Fermi acceleration and finally (iii) to compare with available observations of supernovae remnants and gamma-ray bursts. In order to do this a pre-existing code (PSC) will be used, to test its portability and scalability on the Stokes cluster in preparation for a future Class A project.

ASTROPHYSICS

STOCHASTIC ACCELERATION OF COSMIC RAYS IN GIANT RADIO LOBES.

Dr Stephen O’Sullivan,Dublin City University

The target of this investigation is the acceleration of protons via the second-order Fermi process in the giant lobes of radio galaxies. Such sites are a candidate for the source of ultra-high energy cosmic rays. I will seek to confirm predictions for conditions necessary for consistency in the relevant timescales in the quasi linear framework via numerical simulation.

ANNUAL REVIEW 2009 PAGE 47PAGE 46 ANNUAL REVIEW 2009

CHEMISTRY

DEFECT PROPERTIES OF TIN MONOXIDE

Dr. Jeremy Allen & Prof. Graeme WatsonSchool of Chemistry, Trinity College Dublin

The identification and understanding of p-type conductors is particularly attractive for use in the generation of thin-film transistors (TFTs). One such material which has been highlighted is tin monoxide. Epitaxially grown thin film structures of tin oxide have been shown to give enhanced performance over previously reported p-type channel oxide TFTs.

Tin monoxide has been previously studied using standard DFT, in terms of understanding both the electronic distribution and structure of the materials. However, accurate simulation of intrinsic defects has been lacking. In this study we will use hybrid-DFT with additional attractive energy corrections to account for the long-range van der Waals interactions not modelled using DFT methodologies. This will enable us to elucidate not only the correct electronic structure of defects but as the conduction mechanism in SnO.

CHEMISTRY

GROWTH OF RARE EARTH METAL OXIDES AS HIGH-K DIELECTRICS

Simon D. Elliott, Michael Nolan and Aleksandra ZydorTyndall National Institute

Researchers at Tyndall National Institute used ICHEC resources as part of a collaboration called REALISE (http://www.tyndall.ie/realise), funded by the European Commission under the Sixth Framework programme. The project ended in 2009 with the successful development of a new dielectric material, along with a processing technique that is suitable for semiconductor manufacturing. The new material is based on zirconium oxide, with the addition of just enough lanthanum oxide to ensure an atomic structure that responds strongly to electric fields. Two areas of information technology were chosen to test the new material and process – mobile phone components and flash memories – with results that exceed today’s best technologies by a factor of three.

Everyone who uses a games console or digital camera is amazed by the greater and greater amounts of data that memory cards and USB flash drives can hold. One of the obstacles to storing more data in flash memory is the interference between adjacent memory cells. Inserting new dielectric materials between adjacent cells can help.

The partners in the REALISE project devised one such material, a mixed oxide of lanthanum and zirconium. The deposition chemistry was examined through in situ experiments at the University of Helsinki (Finland) and via simulations at Tyndall National Institute and ICHEC, so that the technique could be perfected for laying down the new material in nanometre-

thin films with the correct atomic structure. Nanoscale measurements at CEMES (France) and MDM laboratory (Italy) confirmed the structure and also indicated a three-fold improvement in the relevant dielectric properties over alumina. For Numonyx (Italy), Europe’s leading flash memory manufacturer, the new material is a candidate for flash memory cards and USB drives. According to Numonyx, “new materials like this are needed for our higher density flash memories that should hit the market in 2014.”

The second test structure was for mobile phones. Today’s mobile phones contain about a dozen chips, connected together with hundreds of more mundane electronic components. Decoupling capacitors are one such component, with over 100 of them hiding inside the typical mobile phone. The size and cost of the phone can be massively reduced by patterning these components onto the chips. The challenge however is to develop a high performance capacitor that can be fabricated as part of the silicon chip.

In the REALISE project, the mixed oxide of lanthanum and zirconium was deposited in a thin sandwich between metal layers in order to fabricate a capacitor. High purity lanthanum and zirconium chemicals were developed by the University of Liverpool (UK) and SAFC Hitech (UK). SAFC Hitech say that “we are now marketing the new chemicals from the REALISE project, reinforcing our position as a leading supplier to the electronics industry.” Crucially, the processing technology was brought to the quality needed for chip manufacture by ASM Microchemistry (Finland). NXP semiconductors (Netherlands) were thus able to fabricate high performance decoupling capacitors that are three times smaller than the current record, with the bonus of double the working lifetime. NXP say that “in terms of surface area on the chip, the new capacitors should cost 70% less to produce.”

Quantum mechanical calculations yielded this model of the chemical reactions during atomic layer deposition of lanthanum in LZO (blue=La, red=O, grey=C, white=H).

High resolution transmission electron microscopy atomic structure image near the interface of LZO material deposited on silicon, with a schematic showing its location in the flash memory stack.

Flash memory chips manufactured by one of the project partners.


CHEMISTRY

EFFECTS OF ELECTROMAGNETIC FIELDS ON PROTEINS


The dominant focus of this research project is to investigate the possible nature of athermal effects of electromagnetic (e/m) fields on the structure and dynamics of enzymes. Experimental studies of (e/m) fields have shown potential destruction of enzyme activity after relatively brief exposure to e/m fields. In order to better understand these effects we have carried out non-equilibrium molecular dynamics (MD) simulations of various mutants of Hen Egg White lysozyme (HEWL) to gain atomic level detail of these effects.

These simulations were performed at 300 K and 1 bar in the presence of both external static electric and low-frequency microwave (2.45 GHz) fields of varying intensity. Significant non-thermal field effects were noted, such as marked changes in the protein’s secondary structure relative to the zero-field state, depending on the field conditions, mutation and orientation with respect to the applied field. This occurred primarily as a consequence of alignment of the protein’s total dipole moment with the external field, although the dipolar alignment of water molecules in both the solvation layer and the bulk was also found to be influential. Substantial differences in behavior were found for proteins with and without overall

net charges, particularly with respect to translational motion. Localized motion and perturbation of hydrogen bonds was also found to be evident for charged residues.

All simulations were carried out using a modified version of GROMACS 3.3.1, with a time varying e/m field. Starting coordinates for HEWL (PDB ID: 2lzt) were obtained from the PDB Data Bank. The enzyme was solvated in SPC water and heated to 300K in steps of 60K starting from 60K.

A steep rise in the C - RMSD is seen in the first 1.5 ns of a 0.05 VÅ-1 static electric field simulation of the wild-type protein with the formation of a plateau in the latter 2.5ns of the simulation (cf. Fig. 1a) [1]. It is of interest to ascertain which regions of the protein experience the most significant changes in structure and mobility in the first 1.5 ns of the trajectory; C - RMSF curves for the two regions of the simulation are shown in Fig. 1b. A striking difference appears in two particularly charged regions of the protein (residues 43-55 and residues 65-75). The two segments are located in the beta domain of the lysozyme and are relatively solvent exposed. Within the first 150ps of the e/m field simulation, these hydrogen bonds are disrupted due to localized translational forces on their charged residues and this area experiences significant positional fluctuations (cf. Fig. 1b) 1. The relative exposure of these regions to water results in easier perturbation of the hydrogen bonds arising from the localized translational motion.

1. N.J. English, G.Y. Solomentsev and P. O’Brien, J. Chem. Phys., 131(3), 035106 (2009).

Figure1: C – RMSD (a) and RMSF (b) curves for the wild-type protein in a 0.05 VÅ-1 static electric field acting along the laboratory + z-axis. The RMSF values in panel b were calculated for time 0 – 1.5ns (denoted by the black curve) and for time 1.5ns – 4ns (shown by the red curve).

CHEMISTRY

QM/MM STUDY OF ISOLATED SURFACE DEFECTS ON CERIA

Natasha M. Galea and Graeme W. WatsonSchool of Chemistry, Trinity College Dublin, Ireland

Ceria (CeO2) plays a significant role in several industrial catalytic processes, for example, aiding the removal of environmentally harmful molecules from exhaust fumes. The technological importance of CeO

2 is centered on its oxygen storage capacity (OSC); it’s ability to rapidly store / release oxygen atoms via a reversible oxidation / reduction mechanism. Originating from the ease with which oxygen atoms are removed to create oxygen vacancies (defects), the OSC mechanism of ceria is aided by the reduction of two neighbouring cerium ions positioned adjacent to the defect site which removes the need for any alteration in overall charge;

Based on the OSC properties previously observed experimentally and theoretically for low index surfaces of ceria, it is important that we fully understand the mechanisms involved during the formation of defects. Within the literature, high level calculations focus on materials containing a large concentration of defects, requiring periodic low index crystal surfaces, where adjacent oxygen vacancy sites have the ability to influence each other. However, few high level computational studies are available with respect to lowconcentration / isolated oxygen vacancies. In light of the nature of previous studies, we have examined the formation of an isolated oxygen vacancy on the 111 surface of CeO

2 with the embedded-cluster quantum mechanics / molecular mechanics approach.

Embedded cluster QM/MM (using the software ChemShell) enables a central cluster of atoms within a larger hemisphere to be treated using a higher accuracy computational approach (QM: Hartree-Fock), while the remaining atoms on the periphery of the hemisphere are treated via a less accurate and therefore less time-consuming method (MM: interatomic potentials). A layer of point charges surrounds the entire hemishphere. This technique permits us the opportunity to model materials with a large nanostructure in a non periodic environment.

Embedded QM cluster : Central (green) atom illustrates O ion removed (defect), Ce ions illustrated by neighbouring (white) atoms, and interface (Ce) ions in grey.

Embedded-cluster QM/MM allows the comparison of reduced surfaces of ceria containing low and high concentrations of oxygen vacancy defect sites.


CHEMISTRY

FIRST PRINCIPLES SIMULATIONS OF SHAPE MEMORY ALLOYS BASED ON NiTi


In this project, we proposed to study the growth of oxides and the resulting interfaces of the shape memory alloy NiTi using first principles density functional theory (DFT). These aspects of NiTi are crucial, since a TiO

2 passivating layer grows naturally at the surface of an NiTi sample. This work is in collaboration with Dr. Tofail Syed at the University of Limerick.

During 2008, we made heavy use of the new Stokes cluster making substantial progress in studying the reaction of oxygen at the (110) surface of NiTi. Oxygen coverages from a single O atom to 1 monolayer of O

2 at the surface have been studied and some resulting adsorption configurations are shown in the accompanying figure.

Oxygen adsorption is strongly exothermic, and results in Ti at the surface being pulled out of the surface layer. When O2 impinges at the surface it adsorbs dissociatively, with each O atom coordinating to surface Ti, and pulling these atoms out of the surface layer. Addition of further O

2, results in O2 dissociation and interaction with Ti and at full coverage, all surface Ti are pulled out of the surface layer, leaving an Ni-rich region. This latter result is consistent with experimental and provides useful insights into understanding growth of the TiO

2 layer at the NiTi surface

Initial work on atomic models of the TiO2-NiTi interface has considered the (001) TiO2 surface on the (110) NiTi surface, initially assuming both are perfect. Preliminary results indicate Ti atoms in the region of the interface at intermediate between metallic Ti and Ti4+ and that Ti vacancies can form rather easily at the interface.

CHEMISTRY DOPED FULLERENES AND PEAPOD NANOTUBES – A FIRST-PRINCIPLES INVESTIGATION

Dr. Andreas LarssonTyndall National Institute

Within this project we have investigated how metal-phthalocyanine molecules (specifically, SnPc, PBPc and CoPc) interact with the Ag(111) surface. This study has been performed by the doctoral candidate Mr. Jakub Baran and Dr. Andreas Larsson at the Tyndall National Institute, in collaboration with the experimental Nanoscience Group led by Prof. Philip Moriarty at the University of Nottingham. With the use of the ICHEC facilities we have been able to compute these systems, which contain more than 220 atoms, using quantum mechanical methods in the form of density functional theory (DFT) simulations to compare with the experimental normal-incidence x-ray standing wave (NIXSW) spectroscopy, x-ray photoelectron spectroscopy (XPS) and scanning tunneling microscopy (STM). The molecule-surface distances predicted by the DFT calculations are in good agreement with the NIXSW results, and the STM images can be matched with our calculated electron densities in the relevant energy ranges.1 We have also studied the conformational inversion of the shuttlecock-shaped phthalocyanine molecules GePc, SnPc and PbPc using DFT, which has been proposed as a mechanism for energy harvesting and nanomechanical devices. We have found the same mechanism of inversion for GePc and SnPc but a different one for PbPc. Inversion proceeds through two transition states, separated by a planar local minimum, for GePc and SnPc, but through one transition state distorting the phthalocyanine macrocycle for PbPc. The energy barrier of inversion is 4.27 eV for PbPc and 2.12 and 3.16 eV for GePc and SnPc, respectively. Such high barriers are unlikely to be overcome at normal experimental conditions, and in many cases alternative explanations for switching between “up” and “down” conformation need to be sought, such as ionization assisted inversion or even flipping over of the molecules. 2

1 Theoretical and Experimental Comparison of SnPc, PbPc, and CoPc Adsorption on Ag(111), J. D. Baran, J. A. Larsson, R. A. J. Woolley, Y. Cong, P. J. Moriarty, A. A. Cafolla, K. H. G. Schulte, Phys. Rev. B 81 (2010) 075413.2 Inversion of the Shuttlecock Shaped Metal Phthalocyanines MPc (M = Ge, Sn, Pb) – A Density Functional Study, J. D. Baran, J. A. Larsson,

Phys. Chem. Chem. Phys. (2010) DOI: 10.1039/b924421b.

CHEMISTRY ELECTRONIC STRUCTURE STUDIES OF THE DEFECT CHEMISTRY OF TIO2

Dr Benjamin J. Morgan and Prof. Graeme W. WatsonSchool of Chemistry, Trinity College, Dublin 2, Ireland.

ABSTRACT

TiO2 is one of the most widely investigated metal oxides, with numerous technological applications including photocatalysis of environmental pollutants, solar energy production, water splitting for hydrogen production, and possible spintronic devices. In all these cases crystalline defects can have large effects on the behaviour of experimental samples, and an understanding of both the relative stability of competing defects under variant synthesis conditions, and the effect of these on the electronic properties of the material, is necessary to optimise samples for real-world applications.

We will use high quality density functional theory calculations to study a range of intrinsic and extrinsic TiO2 defects in both rutile and anatase. This will allow us to study the electronic and geometric structures of these competing defects, as well as their relative stability through calculating the formation energies. A detailed analysis of the electronic structure of comparable defects between the two polymorphs, rutile and anatase, will also be undertaken, to facilitate a greater understanding of the experimentally observed differences in photochemical behaviour.

CHEMISTRY METAL OXIDE SURFACES AND INTERFACES: AT THE FRONTIER OF FIRST PRINCIPLES SIMULATIONS


ABSTRACT

The surfaces and interfaces of metal oxides are of paramount importance in many technologies, as well as being of fundamental interest. However, despite experimental advances, an atomic level understanding of the properties of metal oxide surfaces, thin films and interfaces – all of which play a role in the use of oxides - are lacking. In this regard, modern first principles calculations on high performance computing are positioned to provide this atomic level detail, allowing both an understanding and predictions of surface and interface behaviour.

Areas of particular interest include metal clusters and oxide clusters on oxide surfaces, the liquid-oxide interface, thin film growth and surface reactivity. Within these areas we are interested in (i) Cu

2O surfaces and reactivity, (ii) metal thin film growth on TiO2, (iii) doping of oxide surfaces and (iv) supported oxide and metal clusters on different oxide substrates. First principles density functional theory calculations will allow new insights into surface and interface properties at the atomic level.

Adsorption sites for oxygen at the (110) surface of B2 structured NiTi. (a) atomic O, (b) O2, (c) 3O2, (d) 4O2


CHEMISTRYOXIDATION OF TERNARY SHAPE MEMORY ALLOYS BASED ON NITI USING FIRST PRINCIPLES SIMULATION


In 2009, this project investigated the initial stages of oxidation of NiTi shape memory alloy and the atomic level details of the resulting interface using first principles density functional theory (DFT) simulations with the VASP code. This is an important process to study because a passivating TiO

2 layer grows spontaneously on NiTi that has been treated to produce the shape memory effect, but the atomic level details of this process are as yet unknown and knowledge of the oxidation properties of NiTi will be useful for optimizing processing of the alloy. To study the oxidation of NiTi, we allowed oxygen molecules, O

2, to react with the (110) surface of the NiTi alloy and relaxed the surface-adsorbate structure using DFT. We find that upon reaction of O2 with the NiTi alloy, oxygen spontaneously dissociates and pulls Ti out of the surface forming TiO2. This overall finding is consistent with the emergent consensus on formation of the TiO2 layer, as a process in which oxygen consumes Ti from the alloy.

A detailed analysis of the interaction of successive O2 molecules with the alloy shows that (i) O2 always dissociates with a large energy gain of at least 3 eV per oxygen, (ii) upon interaction with oxygen, surface Ti atoms are pulled out of their lattice sites to form Ti-O bonds, weakening, and ultimately breaking, Ni-Ti bonds, (iii) an Ni-rich region will be found below the oxide layer and (iv) these Ti show characteristics of both metallic Ti and oxidised Ti (as in TiO

2).

Upon addition of 2 monolayer of O2 (that is 8 O2 molecules in our model), all surface Ti atoms are pulled out of their lattice sites and are present in a single thin layer of TiO2 atop an Ni-rich NiTi surface. This process is indicated in the accompanying figure showing the final structure after reaction with 1O2, 4O2 and 8O2. This occurs entirely spontaneously and provides an explanation for many experimental findings. The work has been published in Biomaterials (Biomaterials, 2010, vol. 31, p. 3439) and presented at a number of conferences.

CHEMISTRY BURNQUEST: TOWARDS A WORLD CLASS COMBUSTION CHEMISTRY CENTRE

Prof. John SimmieNational University of Ireland, Galway

BurnQuest, which was funded by an EU Marie Curie Transfer of Knowledge award, finished at the end of August 2009. The objective of the four year project was to assist in the creation of a world-class combustion chemistry centre and this has been successfully achieved. The Combustion Chemistry Centre (C3) in NUI Galway is pre-eminent in this specialised field and has been recognised as such by funding agencies such as Science Foundation Ireland, the EU Framework Programme 7 and just as importantly by major industrial companies in Canada, the USA, continental Europe and the Middle East. In other words, C3 is now a global player — Ireland does not manufacture either internal combustion engines or gas turbines and has no indigenous liquid transport fuel industry to speak of — which was always our objective.

The detailed aim of the project was to learn from experts in the fields of computational quantum chemistry, mechanism reduction and sensitivity analysis in order to complement our unique experimental facilities and our skills in the formulation of detailed chemical kinetic models.

Naturally in order to apply high-level advanced quantum methods to the calculation of the energetics and kinetics of reactions a supercomputer was required and the original ICHEC grant application used our input to secure the funding.

So we were involved from the earliest days on a succession of systems of variable usefulness and unfortunately frequently hampered by ill-judged time constraints on job times. It frequently seemed to us that only certain types of jobs were favoured and that our requirements for long run times with a small number of processors was out of favour with ICHEC management. Consequently for some of our most challenging work using coupled cluster methods in the complete basis

set limit — in the jargon of the trade, coupled cluster single point energy calculations on doublet transition states of six “heavy” (that is, non-hydrogen) atoms with single and double excitations and incorporating perturbative treatment of triple excitations, or CCSD(T), and using Dunning’s correlation consistent basis sets from cc-pVDZ to the computationally-very-expensive cc-pVQZ — this had to be done at computer centres outside of Ireland which was a pity.

This hindrance remains for high-precision benchmark calculations capable of sub-kilojoule per mole accuracy — surely we are not the only research group in Ireland with these requirements?

On the positive side we have embedded in our Centre the sought-for expertise and we have increased our publication frequency in good journals and expanded the diversity of topics covered.

Thus for example work on hydroperoxides and peroxy radicals (the latter are not just of importance in low-temperature combustion but are also of biological interest), and work on alkyl furans which are of interest as ‘next-generation’ biofuels and where we identified these molecules as possessing the strongest known C—H bond strengths.

We have since moved on to an ambitious programme of computing the rates of hydrogen atom abstraction from ethers, ketones and alcohols by the atmospherically important hydroxyl radical which is also of importance in the combustion of oxygenated compounds.

Miscellaneous outputs arising from the project include the formation of a new COST Action by myself, together with a French and a Swedish colleague, which brings together some 18 EU countries in a multi-pronged attack on improving energy efficiency and pollutant reduction in the combustion of transport fuels. One of the most important work packages of this Action concerns the theoretical and experimental determination of the thermochemistry and kinetics of elementary reactions — that is, a continuation of BurnQuest on a Europe wide scale.

Reaction of O2 with NiTi alloy - dissociation of Oxygen, formation of Ti-O bonds


CHEMISTRY ATOM-SCALE STRUCTURE, DYNAMICS AND ENERGETICS OF ALKANETHIOL BILAYERS AS NANOSTRUCTURE-DIRECTING AGENTS

Greg Gannon*, Andreas Larsson*, Colm O’Dwyer** and Damien Thompson** Electronics Theory Group, Tyndall National Institute, University College Cork, Ireland and **Materials and Surface Science Institute, University of Limerick, Ireland.

Many emerging applications of nanostructures involve organic molecules chemi- or physisorbed at their surfaces. Lamellar metal-oxides have applications ranging from energy-storage, electrocatalysis, the harnessing of electrochromic and photoelectrochromic properties, application in display devices, photovoltaics, to novel energy-conversion systems, protonpump electrodes, sensors, or chemiresistive `artificial nose’ detectors. As the nanostructures approach the molecular scale, the conformation of molecules adsorbed at their surfaces will inevitably be influenced by the nanoscale geometry. The motivation for this work stems from the profound attention that non-carbonaceous nanostructures are currently receiving.

One-dimensional nanomaterials, such as nanotubes, nanowires, and nanobelts or nanoribbons have attracted considerable attention in the past decade because of their novel and useful physical properties such as semiconductivity and acting as a transparent metal in its doped state, leading to immediate applications. Besides the use of 1D nanostructures in electronics as functional components and interconnects in dense, high-speed circuits, they also have numerous applications among others in the design of ultra-small sensors, optical elements for optoelectronics, non-linear optical converters and information storage devices. The objective of the work outlined in this proposal is to use all-atom MD simulations to clarify and understand the role of organic surfactant molecules on polymorphisms of metal-oxide lamellar nanostructures.

Previous structural models for bending, and in some cases (depending on the nature of organic structure-directing agent) nanotube formation, are too simple and consistently reiterated a condensation mechanism and eventual scrolling; the reason behind layer scrolling, bending and breakage has proven elusive for these technologically important materials. The layered turbostratic structure of metal-oxide nanofibers, when preserved, could provide a new organic templated-synthesis route for the fabrication of thin-film layered nanostructures. The advantage of template-based growth methods is the ability of fabricating unidirectionally aligned and uniformly sized nanostructured arrays of a variety of materials. The knowledge of the atomic arrangements in these crystalline compounds is a key point for the understanding of the chemical and physical properties which could even range to controlled

CHEMISTRY THE INTERACTION OF ROOM TEMPERATURE IONIC LIQUIDS WITH PHOSPHOLIPID BILAYERS: A COMPUTATIONAL INVESTIGATION ICHEC CAPACITY

Damien Thompson* and Pietro Ballone*** Electronics Theory Group, Tyndall National Institute, University College Cork, Ireland ** Atomistic Simulation Group, Queen’s University of Belfast

AbstractA vast number of new chemical compounds are screened every year to assess their toxicity or to search for new pharmaceutical principles. In this respect, the first relevant aspect arguably is the interaction of the new compound with biological membranes. A case in point is represented by the interaction between biomembranes and room temperature ionic liquids (IL), which are a large class of organic ionic compounds whose melting temperature falls below 100 °C. As well as their potential for use as pharmacological agents, these compounds are the crucial ingredient of innovative approaches (Green chemistry) meant to decrease the environmental impact of industrial processes. However, recent experiments carried out on model bio-membranes have shown that, at sufficiently high concentration, several IL’s damage or even destroy phospholipid bilayers. The observation refers to concentrations far above those expected from industrial processes, but the interaction of IL’s and phospholipids remains an important issue to be explored in detail.

Biomembrane simulations are at present one of the most active subjects of computational biophysics. Computer simulations, in particular, are being used to determine the phase diagram, mechanical and dynamical properties and even the formation process of lipid bilayers. For reasons of computational complexity, much less has been done to investigate the interaction of membranes and third chemical species in solution, which, however, is a crucial aspect for many biotechnology and health applications. The proposed simulations of a phospholipid bilayer in contact with ionic liquid aqueous solutions will complement ongoing lab experiments and could pave the way to important new applications.

nanocantilever action and articifial muscles. We want to study the atom-scale mechanism behind laminar curvature in a SAM-intercalated system, by explaining the influence of synthesis parameters (specifically the organic molecule and temperature), prerequisite for the (eventual) elucidation of the mechanism of layer scrolling to form the tubular shape.

CHEMISTRYDENOX CATALYSIS ON CERIA SURFACES

Prof. Graeme WatsonTrinity College Dublin

The advent of increasingly stringent legislations on deNOx

emissions has resulted in an explosion of interest in developing new catalysts, and in adapting the present catalysts to improve their performance. Whilst much attention has been focused on the improvement of these catalysts, the actual mechanisms involved in the catalysis are still not well known. One of the current industry standards for deNO

x catalysis is ceria, CeO2. Ceria can adopt the role of a support or as a catalyst itself, and can be productive in both oxidative (CO to CO2 oxidation) and reductive (NO2 to N2) reactions, which stems from the oxygen storage capability (OSC) of the system.

In this study we have revisited the behaviour of CO and NO2 adsorption on the (110) surface of CeO2 using first principles methods. Our calculations revealed:1 (i) a novel oxygen vacancy structure on on the (110) surface, denoted a split vacancy, which is 0.32 eV more stable than the vacancy structure found previously, and which featured reduction of a surface Ce and a subsurface Ce, Figure 1, (ii) the importance of the geometry of the CO molecule when in contact with the surface was highlighted, with a tilted adsorption mode being the most energetically favourable, with the formation of a carbonate anion coupled with surface reduction, and (iii) a new bidentate adsorption mode is observed for NO

2

adsorption, accompanied by partial surface oxidation. These novel structures observed on reduction, CO adsorption, and NO2 adsorption were all energetically more stable than the previously reported structures. In all of the systems, it was found that the location of the CeIII ions did not have a strong effect on the energetics, although they were coupled to strong local distortions of the structure.

The energetics of catalytic processes at this surface show that the steps involved are exothermic and balanced, indicating that the (110) is likely to be highly active in both oxidation and reduction catalysis. However, it should be noted that the activation of NO

2 is less than previously observed on other surfaces. These results provide a detailed investigation of the interactions involved in surface reduction and the adsorption of CO and NO

2 on the (110) ceria surface.

1 D.O. Scanlon, N. M. Galea, B. J. Morgan and G. W. Watson, J Phys. Chem. C, 113, 11095-11103 (2009).

Figure1: Spin densities for (a) the simple vacancy and (b) the split vacancy on the (110) surface of ceria. Oxygen atom are red, cerium atoms are grey, and the spin isosurface at 0.05 e Å-3 is green.


CHEMISTRY

OPTIMISATION OF IONIC CONDUCTIVITY IN INTERMEDIATE SOLID OXIDE FUEL CELLS


Ceria, CeO2, is known to be an important material for a wide range of processes. CeO2 has been employed extensively in the catalysis of automotive emissions, either as a supporting material or as a catalyst itself. Furthermore CeO2 and CeO2 based materials been shown to be a promising candidates for the next generation of electrolyte materials in Intermediate-Temperature Solid Oxide Fuel Cells. Recently it has been shown that nano-scale CeO

2 displays ferromagnetic behaviour, indicating that CeO2 may have applications in spintronics and dilute magnetic semiconductors.

Many of these properties are due to the high Oxygen Storage Capacity (OSC) of CeO2. This refers to the ability of CeO2 to either release oxygen, or absorb it into its bulk, depending on the outside environment. In Kröger-Vink notation, this process is written as:

From the above equation, we see that upon the removal of oxygen form CeO

2 two excess electrons will localise on two CeIV ions, reducing them to CeIII. It has been postulated exchange interactions between these excess electrons could be the cause of the ferromagnetism that had been observed in CeO2.

To investigate these claims, we carried out DFT+U calculations on CeO2 with varying concentrations of oxygen vacancies within the bulk and oxygen vacancies upon the low index surfaces. Our results found that there was no exchange interaction between the excess electrons and that there was no preference for either ferromagnetic (FM) or anti-ferromagnetic (AFM) ordering of the spins. Hence oxygen vacancies are unlikely to be the cause of ferromagnetism in CeO

21.

Pure CeO2 is known to be a poor ionic conductor. However, the introduction of aliovaltent dopants can be used to create oxygen vacancies within the lattice which provide pathways for conduction of anions through the material. The advantage of oxygen vacancies formed in this way is that they, as opposed to those formed through the reduction of CeO

2, is they increase ionic conductivity while hindering any unwanted electronic conductivity. Therefore doped CeO2 is a suitable material for solid oxide fuel cell electrolytes.

We have carried out DFT+U calculations for CeO2 with a series of dopants to determine if these materials are readily formed and that oxygen vacancies are present. The next step will be to perform elastic band calculations to discover how these vacancies affect the ionic conductivity.

1Keating et al. 2009, J. Phys.: Condens.Matter 21 405502

Figure1: Charge density slices of an oxygen vacancy in cells with (a)12.5% and (b)1.57% concentration of vacancies. Note that charges are

localised on neighbouring Ce ions and do not interact with each other.

CHEMISTRY OPTIMIZING THE OXYGEN STORAGE CAPACITY OF CE1-XMXO2 (M = NOBLE METAL)


With the onset of strict automotive legislations around the world, the search for improved exhaust catalysts has become inexorable. CeO2 has long been regarded as one of the key materials in modern three way catalysts (TWCs), where it has been highly effective in the catalysis of automotive emissions, both as a support and as a catalyst itself. Ceria’s importance as a catalyst and as a support stems from its oxygen storage capacity (OSC), which allows it to release oxygen under reducing conditions and to store oxygen by filling oxygen vacancies under oxidizing conditions. Increasing the number and mobility of oxygen vacancies in ceria enhances the OSC, and hence enhances its catalytic activity. These stringent automotive regulations have fuelled the need for more efficient catalysts, and to this end the doping of Ceria has gained much interest.

Figure1: Noble metal dopant in a 2x2x2 96 atom CeO2 cell, showing the distortion of the oxygen sub-lattice. Cerium ions are cream, oxygen ions are red.

Recently the incorporation of the noble metals (NMs, which normally sit atop the CeO2 support in TWCs) into the lattice sites has received much experimental attention. It has recently been reported that doping CeO2 with the noble metals that normally sit atop a CeO2 support in automotive three-way catalysts, causes a remarkable increase in performance. However, the exact mechanism for this improvement is still not understood. In this study we utilized state of the art ab initio methods to examine the incorporation of noble metals into the CeO

2 lattice at a nanoscopic level, and elucidated the origin of its enhanced OSC performance.

Our results indicate that the role of the NM dopant cation in the structure is to weaken the bonding surrounding lattice oxygen in the host oxide, making their removal much easier, Figure 1. These results are currently being written up for publication, and it is expected that the insights obtained from this study can be used as a general guide for optimizing the oxygen storage capacity of materials through both aliovalent and isovalent doping. The effect of these NM dopants on catalytic selectivity is currently being assessed.


CHEMISTRY THE BEHAVIOUR OF HYDROGEN DEFECTS IN P-TYPE OXIDES


Hydrogen is ubiquitous in metal oxides, and has been shown both experimentally and theoretically to be the source of unintentional shallow donors in many wide band gap n-type oxides such as ZnO, SnO

2, and In2O3. Indeed it was shown that interstitial hydrogen and hydrogen on the perfect anion site, are the likely sources of n-type conductivity in “undoped” ZnO, and that intrinsic n-type defects could not be the source. To date, however, the behaviour of hydrogen in oxides that do not possess large optically transparent bandgaps, or that do not exhibit n-type conductivity has not been comprehensively elucidated.

The behaviour of hydrogen in p-type oxides has received very little experimental attention, and the few experiments that have been carried out have left many questions unanswered. A recent muonics study has indicated that hydrogen behaves as a deep donor in p-type Ag

2O (~0.25 eV) but that hydrogen

Figure1:. Hydrogen (green sphere) sitting on a perfect oxygen lattice site in a 2x2x2 48 atom cell. Copper ions are blue, oxygen anions are red.

in Cu2O has quasi-atomic character and causes an electrically active level ~1 eV below the conduction band minimum (CBM). However, this study could not identify if the transition level in Cu

2O represented hole or electron ionization, and could not infer any site identification due to motional narrowing. Previous predictions from some “generalized” theory papers indicates that H in Cu

2O should behave as a n-type donor, although these predictions have been recently called into question. In light of these unanswered questions, we have examined the electronic behaviour of hydrogen impurities in natively p-type oxide Cu

2O using the state of the art ab initio screened hydrid density functional HSE06, as implemented in the VASP Code.

These detailed calculations (which are up to a factor of 50 times more expensive than a GGA/GGA +U study), are currently submitted for publication. Our results have revealed for the first time the site with “quasi-atomic” character that could not be identified in the muonics study, and have elucidated the nature of the mid-band-gap ionization. We have also investigated for the first time the effect of hydrogen on the electrical properties of Cu

2O, and propose a guide to experimentalists to increase the conductivity of Cu2O based devices.

COMPUTINGOPTIMISING HYBRID MACHINE TRANSLATION

Patrik Lambert, Yanjun Ma, Sergio Penkale and Ankit SrivastavaMachine Translation Group, Centre for Next Generation Localisation, School of Computing, Dublin City University

ABSTRACT

In the current approach to machine translation, translational correspondences between source and target language words, called word alignment, are first induced automatically. In a second and separated stage, bilingual phrases are extracted from these correspondences and relative probabilities are calculated for each bilingual phrase, which constitutes the probabilistic dictionary used for translation. In this project, we want to bring the word alignment stage into relation with the end product (the translations) by optimising our alignment system according to machine translation metrics. We also want to optimise the way of combining various information sources to extract the phrases used to build the probabilistic dictionary.


EARTH SCIENCE COUPLED OCEAN-ATMOSPHERE CLIMATE MODELLING

Dr. Michael HartnettNational University of Ireland, Galway

The project is funded by HEA under Cycle 4 of the Programme for Research in Third Level Institutions (PRTLI-4) and by EPA under STRIVE Programme. It is part of an initiative to develop an infrastructure-system to enable true national collaboration and capacity development in Environment & Climate Change assessment in the context of Ireland. This research is being undertaken within the Environmental Change Institute (ECI), NUI Galway.

Currently there is no regional (or global) scale coupled ocean-atmosphere climate model that has the capability to treat air-sea exchange of particulate matter/aerosols and ozone in terms of their linkage to marine biogeochemical cycles driven by the marine biota. State-of-the-art ocean and atmosphere modelling systems were acquired in order to develop capabilities for advanced regional climate change modelling. In terms of marine aerosol production, the REMOTE regional climate model has, amongst other capabilities, the most advanced treatment of aerosol processes. As regards the ocean model, the Max Plank Ocean Model (MPI_OM) is one of the few models worldwide that has been developed for climate change modelling and includes biogeochemical processes modelling, widely known as the HAMOCC model. OASIS3 coupler is currently used by approximately 15 climate modelling groups worldwide and has also been applied in this study. It acts as a separate mono process executable, which main function is to interpolate the coupling fields exchanged between the component models and communicate this information between the modelling system components.

The ocean model grid has been constructed in such a way as to allow achieving high resolution over the Irish coastal waters. The model encompasses the entire globe and uses the bipolar orthogonal spherical coordinate system. The computational mesh consists of 241 meridional lines and 164 zonal lines with one pole located over Western Europe and second pole over Canada. The model has been spun-up using the atmospheric forcing data provided by NOAA from its NCEP/NCAR Reanalysis Project in order to achieve ‘initital conditions’ for year 2006. The data consisted of 8 parameters and included total cloud cover, precipitation, solar radiation, dew point temperature, 2m air temperature, 10m wind speed and zonal and meridional wind stress components. Year 2006 will be used as a ‘control year’ for future coupled climate change simulations. Recently, the atmospheric data from the Max Planck Insittute’s A1B climate scenario simulation has been acquired. It is currently used to carry out MPIOM/HAMOCC simulations to spin up the model and to produce restart files for the model simulations for 2020, 2030, 2050 and 2100.

The ocean model reproduces the sea surface temperatures in Irish coastal waters with high degree of accuracy. The model

predictions correlate very well with the temperatures measured at M1-M6 marine buoys located off the eastern, southern and western Irish coasts and also off the continental shelf.

Predicted phytoplankton distributions in the world’s oceans are qualitatively good and yield similar results to other modelling studies. Higher chlorophyll concentrations are simulated in each hemisphere during the productive seasons. Low chlorophyll-a concentrations observed in the subtropical regions are reproduced by the model. As regards the North Atlantic waters, the concentrations increase rapidly in the April to June interval, which is the correct timing for the spring phytoplankton blooms in the areas north of 45ºN. Rapid depletion of nutrients (nitrates and orthophophates) that takes place in the North Atlantic waters at the time of spring bloom is also well predicted by the model.

The model reproduces both global distributions of sea surface pCO2 and sea-air pCO2 differences well including the climatic changes in global CO2 atmospheric mixing ratio. Detailed comparisons of sea-air pCO2 differences were carried out for four stations from which measurements are available; two located in the North Atlantic, one around Iceland and one in the area of the North Atlantic Bloom Experiment (NABE) close to the shores of North America, and two located in the North Pacific.

The global ocean model has also been used to provide boundary conditions for a fine resolution hydrodynamic model of the Irish Sea in order to investigate the climatic change to hydrography of the region with particular focus on the western Irish Sea, which is subject to strong thermal stratification in the summer each year.

EARTH SCIENCE

JOINT INVERSION OF MAGNETOTELLURIC AND SURFACE WAVE MEASUREMENTS IN ANANISOTROPIC EARTH: APPLICATION TO REAL DATASETS FROM CENTRAL GERMANY AND NORTHERN CANADA

Estelle RouxDublin Institute for Advanced Studies

Joint inversion of different kinds of geophysical datasets has the potential to improve the model resolution. Joint inversions have been commonly undertaken with datasets sensitive to the same physical parameter. This problem is more challenging when datasets are sensitive to different physical parameters.

EARTH SCIENCE PRESENT AND FUTURE STATISTICS OF EXTREME EUROPEAN STORMS

Rodrigo CaballeroUniversity College Dublin

ABSTRACT

In December 1999, three extreme storms devastated northern Europe, causing unprecedented damage and loss of life. How do such storms arise? What are the chances of such a storm hitting Ireland on any given winter? Will such storms occur more often in future? We will study these questions by generating a very large number of ‘virtual’ storms using a numerical atmosphere model, allowing us to build a robust and realistic picture of typical storm features and controlling conditions.

We will generate a large (1000 year) ensemble of highresolution simulations of the North Atlantic storm track spanning present and future climate conditions, providing an unprecedented combination of statistical accuracy and simulation realism.

Our approach is based on a joint inversion using a Genetic Algorithm (GA) for a one-dimensional isotropic structure using long-period MT data and teleseismic receiver functions (Moorkamp et al., 2007, 2010). GA are a class of stochastic optimisation algorithms particularly well suited for the solution of multi-objective optimisation problems. In comparison with linearized methods, they have the potential to escape local minima. But, as it is a stochastic method, there is no guarantee that it does. Repeated runs of the GA can avoid this problem. The disadvantage then is that it requires much higher number of function evaluations. This problem has been overcome thanks to the facilities available on the ICHEC cluster.

We used our own implementation of NSGA II (Deb et al., 2002) to jointly invert MT and SW dispersion curves for a one-dimensional anisotropic structure with both synthetic and real datasets.

When applying this anisotropic joint inversion to real datasets, the main difficulty is to find a suitable region where good quality MT and seismic datasets coincide. For this reason, two regions have been selected. The first one is Central Germany where previous studies include comparisons between seismic and electrical structures but no joint inversion. The second area is the Slave Craton in Northern Canada. The Slave Craton is one of the oldest place on Earth and a highly debated subject where a lot of seismic anisotropy work has been carried out but no MT anisotropy so far.

We jointly invert MT data from one selected site in Central Germany (more details in Leibecker et al., 2001) together with SW data (GRSN network) sampling the same area (see map in fig. 1).

Several runs of GA have been performed using a population size (i.e., the number of models in each iteration) of 900 members for 200 iterations.

We thus obtained a one-dimensional anisotropic model for Central Germany which fits equally well MT and seismic datasets.

We have constrained the depth of the LAB at 83 km and resolved two main anisotropic layers, one in the lower crust with a NE / SW anisotropic direction and another strongly anisotropic layer in the asthenospheric upper mantle. This layer is characterized by a nearly East – West most conductive

/ seismic fast axis direction. Such a result agrees well with previous studies made in this area (Gatzemeier et al., 2005; Brechner et al., 1998) but can-not be fully explained by the absolute plate motion as it is often the case when the most conductive direction and the seismic fast propagation coincide.

Figure 1: Joint electrical (A) and seismic (B)anisotropic structure in Central Germany.


Machine Compiler # CPUs

Optimization Flags

Time (hr)

Walton Pathscale 24 -O3 -ipa-OPT: Ofast -fno-math-errno

3.75

Stokes Intel 24 -O2 -xT –ip 1.2

Stokes Intel 24 -O3 -xT -ip -no-prec-div

1.15



Stokes Inte 96 -O2 -xT –ip 0.4

EARTH SCIENCE CLIMATE CHANGE AND THE FUTURE WIND ENERGY RESOURCE OF IRELAND

Prof. Peter LynchUniversity College Dublin

INTRODUCTIONAt the UCD Meteorology & Climate Centre, we are using the method of Regional Climate Modelling to examine the potential impact of global climate change on the wind energy resource of Ireland.

The impact of greenhouse gases on climate change can be simulated using Global Climate Models (GCMs). However, long climate simulations using coupled atmosphere-ocean general circulation models are currently feasible only with horizontal resolutions of 50 km or greater. This is inadequate for the simulation of the detail and pattern of climate change and its effects on the wind resource of Ireland. The Regional Climate Model (RCM) dynamically downscales the coarse information provided by the global models and provides high resolution information, on a sub-domain covering Ireland.

EXPERIMENT SETUPThe Irish climate was simulated using the CLM-Community’s COSMO-CLM Model (CLM) at 7km resolution. The model domain has 90x94 grid points and in the vertical there are 32 unequally spaced levels. The model was integrated with a time step of 40 seconds. The wind fields are output at one hour intervals. The COSMO-CLM 7km simulation was driven at the lateral boundaries by CLM consortial simulation data at 18km resolution. A preliminary experiment was carried out to determine the optimal number of CPUs and to establish the compiler optimization flags. Table 1 presents the results. It was decided to run the code on 48 processors and so as to maintain accuracy, to use the safe optimization flags “-O2 -xT –ip”.

RESULTSThe CLM model was validated by performing a 20-year simulation (1981-2000) of the past Irish climate, driven at the lateral boundaries by ECMWF ERA-40 data, and comparing the output to observations. Projections for the future Irish climate were generated by downscaling the Max Planck Institute’s ECHAM GCM data using CLM. Simulations were run for the 40-year future period (2021-2060) and a control period (1961-2000). The future period was simulated using the A1B green-house gas emission scenario. Future projections show a marked increase in the amplitude of the annual cycle in wind strength with about 10% more energy available during winter and 10% less during summer. The results are presented in Figure 1. The projected changes for summer and winter were found to be statistically significant over most of Ireland.

Figure1:. The CLM ECHAM5 A1B projected percentage change in the 60m mean wind power density for (a) winter and (b) summer

Table 1. The Speed-Up of CLM with increasing CPUs (for a one-month simulation)

EARTH SCIENCE SEISMIC SOURCE MODELLING AND PROPAGATION IN COMPLEX 3D EARTH MODELS

Dr Gareth Shane O’BrienSchool of Geological Sciences, University College Dublin

Seismic wave generation, by earthquakes or artificial sources, and their propagation through the Earth is a key tool in geophysics and one of the best available methods for studying physical processes in the Earth. Interpreting the observed seismic waveforms is a difficult task and exacting information about the source and medium is non-trivial due to the complex structure of the Earth and the unknown nature of the source mechanisms. Therefore, to interpret the observations it is essential to calculate, in detail, the propagation of the waves through modelled structures and then compare the resulting synthetic seismograms with real observational data. Numerical modelling of synthetic seismograms in 3D complex Earth models is only feasible using high-end computational techniques and resources. The focus of this project is to develop and apply such high-end computational wave propagation algorithms to two main areas; i) CO

2 sequestration, and ii) volcano seismology.

Capture and geological storage of CO2 provides a way to reduce CO2 emissions into the atmosphere, by capturing it from major stationary sources (such as fossil fuel burning power stations), transporting it, usually by pipeline, and injecting it into suitable deep rock formations.

A numerical simulation of the seismic wavefield propagating across the surface of Ubinas volcano, Peru. The wavefield is generated by an explosive event at a depth of 1 km directly under the summit crater. The blue colour indicates upward ground displacement and the red colour shows downward ground displacement. The circular wave-front is the direct wavefield expanding from the epicentre whilst the complex wavefield behind this front is created by scattering of the wavefield from the topography and subsurface structures.

The fate of the injected CO2 is of critical importance if the sequestration is to be successful in mitigating the atmospheric pollution. Repeat seismic studies can be used to image the evolution of the CO

2 plume in the subsurface and numerical simulations aid the interpretation of these seismic experiments. These numerical simulations are also used to forecast the future behaviour of the CO

2 plume.

Volcanoes can produce sounds as the magma (lava), gases and water flow inside the volcano.

These sounds are inaudible but can be recorded as seismic signals. Different types of signals can indicate different fluid motion. By understanding how these signals are created we can better understand the volatility of the volcano. The high degree of variability in the generation of volcano seismicity presents a difficult challenge in modelling seismicity in volcanoes. Therefore, using seismic signals as a precursor to an eruption is difficult and only through a better understanding of the dynamics can we increase our forecast capabilities. Even though the role of fluid dynamics in generating volcano seismicity has been well established the actual physical processes are not yet fully determined. By a combination of computational fluid dynamics and computational seismology using high-end computing we can study the role of fluids in generating the observed signals recorded on active volcanoes.

(a)

(b)


ENGINEERING EVALUATION OF NOVEL FEATURES

Marco GrimaldiUniversity College Dublin

In the field of Speaker Identification it is a well known fact that at the present time there is no scientific process that enables one to uniquely characterize a person’s voice or to identify with absolute certainty an individual from his or her voice3. In fact, a number of researchers have evaluated how different approaches in speech modeling, signal analysis and machine learning impact the ability of generalize the concept of speaker identity (e.g. 4, 5, 6). Despite the enormous effort of the research community, automatic speaker recognition it is still an open problem, only solved in very particular scenarios. The scientific community working on forensic speaker identification 3, 9, 10 stresses the fact that the problem may be solved in “ideal world” scenarios, but that the state of the art in Automatic Speaker Identification can not be applied straightforwardly in real world cases.

The main objective of the project was to test a novel parameterization of speech that is based on the AM-FM 7 representation of the speech signal and to assess the utility of these features in the context of speaker identification. In order to explore the extent to which different instantaneous frequencies due to the presence of formants and harmonics in the speech signal may predict a speaker’s identity, this work evaluated three different decompositions of the speech signal within the same AM-FM framework: a first setup has been used previously for formant tracking 7; a second setup has been designed to enhance familiar resonances below

4000 Hz, and a third setup has been designed to approximate the bandwidth scaling of the filters conventionally used in the extraction of MFCCs. From each of the proposed setups, parameters have been extracted and used in a closed-set text-independent speaker identification task.

The speech samples used in this project are extracted from the CHAINS corpus 1, 2. The CHAINS corpus is a novel speech corpus designed to facilitate research into the characterization of individual speakers. It offers speech samples obtained in two different recording sessions with the unique possibility of studying the effect of non-modal voice (whispering) in speaker identification. The first recording session used a very high quality recording environment and apparatus, while the second recording session is more typical of data recorded in a quiet office using a near-field microphone. Across the two sessions, each speaker provides recordings in six qualitatively different speaking styles.

The computing facility provided by ICHEC have been used to guarantee the reproducibility of our results: varying the parameters of both the identification system and the AM-FM parameterization of the input signal, we have been able to provide recognition curves in a variety of experimental setups. By repeating each experiment several times, we have averaged out errors introduced by the initialization procedures common to any machine learning technique and have been able to straightforwardly compare results obtained using more standard approaches to speech parameterization, such as MFCC and RASTA features.

Figure 3: Accuracy of the identification system varying speech encoding.Figure 1: Pyknogram of a sample speech file, the new signal representation provided by the AM-FM approach.

Figure 2: Accuracy of the identification system varying the AM-FM setup and the number of parameters extracted

The quality of the work done to date and the novelty of the approach adopted in speech parameterization has been recognized by international peer reviewers and IEEE Transaction on Audio Speech and Language Processing have accepted our work for publication.

Preliminary details of the publication are as follows: M .Grimaldi, F. Cummins. Speaker Identification Using Instantaneous Frequencies, to appear in IEEE Transaction on Audio Speech and Language Processing, 2008.

1 http://chains.ucd.ie/index.php2 F. Cummins, M. Grimaldi, T. Leonard, and J. Simko. The CHAINS corpus: CHAracterizing INdividual Speakers. In Proceedings of SPECOM’06, pages 431--435, St Petersburg, RU, June 2006.3 J.-F. Bonastre, F. Bimbot, L.-J. Boë, J. Campbell, D. Reynolds, I. Magrin-Chagnolleau. Person authentication by voice : a need for caution. In: Proceedings Eurospeech, Genève, Suisse, 2003.4 Plumpe, M.D.; Quatieri, T.F.; Reynolds, D.A. Modeling of the glottal flow derivative waveform with application to speaker identification. Speech and Audio Processing, IEEE Transactions on, pages: 569-586, Volume: 7, Issue: 5, Sep 1999.5 A. Reynolds. Experimental evaluation of features for robust speaker identification. IEEE Trans. Speech Audio Processing, vol. 2, pp. 639-643, Oct. 1994.6 Bou-Ghazale, S.E.; Hansen, J.H.L. A comparative study of traditional and newly proposed features for recognition of speech under stress. Speech and Audio Processing, IEEE Transactions on, pages: 429-442, Volume: 8, Issue: 4, Jul 2000.7 A. Potamianos and P. Maragos. Speech formant frequency and band-width tracking using multiband energy demodulation. Journal of the Acoustic Society of America, 99:3795–3806, 1996.


ENGINEERING A COMPUTATIONAL ANALYSIS OF THE MICROSCALE FORCES THAT DRIVE CELL-DERIVED TISSUE FORMATION: A TISSUE ENGINEERING SOLUTION

Dr Adam Stops, Dr Noel Harrison, Prof Peter McHughNational university of Ireland, Galway

Tissue engineering attempts to grow tissue from individual cells in a laboratory setting. As with many building processes a scaffold is used to provide a temporary structure to organise cell function and encourage tissue growth. An artificial foam-like biomaterial provides this host environment; yet, the scaffold does not just assist in delivering a geometrical structure to organise tissue shape, but it also interacts with the seeded cells in the form of biophysical signalling (biophysical refers to the net product of mechanical stretch and fluid flow). Given that cells are biophysically sensitive and that the type of tissue synthesised by the cells is hugely dependent upon the level of biophysical stimuli, then the biophysical environment within these scaffold is of paramount interest to the tissue engineer.

Typically a bioreactor is used to create the biophysical environment for cell culture. For example, the mesenchymal cells (MSCs) are seeded within the scaffold so that the scaffold-cell construct can then be placed within the bioreactor. Subsequently, the bioreactor is then used to impose a mechanical stretch onto the construct whilst simultaneously enforcing fluid to flow through. This methodology results in a complex scenario whereby mechanical deformations and fluid flows both influence cell function. At the moment, very little is known about the optimal magnitudes of scaffold stretch or fluid velocities to grow tissue. Thus , the work undertaken in this investigation used a computational analysis to determine whether a laboratory setting can impart an optimal level of biophysical stimuli onto the cells seeded within a scaffold, furthermore, this work attempted to link particular levels of mechanical stimuli with specific typos of tissue, i.e. bone, cartilage or fibrous tissue.

The microstructural architecture of a tissue engineering scaffold was collated by using µ-Computer Tomography (µCT) images. From the resulting 3D meshes, two computational methods were employed to simulate the biophysical environment of a cell-seeded scaffold within a bioreactor. Finite Element (FE) analysis was used to model the mechanical deformations, and Computational Fluid Dynamics (CFD) was implemented to reproduce the fluid flow conditions. While the FE software utilises a parallel OpenMP f90 code, viz. The FEEBE suite was written in F90 by Dr Noel Harrison and Dr Denis O’Mahoney (National University of Ireland, Galway, Ireland), the CFD software employed a parallel platform supplied by ANSYS CFX (ANSYS Inc, USA). ICHEC has facilitated the development and application of the custom designed FEEBE software suite through a number of ICHEC Class B and C projects, and also was instrumental in optimising the use of the commercially available software CFX for this specific application on the ICHEC hardware.

Using the computed deformations and fluid velocities, cell functions were predicted for a range of bioreactor settings.

Figure1: The predicted differentiation patterns within the modelled volume are shown here for (a) an FBT dominated volume (where 73.9% of the cells were FBT), which resulted from 5% scaffold strain and 10 µm s!1 inlet velocity, (b) a volume with mostly CHD cells (56.7% CHD) which arose from a 5% scaffold strain and 1 µm s!1 inlet velocity and (c) an OST cell majority (84.9% OST) which was a product of 1% scaffold strain and 1 µm s!1 inlet velocity.

Subsequently, given that a biophysical-sensitive selection decides the differentiated form of the mesenchymal cell , i.e. an osteoblast (OST), fibroblast (FBT) or chondrocyte (CHD), and the differentiated form of the cell dictates the type of tissue (generally, osteoblasts produce bone, fibroblasts produce skin and chondrocytes produce cartilage), tissue formation predictions were offered so that specific bioreactor settings could be linked with particular tissue types.

Interestingly, and optimal combination of mechanical deformation and fluid velocities, as provided by a specific bioreactor setting, was found for osteoblast and fibroblast cells (fig 1). However, chondrocyte cells did not favour one particular combination of biophysical stimuli. Interestingly, this could lead to a development of the current theories used to grow cartilage in a laboratory. Furthermore the results suggested that tissue growth is highly sensitive to bioreactor settings. Importantly, this biophysical environment is the driving force in cell function and so the ability to engineer the desired tissue is of great interest to the tissue engineer. It is hoped that the findings of this work will help experimental researchers to further develop the current tissue engineering procedures.

The ICHEC facility has permitted the combined mechanics and fluid dynamics simulation to surpass previous computational models by being able to represent significant geometries through substantial mesh sizes (the FE model consisted of ~ 7 million elements, while the CFD mesh used ~45million elements. Consequently, a typical FE simulation required 5000-8000 hours CPU time and 12GB of RAM, and were run on the ICHEC systems Stokes and Hamilton. This work has been published in the Journal of Biomechanics, 43 (2010) 618-626.

ENGINEERING MODELLING THE ACTIVE EVOLUTION OF FOCAL ADHESIONS AND STRESS FIBRES IN CELLS AS A RESULT OF MECHANICAL STIMULI

Enda Dowling, William Ronan, Patrick McGarryNational University of Ireland, Galway

INTRODUCTIONActin filaments in chondrocytes dynamically adapt to mechanical loading through Rho-kinase mediated signalling mechanisms. However, in vitro investigations of the response of cells to mechanical stimuli provide limited insight into these mechanisms due to the fact that cellular structure and function actively evolves in response to physical stimuli. Furthermore, a passive elastic computational model of cell behaviour cannot be used to accurately compute stresses in a cell or the evolution of the actin cytoskeleton.

In this study, an active model that describes the assembly of the actin cytoskeleton in response to cell signalling, and the dissociation of the actin cytoskeleton in response to a reduction of intracellular tension is implemented. In vitro experiments of single chondrocytes undergoing shear deformation are simulated. Computations are compared to a passive hyperelastic model of the cell cytoplasm.

METHODSThe actin cytoskeleton is developed via the phosphorylation of myosin and polymerization of actin filaments which combine to form actin-myosin (AM) contractile units. In the present study the formation and dissociation of AM contractile units is governed by a first order kinetic equation. AM contractile units in response to a signal in the cell cytoplasm. Dissociation of contractile units in response to a reduction in tension to a value lower than the isometric tension. AM contractile units are free to assemble in any direction at all integration points in the cytoplasm, with the stress state at that point determining the dissociation of contractile units in any given direction. The AM contractility units are implemented as a user-defined material in ABAQUS. The 3D model of a single chondrocyte is based on in vitro cell geometries and testing parameters, with the cell adhered to a glass slide and a tungsten probe used to apply a shear load. A passive hyperelastic nucleus is included in the model with properties based on published studies.

RESULTSAs can be seen in Fig. 1(a), following 7.5µm of probe indentation a reduction of tension at the front of the cell leads to a localized dissociation of AM contractile units. This gradual dissociation of AM contractile units at the base of the cell is illustrated in Fig. 1(b-d). Comparison of probe forces for active cell simulations and passive cell simulations reveal that the active cell provides significantly more resistance to shear deformation. It should also be noted that the maximum stress computed in the cell nucleus is 24% greater in the case of the passive model.

Figure1: Contour plots of the density of AM contractile units during shear.

DISCUSSIONResults show a strong correlation with experimentally observed behaviour in single chondrocytes experiencing shear. Confocal microscopy reveals an absence of actin at the front of chondrocytes following shearing. Additionally, the disruption of AM contractile units using cytochalasin-D leads to a significant reduction in probe force required to shear adhered chondrocytes. Clearly the active contraction and remodelling of the actin cytoskeleton plays a critical role in the deformation of chondrocytes. Recent in vitro studies have demonstrated the important role of nuclear deformation in gene expression. We have demonstrated that the inclusion of active cytoskeletal evolution has a significant effect on computed nuclear stresses, and consequently on the development of a predictive framework for cell mechanotransduction.


ENGINEERING DEVELOPMENT OF DISCRETE FINITE ELEMENT CELL MODEL

K.M. Coghlan & P.E. McHughNational University of Ireland Galway

Adherent cells exert tractions on their underlying substrate. These forces act as mechanical signals and play an important role in how the cell alters its structure and function, including cell proliferation and differentiation. Wang et al1 experimentally measured the 2D tractions exerted by smooth muscle cells on gels using Fourier transform traction cytometry (FTTC). The aim of this project is to use the Finite Element (FE) method to model a contracting 3D cell on a 3D gel substrate and to predict the resulting displacement field and compare the results to the 2D field observed by Wang et al.

The FE gel substrate and cell were modelled as hyperelastic materials. The height of the cell at the centre was 6 µm and decreased linearly over a length of 80 µm to a height of 0.5 µm at the edge of the cell, as described by Wang et al. To initiate cell contraction, the cell cytoplasm was given a thermal coefficient of expansion, and a negative temperature field was prescribed. was varied until the cell contracted enough to displace the gel a maximum of 5µm at the cell edge.The cell on gel traction model was accurately able to predict the resulting displacement field of the gel due to cell contraction. In addition, the model was also able to capture the 3D displacements that were not observed by the FTTC

Figure1: Experimental compared to FE traction fields

in vitro method. The work of Wang et al suggest out of plane displacements of less than 0.2 µm. The FE displacement field perpendicular to the gel surface revealed a larger than expected out of plane gel activity, with displacements of up to 0.7 µm. The FE model predicted tractions much larger than those calculated experimentally. Figure 1 shows the in vitro traction field as computed from the displacement field compared to the FE traction field.

Wang et al reported maximum tractions of 900 Pa at the periphery of the cell. The FE results show tractions 5x higher at 4500Pa. FE tractions higher than 900 Pa are greyed out. The higher tractions are due to high localized strains at the cell periphery as a result of the soft gel undergoing large deformations. The FTTC method was unable to capture the high localized strains possibly due to the poor resolution of the traction microscopy method. The beads were on average 4µm apart, whereas the high strains occurred in regions of less than 2µm as predicted by the FE model. The FE model presented here eliminates this problem.

The authors wish to acknowledge ICHEC for the provision of computational facilities and support.

1 N Wang et al, Am J Physiol Cell Physiol, 282: C606-C616, 2002

LIFE SCIENCESEARCHING FOR EVIDENCE OF INTERKINGDOM HORIZONTAL GENE TRANSFER BETWEEN BACTERIA AND CTG SPECIES

Dr. David FitzpatrickNational University of Ireland, Maynooth

To date very few incidences of interdomain gene transfer into fungi have been identified. We used the emerging genome sequences of Candida albicans WO-1, Candida tropicalis, Candida parapsilosis, Clavispora lusitaniae, Pichia guilliermondii, and Lodderomyces elongisporus to identify recent interdomain horizontal gene transfer (HGT) events. We refer to these as CTG species because they translate the CTG codon as serine rather than leucine, and share a recent common ancestor.

Phylogenetic and syntenic information infer that two C. parapsilosis genes originate from bacterial sources (Figure 1). One encodes a putative proline racemase (PR). Phylogenetic analysis also infers that there were independent transfers of bacterial PR enzymes into members of the Pezizomycotina, and protists. The second HGT gene in C. parapsilosis belongs to the phenazine F (PhzF) superfamily. Most CTG species also contain a fungal PhzF homolog. Our phylogeny suggests that the CTG homolog originated from an ancient horizontal gene tranfer event, from a member of the proteobacteria.

An analysis of synteny suggests that C. parapsilosis has lost the endogenous fungal form of PhzF, and subsequently reacquired it from a proteobacterial source. There is evidence that Schizosaccharomyces pombe and Basidiomycotina also obtained a PhzF homolog through HGT. Our search revealed two instances of well-supported HGT from bacteria into the CTG clade, both specific to C. parapsilosis. Therefore, while recent interkingdom gene transfer has taken place in the CTG lineage, its occurrence is rare.

We have also investigated the evolutionary history of phenazines in bacterial species. Phenazines are secondary metabolites with broad-spectrum antibiotic activity against bacteria, fungi, and eukaryotes. In pseudomonad species, a conserved seven-gene phenazine operon (phzABCDEFG) is required for the conversion of chorismic acid to the broad-spectrum antibiotic phenazine-1-carboxylate (Figure 2). Previous analyses of genes involved in phenazine production from nonpseudomonad species uncovered a high degree of sequence similarity to pseudomonad homologues. As well as eluciadating the evolutionary history of genes involved in the production of phenazines, we also wished to determine if the phenazine operon has been transferred through horizontal gene transfer. Analyses of GC content, codon usage patterns, frequency of 3:1 dinucleotides, sequence similarities, and phylogenetic reconstructions were undertaken to map the evolutionary history of phenazine genes from multiple bacterial species. Patchy phyletic distribution, high sequence similarities, and phylogenetic evidence infer that pseudomonad, Streptomyces cinnamonensis, Pantoea agglomerans, Burkholderia cepacia, Pectobacterium atrosepticum, Brevibacterium linens, and Mycobacterium abscessus species all contain a phenazine operon which has most likely been transferred among these species through horizontal gene transfer. The acquisition of an antibiotic-associated operon is significant, as it may increase the relative fitness of the recipient species.

Figure 1: Proline racemase maximum likelihood phylogeny with active site alignment. Bootstrap resampling (100 iterations) was undertaken and percentages are displayed. Fungal branches are shown in green. An alignment around the active site is also displayed.

Figure 1: Comparison of phenazine loci from multiple bacterial species with P. aeruginosa PAO1. Gene names are shown within gene boxes. Homologs are also colored coded.


LIFE SCIENCE THE RETINOL BINDING PROTEIN SYSTEM AND TYPE 2 DIABETES

Dr. Gemma Kinsella, Prof. J. B. C. Findlay.National University of Ireland, Maynooth

Two recent scientific discoveries provide a way forward in preventing the seemingly unstoppable avalanche of disability due to type 2 diabetes. The first of these discoveries is the observation that elevations in serum retinol binding protein (RBP) may attenuate the response of the cell to insulin, thus increasing resistance and presaging type 2 diabetes.1 The second is the identification of the plasma membrane receptor for RBP through which its effects are likely to be mediated.2 RBP is responsible for the transport and delivery of retinol (vitamin A) around the body. It is synthesised and released from the liver as a complex with a second protein transthyretin(TTR), a process which is under the control of dietary retinol. RBP then interacts with a receptor in the plasma membrane of virtually all cells, releasing its bound retinol to the transport part of the receptor activity.3 RBP then looses its affinity for both the receptor and TTR and being small is excreted via the kidney. In obesity or in individuals with a genetic pre-disposition to type 2 diabetes, RBP levels in serum rise due to secretion from adipose tissue and loss of secretory control. The connection with insulin action is a very novel one, with evidence now supporting a cross-talk between RBP and insulin, since higher levels of the former are seen to dampen down the intracellular response to insulin. Thus, the hypothesis to be exploited in this project is that agents which reduce serum RBP levels will prevent the genesis of insulin resistance, and consequently that of type 2 diabetes and its cardiovascular complications.

The project aims to use computational drug discovery techniques to discover small molecules that may prevent the elevation in serum RBP and its effects. Two novel drug design phases were developed to determine novel RBP binders and RBP:TTR disrupters. The rationally designed compounds were evaluated experimentally through RBP fluorescence measurements to determine binding to RBP and in RBP-TTR release assays, and surface plasma resonance (SPR) to determine disruption of the RBP:TTR interaction.

From Phase 1, five of the 21 ordered compounds merit further investigation. One is an analogue of the previously identified RTB-70, while RTB-84 & RTB-86 represent novel and different scaffolds. Additionally, RTB-102 & RTB-103 (which are analogues) disrupted the RBP:TTR interaction. From Phase 2, all 10 compounds were deemed of interest after the experimental validation, but in particular RTB-117.

Across 10ns molecular dynamics (MD) simulations the binding mode of A1220 in the 3FMZ crystal structure exhibits three HBs involving the carboxylic group and Gln98, Tyr90 and Arg121 (Figure 1). We examined the best docked binding poses of our hit compounds after refinement with the ligX protocol in MOE and 10ns MD simulations. A steady interaction involving Arg121 was the most prevalent with all our identified hit compounds with subtle changes in the loop regions of the protein.

Furthermore, RTB-70 was validated in mouse work and is the basis for patenting “first in class” drugs. For this, we need to optimise the “hits” to “leads” and establish robust structure/activity relationships. In silico ADMET profiling will give an indication of any potential adverse effects.

The authors acknowledge IRCSET postdoctoral funding.

Publication in Preparation –“Fenretinide derivatives act as disrupters of Retinol Binding Protein (RBP4) interactions with TTR and STRA6”

1 Yang, Q.; Graham, T.E.; Mody, N.; Preitner, F.; Peroni, O.D.; Zabolotny, J.M., Koyani, K.; Quadro, L., Kahn, B.B., Nature, 2005, 436, 356-362.2 Kawaguchi, P.; Yu, J.; Honda, J., Hu, J., Whitelegge, J.; Ping, P., Wiita, P.; Bok, D., Sun, H., Science, 2007, 315, 820-825.3 Burke, B.J., Redondo, C., Redl, B., Findlay, J.B.C., 20505, In Lipocalins (Aherstrom), Landes Biosciences, Austin, Texas, USA.4 PyMOL (De Lano Scientific, USA).

Figure 1: Co-crystal structure of human RBP4 with A1120 (on left). A zoomed-in image of the binding pocket with A1120 is shown on the right. Images were produced using Pymol.4

LIFE SCIENCE HIGH RESOLUTION MODELLING OF BONE LOSS IN VERTEBRAL TRABECULAR BONE

Prof. Peter McHugh, Pat Mc DonnellNational University of Ireland, Galway

Trabecular bone is severely affected by osteoporosis, with up to 50% reduction in bone mass and damaging changes in the architecture. This results in greatly increased fracture risk, particular at the anatomic locations of the distal radius, the femoral head and the vertebrae. The objectives of this project are to investigate the mechanisms and effects of bone loss in vertebral trabecular bone. To achieve this goal, a custom finite element analysis package (FEEBE), which has been specifically developed to run on parallel processors, has been used on Hamilton to perform multiple analyses on high resolution models of human vertebral trabecular bone. The models have been generated using the FEEBE package, by discretising micro-computed tomography (µCT) scans of real bone into brick-shaped voxel elements. An element removal algorithm (ERA) has been implemented and run on Hamilton to virtually remove surface elements from the trabecular architecture according to the levels of strain and microdamage that are experienced under physiological compression loads. The models that have been analysed to date comprise 2 million elements. Each analysis requires approximately 48 hours cpu time for completion, using 8 processors. Analyses have been completed on models that have been simulated with progressively increasing adaptive and micro-damage based bone loss. The results of these initial analyses indicate that this element removal and finite elelment analysis process is capable of producing realistic osteoporotic trabecular architectures from a baseline healthy architecture (see Fig. 1). Analyses are currently ongoing and it is planned that by mid-September, sufficient data will be available for publication of a paper in an international journal.

a) Original trabecular architecture

b) Simulated bone loss

Figure 1: Slice removed from model of trabecular bone showing a) original architecture, and b) osteoporotic architecture after a number of iterations of simulated bone loss.


LIFE SCIENCE

THE TREE OF ONE HUNDRED PERCENT

Dr. James McInerneyNational University of Ireland, Maynooth

In 2006 Ciccarelli and coworkers attempted to infer a tree of life from a dataset comprised of 31 genes involved in translation. While this is recognized as an important contribution to our understanding of life on the planet, there have been some criticisms of the approach that involves using such a small sample of genes. In particular, it has been pointed out that on average a prokaryote proteome is comprised of approximately 3,000 genes. A study of just 31 selected genes equates to only 1% of the proteome resulting in a ‘skimpy’ tree of life.

We have attempted to maximize the amount of data used in the inference of a tree of life in order to improve accuracy and to provide a more complete picture of the evolutionary history of life. Phylogenies based on single genes often do not contain enough phylogenetic information to adequately address the question. Using a single gene can result in sampling effects that can lead to several parts of the tree of life being poorly resolved. For this reason the tree of life that we built using ICHEC facilities was constructed from 393 whole genomes and additionally, duplicated genes were not excluded from the

analysis. Duplicated genes are often neglected in phylogenetic analyses. Apart from a small number of attempts, most studies only utilise single copy loci for their phylogenetic inferences. These studies that eliminate duplicated genes from a phylogenetic analysis place major restrictions on genomes that have undergone whole genome duplications.

In order to include duplicated genes in the inference of our tree of life, a number of computational methods were employed to reconcile these duplications. This was a very computationally intensive step in the analysis. Considering that the number of potential trees increases rapidly with the number of species and finding the optimal species tree is NP-complete, heuristic methods must be employed for this inference. Only one published software program is available for solving the gene duplication problem on a large scale, this is the DupTree software. However this software also has major limitations. For the aforementioned reasons we have employed a novel approach to infer species trees from large sets of duplicated gene families spanning 393 genomes in the hope of deriving a tree of 100%. Prior to this study the largest number of duplicated gene families that have been employed to infer a tree was 3,978 and this was done using the aforementioned software, DupTree.

This study inferred a tree of life from 6,455 duplicated gene families, making it the most comprehensive tree of life that has been constructed using duplicated genes.

Our results have identified a number of interesting groups on the tree. Many of these groups have been identified using previous methods, however our approach can also include information on duplication and losses throughout evolution.

LIFE SCIENCE FUNCTIONAL VARIATION IN EVOLUTION & POPULATIONS

Prof. Denis ShieldsUniversity College Dublin

BACKGROUNDOur research group is focused on discovering by bioinformatics methods patterns in biological sequences or in biological data that reveals functional importance. We apply an evolutionary perspective in ranking and evaluating potential regions of proteins, and genomes, as being of functional importance. We receive support from ICHEC for this research programme across five areas: novel motif discovery; genetic variation in populations; structural modeling of peptides; mapping from peptides to peptidomimetics; and in order to improve methods for discovering shared short motifs, remote homology searching across protein interaction networks.

These analyses contribute insights into the function of short motifs in adhesion processes, including platelet clotting associated with heart attacks and epithelial to mesenchymal transition associated with kidney disease, the function of short motifs in viral infections of host cells, and functional variation associated with neuropsychiatric disease. We anticipate that novel computer methodologies will emerge from this research, in particular for mapping from peptide sequences that encode discovered short motifs, to peptidomimetic compounds that are useful reagents in teasing apart cellular signaling, and that may potentially act as the template for the design of therapeutic drugs. Our recent investigation in Nature Chemical Biology of bioactive peptides combined evolutionary modeling with experimental validation1. We wish to refine further the rules that establish what makes a bioactive peptide. We also aim to go beyond this, to computationally predict not only rules that help predict which peptides are bioactive1, 2 but peptidomimetics that may be worthy of experimental validation3. We are interested not only in evolutionary variation in protein and gene structure and function, but also in population variation that may give insights into fundamental processes that lead to disease in certain individuals.

DOCKING APPROVED DRUGS TO PEPTIDE-BINDING DOMAINSDr. Fergal Casey, IRCSET-funded postdoctoral fellow, led a research project to discover FDA approved drug compounds that dock to the peptide binding sites of a set of structurally characterized domains. This relied on the use of the software package AUTODOCK4 on ICHEC systems to allow the comparison of docking scores across multiple targets and drugs. Essentially this

Figure 1: Discovery of small molecule inhibitors of protein-protein interactions using combined ligand and target score normalization5.

two-way normalization of scores permitted better assignment of interesting matches. The findings from this survey have been published in the Journal of Chemical Information and Modeling5.

SHORT LINEAR PROTEIN MOTIFSAnalysis of short linear protein motifs in viruses is led by SFI-funded PhD student Ravindra Pushker, who is surveying the distribution of eukaryotic linear motifs in viral proteins to determine the extent, if any, of their over-enrichment. This builds on our previous work on the discovery and analyses of short linear motifs6, and software we have developed in the form of the SlimSuite package of programs and associated web server7,8. Dr. Niall Haslam (SFI funded postdoctoral fellow) has been working with ICHEC staff (Kashif Iqbal and Simon Wong) to integrate some of these programs in the BioPortal project. This work is carried out in close collaboration with our colleagues Dr. Richard Edwards at the University of Southampton and Dr. Norman Davey at the European Molecular Biology Laboratory (EMBL) in Heidelberg.

GENOME WIDE ASSOCIATION OF POLYMORPHISMSWe are investigating the role of non-synonymous SNPs (single nucleotide polymorphism) in human disease. This requires imputation of missing genotypes (by merging the HAPMAP haplotype resource with Wellcome Trust Case Control Consortium data for seven diseases, 2000 cases for each disease versus 3000 controls, with over 400,000 genotyped SNPs and approx. 12,000 non-synonymous SNPs imputed). The software IMPUTE has been employed for this task and a custom pipeline developed for quality control of the imputed data. The next major computational task is the permutation testing of significance. Our goal is to discover unknown associations that can be validated in separate populations, and to characterize the role of context (disordered protein regions versus ordered) and evolutionary rate in determining the risk associated with a SNP variant.

1.Edwards, R.J., Moran, N., Devocelle, M., Kiernan, A., Meade, G., Signac, W., Foy, M., Park, S.D., Dunne, E., Kenny, D. & Shields, D.C. (2007) Bioinformatic discovery of novel bioactive peptides. Nat. Chem. Biol. 3, 108-112.2.Parthasarathi, L., Devocelle, M., Sondergaard, C., Baran, I., O’Dushlaine, C.T., Davey, N.E., Edwards, R.J., Moran, N., Kenny, D. & Shields, D.C. (2006) Absolute net charge and the biological activity of oligopeptides. J. Chem. Inf. Model. 46, 2183-2190.3.Baran, I., Svobodova-Varekova, R., Suchomel, S., Casey, S. & Shields, D.C. (2007) Identification of potential small molecule peptidomimetics similar to motifs in proteins. J. Chem. Inf. Model., in press.4.Goodsell, D.S., Morris, G.M. & Olson, A.J. (1996) Automated docking of flexible ligands: applications of AutoDock. J. Mol. Recognit. 9, 1-5.5.Casey, F.P., Pihan, E. & Shields, D.C. (2009) Discovery of small molecule inhibitors of protein-protein interactions using combined ligand and target score normalization. J. Chem. Inf. Model. 49, 2708-2717.6.Davey, N.E., Shields, D.C. & Edwards, R.J. (2006) SLiMDisc: short, linear motif discovery, correcting for common evolutionary descent. Nucleic Acids Res. 34, 3546-3554.7.Davey, N.E., Edwards, R.J. & Shields, D.C. (2007) The SLiMDisc server: short, linear motif discovery in proteins. Nucleic Acids Res. 35, W455-459.8.Edwards, R.J., Davey, N.E. & Shields, D.C. (2007) SLiMFinder: A Probabilistic Method for Identifying Over-Represented, Convergently Evolved, Short

Linear Motifs in Proteins. PLoS ONE 2, e967.


LIFE SCIENCE IDENTIFYING TAXON RESTRICTED GENES IN HIGHER PLANTS AND EPIGENOMICS OF GENOME DOSAGE IN ARABIDOPSIS THALIANA

Dr. Charles Spillane University College Cork

TAXON RESTRICTED GENES IN HIGHER PLANTS All sequenced genomes contain a proportion of genes that are lineage-specific, with no sequence similarity to any genes outside the lineage. The origins and functions of such genes are largely unknown. As more genomes are sequenced the opportunities for understanding evolutionary origins and functions of lineage-specific genes are increasing.

This study provides a comprehensive and systematic analysis of the origins of lineage-specific genes (LSGs) in Arabidopsis thaliana that are restricted to Brassicaceae (mustard family). Lineage-specific genes in the nuclear (1761 genes) and mitochondrial (28 genes) genomes are identified. The evolutionary origins of two thirds of these lineage-specific genes are elucidated in this study. Almost a quarter of lineage-specific genes originate from non-lineage-specific paralogs, while ~10% of lineage-specific genes contain DNA exapted from transposable elements (which is twice the proportion observed for non lineage-specific genes). Lineage-specific genes are also enriched in genes that have overlapping CDS. Over half of the subset of the 958 lineage-specific genes found only in Arabidopsis thaliana have alignments to intergenic regions in Arabidopsis lyrata. A smaller number of lineage-specific genes with incomplete open reading frames, across different Arabidopsis thaliana accessions are further identified as accession-specific. Putative de novo origination for two of the Arabidopsis thaliana-only genes was identified via additional sequencing of closely related sister species. Lineage-specific genes are demonstrated to have high tissue specificity and low expression levels across tissues and developmental stages. Finally, stress responsiveness is identified as distinct feature of lineage-specific genes; where lineage-specific genes are enriched for genes responsive to a wide range of abiotic stresses.

By identification of the lineage-specific genes in Arabidopsis thaliana and their origins, this study comprehensively elucidates the evolutionary mechanisms for producing novel ORFs and the relative importance, in terms of prevalence of each mechanism. Insights regarding the functional roles of lineage-specific genes are advanced through identification of enrichment for stress responsiveness in lineage-specific genes, highlighting their possible importance for environmental adaptation strategies.

EPIGENOMICS OF GENOME DOSAGE IN ARABIDOPSIS THALIANAEpigenetics refers to heritable and potentially reversible changes in genome function that occur without altering DNA nucleotide sequence. Understanding epigenetic regulatory systems is essential for development of novel biotechnologies to better harness epigenetics for crop improvement. Genome dosage effects and genomic imprinting are important epigenetic effects

on gene regulation, now amenable to investigation using high-throughput genomics and systems biology approaches. Gene dosage changes at the whole genome (e.g. auto-polyploids) or chromosomal (e.g. aneuploids) levels can elicit important phenotypic effects in crops.

As genome dosage effects can be genotype dependent or independent, a microarray experiment on Col-0 accession has been performed using AtSNPtile1 chips, which is a custom array from Affymetrix. Preliminary analyses have looked for structural changes between the diploid and tetraploid parents. Single feature polymorphisms (SFP) and insertions or deletions (indels) were searched for. The reasoning behind his analysis is two fold; (1) To identify structural genomic differences of plants of different ploidy that may effect gene regulation, (2) Filter out polymorphic and indel sites from downstream transcriptome analysis.

DNA treated arrays were preprocessed (spatial correction and log transformed), quantile normalized and then analysed using significance of microarray analysis (SAM), part of the siggens package (Bioconductor). For groups of features a single t-statistic for the diploid/tetraploid difference was derived by regression; false discovery rate (FDR) was determined by permutation. SFP analysis indicates no significant SFPs between the diploid and tetraploid plants at a FDR of 0.05. Indels between diploid and tetraploids were identified using a segmentation algorithm on the genomic hybridization intensities. P-values from one-sided two sample t-tests for the alternative hypothesis H1: tetraploid > diploid provided the probe level data for the segmentation algorithm. Both Bayesian (BIC) and Akaike information criterion (AIC) were used to define segments. Deletions and duplications were defined as segments with a median p-value > 0.99 and a median p-value < 0.02, respectively. Using BIC to define segments revealed no indels. In contrast, using AIC three duplications and six deletions were identified in the tetraploid compared to the diploid. The duplications sizes range from 104 to 317 bp. The deletion sizes range from 133 to 189 bp. We have identified differentially expressed genes across the ploidy series using the following linear model: Intensity = ploidy + male excess + female excess + error.

Preliminary results indicate a largely stable chromosome structure across the ploidy series. Notably all but one of the indels between the tetraploids and diploids overlap gene model locus within the genome and may effect plant development.

PHYSICSMULTIFERROIC TUNNELING JUNCTIONS

Thomas Archer, Nuala Caffrey and Stefano SanvitoTrinity College Dublin

ABSTRACTSpintronics plays an important roll in modern technology. Most notable is the discovery of the the giant magneto resistance (GMR) effect, where the current through a non-ferromagnetic spacer sandwiched between two ferromagnetic layer is strongly affected by the external magnetic field. This discovery has revolutionized the read head in hard disks, allowing the increase in storage density of an order of magnitude.

Ferroelectric materials are also starting to play an important roll as nonvolatile ferroelectric RAM. However to read from such a memory is significantly slower than from standard memory, since the data is wiped by the reading process.

Combining ferroic attributes in a single device may allow devices to be created that bypass the problems of each single ferroic attribute, allowing the creation of a range of new multiferroic devices. Currently no usable single phase material exists, any usable device would most likely need to be a heterostructure of ferroic materials.

In this work we intend to calculate from first principals the surface structure and the transport characteristics of a tunneling junction where the spacer material is replaced with a ferroelectric material.

PHYSICS SCANNING TUNNELING MICROSCOPY FOR MAGNETIC MOLECULES

Nadjib Baadji and Stefano SanvitoTrinity College Dublin

Magnetic molecules are among the most promising material systems for spintronics applications. Their small size together with the possibility of constructing large molecular superstructures make them an attractive platform for magnetic data storage. The understanding of their electronic structure and its correlation to their physical properties is a big challenge for both theory and experiment. From the theoretical point of view most of the existing ab initio approaches usually fail to describe properly the electronic structure of magnetic molecules as well as their physics. Furthermore the small size and reduced symmetry of molecules in general complicate the possibility of a detailed experimental characterization.

When deposited on surfaces at low coverage the sensitivity of the most commonly used experimental tools for investigating the electronic and geometric structure of materials are not sufficient.

This problem can be overcome by using Scanning Tunneling Microscope (STM), which is a powerful technique for imaging surfaces at the atomic level. The high resolution provided by STM allows one to study of molecules and nanostructures on surfaces. STM can be also used to control and modify nanostructures1, which makes it a suitable tool for investigating the physical properties of magnetic molecules. The interpretation of the obtained STM images requires the modeling of the electronic structure. The most common theoretical scheme is based on the Tersoff-Hamann (TH) approximation2, where the tip is assumed to be a single atom with s-like spherical symmetry. This model is widely used to relate the STM patterns and the electronic structure of systems under study. The general agreement between theory and experiment is mainly due to the recognition of certain symmetries that samples possesses. The TH model is however of limited use when the molecule has little symmetry and a simple eyes-guided comparison between theory and experiments is not informative. Furthermore neglecting the electronic structure of the tip may result is neglecting important features of the STM images, which are not attributable to the molecule to investigate but simply to the scanning tip.

We have developed a new computational scheme where both the tip and sample are treated on the same footing, and we were able the show that the geometry of the tip as well as its nature can dramatically change the simulated STM pattern. We did apply the model to Co-salen molecules on Cu surfaces and the agreement of both the topographic as well as the mapping with experiments is extremely good. We used for the basic electronic structure the siesta code4, which employs local basis set and allows us to simulate such big molecules. The electronic structure obtained with siesta will be the input for simulating the STM images. As a benchmark we also plan to carry out transport calculations for the same molecules by using the non-equilibrium Green’s function formalism as implemented in the SMEAGOL code5 developed by our group. The STM of other molecules are simulated and compared to experiments.

1 H. C. Manoharan, C. P. Lutz, and D. M. Eigler, Nature 403, 512 (2000).2 J. Tersoff and D. R. Hamann, Phys. Rev. Lett. 50, 1998 (1983); Phys. Rev. B 31, 805 (1985)3 C.J. Chen, “Introduction to Scanning Tunneling Microscopy”, Oxford University Press, Oxford, (1993).4 J.M. Soler, E. Artacho, J.D. Gale, A. Garcia, J. Junquera, P. Ordejon and D. Sanchez-Portal, J. Phys.: Condens. Matter 14 2745 (2002)5 A.R. Rocha, V.M. Garcia-Suarez, S.W. Bailey, C.J. Lambert, J. Ferrer and S. Sanvito, Phys. Rev. B. 73, 085414 (2006); www.smeagol.tcd.ie


PHYSICS DEFECTS AND IMPURITIES INDUCED MAGNETISM IN OXIDE SEMICONDUCTORS AND INSULATORS

Andrea Droghetti, Sri Chaitanya Das Pemmaraju and Stefano SanvitoTrinity College Dublin

INTRODUCTIONThe so called d0 magnets are wide-gap semiconductors and oxide insulators displaying magnetic properties which cannot be ascribed to the presence of partially filled d or f shells1. Prototypical examples are diamagnetic materials in their well crystallized bulk form, which present magnetic features when grown as defective thin films or doped with light elements such as C and N. Despite the increasing number of reports of d0 magnets the phenomenon remains rather obscure since it is characterized by a poor degree of experimental reproducibility and lacks of a convincing theoretical framework.

Another important, but yet scarcely studied, class of d0 magnets is represented by zincblende (ZB) II-V or II-IV 2, 3, 4, 5 compounds such as CaP, CaAs, CaSb or CaC, SrC and BaC, which are weakly covalent solids with either one or two holes per formula unit, and they are generally predicted to be half-metallic.

ACHIEVED RESULTSDFT calculations might be very helpful in order to investigate the physical properties of defects and impurities related magnetism. It shows that the magnetic moment formation is associated to spin-polarized holes residing on either cation vacancies or impurities. Critically most of the studies appeared until now are based on calculations performed with local approximations of the exchange cor-relation potential (LSDA or GGA). These notoriously underestimate Coulomb repulsion and tend to over-delocalize the charge density returning a metallic (often half-metallic) ground state and usually extremely large magnetic interaction. We showed that strong electron correlations may play a fundamental role in the magnetic moment formation, which subtly depends on the interplay between covalency, Hund’s coupling and polaronic distortion around the impurity. Corrections to the LSDA/GGA such as self-interaction correction (SIC) schemes often return an insulating ground state, holes trapped by polaronic distortions and no long range magnetism. For instance in rock-salt oxide MgO doped with N substituting for O, one finds that the extra hole entirely localizes around one of the 2p orbitals as a consequence of the coupling with phononic modes (as shown in Fig. 1). Moreover we demonstrated that the magnetic moment is stable against charge fluctuations.

In addiction to d0 magnetism in defective or doped materials, we also investigated the magnetic ground state of RS MgN, which was already predicted as half-metal in the ZB phase4. This is an extremely interesting material because of its structural similarity to the parental MgO, widely used as tunnel barrier in magnetic tunnel junctions6.

We demonstrated that MgN is a half-metal (as clearly shown in Fig. 2, where the band structure and the Density of states are displayed) with negligible spin-orbit interaction and negative formation enthalpy. This means that, although it is not the thermodynamically stable phase, it can be grown as a metastable compound in the form of either thin films or as nanoclusters inside MgO matrices.

1 J.M.D. Coey, Solid State Sci., 7, 660 (2005).2 K. Kusakabe, M. Geshi, H. Tsukamoto and N. Suzuki, J. Phys.:Condens.Matter 16, 5639 (2004).3 O. Volnianska, P. Jakubas and P Bogus lawski, J. Alloys Compd., 423, 191(2006).4 M. Sieberer, J. Redinger, S. Khmelevskyi and P. Mohn, Phys, Rev. B 73,024404 (2006).5 G.Y. Gao et al., Phys. Rev. B,75 174442 (2007).6 S.S.P. Parkin et al., Nature Mater. 3, 862 (2004).

PUBLICATIONS RELATED TO THIS PROJECTPredicting d0 magnets, A. Droghetti, C. D. Pemmaraju and S. Sanvito, Phys. Rev. B, 78, 140404(R) (2008).Electron doping and magnetic moment formation in N- and C-doped MgO, A. Droghetti and S. Sanvito, accepted for the pubblication in Applied Physics Letters.MgN:a new promising material for spintronic applications,A. Droghetti, N.Baadji and S. Sanvito, submitted to Physics Review Letters.Ground state properties of cation vacancies in MgO, A. Droghetti, C. D. Pemmaraju and S. Sanvito, in preparation.

Figure 1

Figure 2

PHYSICS FUNCTIONAL PROPERTIES OF MULTIFERROIC HETEROSTRUCTURES

Dr. Claude EdererTrinity College Dublin

In this project we will explore the coupling between magnetic properties, strain, and electric polarization in epitaxial heterostructures containing both magnetic and ferroelectric components. Such structures are extremely promising for the design of new types of memory devices that combine the advantages of magnetic and ferroelectric RAM, while providing a way to avoid some of the crucial disadvantages of these two technologies. For this research we will employ state-of-the-art first principles calculations based on density functional theory. This method allows us to make quantitative predictions for many characteristic properties of such multiferroic heterostructures, analyze the underlying mechanisms, and identify the most promising materials for further experimental investigations.


Figure 2, unoccupied (acceptor) levels located at energies above the mid-gap of silicon nitride up to the conduction band are sampled by TSCIS. In 2003, it was reported that the use of a high-k oxide (like Al

2O3) as the top dielectric could alleviate the erase saturation problems (TANOS stack technology), thus allowing a thicker tunnel oxide and avoiding the retention issue. Atomic layer deposition (ALD) is therefore proposed as a relatively mild technique for alumina deposition, with the possibility of atomic-level control in the optimum case. Therefore, with the aim of fine-tuning the deposition conditions of the top oxide towards the desired trap density and leakage paths, the atomic-scale structure and reactivity of silicon nitrides surfaces towards trimethylaluminium (TMA) has been studied by us using ICHEC resources. We have investigated the effect of the surfaces of silicon oxide (SiO

2), oxynitride (Si2N2O) and silicon nitride (Si3N4) substrates on the initial growth rates in Al2O3 ALD. The degree of hydroxylation of the surface is a crucial quantity, and the simulations show what coverage of surface-OH groups is favoured under the ALD conditions. The rate estimates of Al

2O3 growth per cycle using the theoretically determined coverage agree well with those estimated using experimental values4.

1 Gigascale Oriented Solid State flash Memory for Europe: www.fp7-gossamer.eu2 M. E. Grillo, S. Elliott, and C. Freysoldt, submitted to Appl. Phys. Lett. (2010).3 A. Suhane, A. Arreghini, R. Degraeve, G. Van den bosch, L. Breuil, M. B. Zahid, M. Jurczak, K. DeMeyer, and J. Van Houdt, submitted to IEEE (2009).4 S. D. E. Elliott, G. Scarel, C. Wiemer, M. Fanciulli, and G. Pavia, Chem. Mater. 18, 3764 (2006).

PHYSICS ATOMIC-SCALE CHARACTERIZATION OF CHARGE TRAPS IN SONOS MEMORY DEVICES, AND OF THE EFFECT OF SURFACE REACTIVITY ON GROWTH RATES IN AL2O3 ALD

Dr. Maria Elena GrilloTyndall National Institute

The present study is part of a large European R&D project (GOSSAMER)1 that addresses for the first time the development of large size NAND Flash memories based on the charge trapping mechanism for technology generation in the range of 36-28nm. Closing the gap between research and industrial exploitation, atomic-scale characterization of the defects in silicon nitride that influence the quality of the memory device was carried out by means of first-principles density functional theory (DFT) on the ICHEC computers. Therefore, a comprehensive characterization of the atomic and electronic structure of the intrinsic charge trap defects in silicon nitride has been performed.2 The present results show that nitrogen vacancies (Figure 1) are the most abundant native defects in silicon nitride, and account for the levels sampled by different charge trap profiling techniques. The unoccupied electronic levels in the gap introduced by the nitrogen vacancies are relevant to interpret the recently reported peak in the trap energy spectrum at 1.6 eV below the conduction band by trap spectroscopy by charge injection and sensing (TSCIS).3 As shown schematically in

Figure 1: Structural model for the neutral nitrogen vacancy (right), and highest negative charge state (q=-3) below the conduction band (left). The silicon nitride host structure is shown in a stick representation. The Si atoms around the nitrogen vacancy are shown in light grey and balls.

Figure 3:Top views of hydroxylated (111) surface of SiO2, and surface electrostatic potential color mapping onto an electronic density isosurface indicating potential reactivity for ALD reactions. Yellow=highest acidity, pink=basic

Figure 3: Schematic representation of the calculated thermodynamic transition level introduce by nitrogen vacancies in the gap, and energy position of the trap density peak reported by TSCIS. The measurement principle of trap energy levels within a SONOS stack is schematized below.

PHYSICS COMPUTATIONAL MODELLING OF MATERIALS FOR ARTIFICIAL PHOTOSYNTHESIS


The dominant focus of this research project is concerned with the detailed classical and quantum molecular simulation modelling of nanoporous photocatalytic titania (TiO

2) films doped with varying amounts of cationic transition metals (e.g., copper, iron) and non-metal anions (e.g., nitrogen, carbon), with a view to enhancing the photoactivity vis-à-vis the pure titania state. Doping with an optimal amount of such impurities is expected to lead to a maximum photo-efficiency, due to a delicate balance between an increase in photo-excited electron trapping by transition metal recombination sites with increasing dopant concentration, and longer lifetimes for charge transfer to the nano-films’ surfaces, where subsequent water-splitting can take place.

First-principles ground-state, Density Functional Theory (DFT) has been used to study doping effects. The VASP 4.6 periodic DFT implementations has been used with all-electronic quantum wavefunctions to optimise the geometry of (doped) bulk and surface-state titania and dye-titania complexes. Electronic structure calculations were performed on these optimised systems to estimate the density of states (DOS) and projected density of states (PDOS). The DFT +U treatment has also been used to estimate more accurate band gap properties for partially filled d orbitals as in titania.

Figure 1: Density of states (DOS) and projected DOS for (a) pure TiO2, (b) N-doped TiO2, (c) W-doped TiO2, and (d) N/W-doped TiO2 and the corresponding PDOS shown in (a’)-(d’). The PDOS plots of (b’)-(d’) are enlarged for clarity.

Mono- and co-doping has been investigated in bulk and surface-state anatase and rutile forms of titania. For codoping, it has been found that transition metals (W, P, Bi) coupled with non-metals (e.g., S, N, C) can be most effective, due to the less apparent prevalence of recombination centres in the band gaps of these doped materials (where released holes and electrons can recombine). The formation energies of these doped systems have been computed, to assess their thermodynamic viability and give insight into the feasibility of experimental fabrication. The DOS and PDOS have also been evaluated to determine the band gap and electronic properties, with a view to selecting the best possible candidates for application as materials in photovoltaic cells (e.g., cf. Fig. 1). In many cases, experimental data on photo-activity in mono- and co-doped samples has been rationalised by elucidating likely mechanisms for photovoltaic activity (or lack thereof ) from electronic band gap information predicted by first-principles simulation 1-9.

1Appl. Phys. Lett., 94, 132102 (2009)2J. Phys. Chem. C, 113, 8373 (2009).3J. Phys. Chem. C, 113, 9423 (2009).4Chem. Phys. Lett., 478, 175 (2009).5J. Phys. Chem. C, 113, 17464 (2009).6Phys. Rev. B, 80, 115212 (2009).7Phys. Lett. A., 374, 319 (2009).8Phys. Lett. A 374, 1184 (2010) 9Chem. Materials, 22, 1616 (2010)


PHYSICS SEMICONDUCTOR AND MOLECULAR NANOWIRE SIMULATION FOR TECHNOLOGY DESIGN

Dr. Giorgos FagasTyndall National Institute

Semiconductor nanowires are under intense investigation in many technology areas, such as nanoscale electronics, photonics and sensors. Recent progress in growth and fabrication techniques has allowed both bottom-up and top-down demonstrations of devices with diameters of only a few nanometers. Silicon nanowires (SiNWs), in particular, offer a path to the ultimate limits of scaling with the potential advantage of ease of integration with existing materials in use in semiconductor technologies.

Si/Ge nanowires may demonstrate a direct band gap due to zone folding; also, as the wire diameter decreases, the band gap increases. Details depend on wire growth orientation and the shape of the cross-section. Most significantly, we have recently shown that surface preparation also changes the electronic structure of SiNWs. In a similar fashion, the unavoidable formation of oxygen-derived defects at the surface of a pure wire core due to oxidation largely determines the properties of Si/Ge nanowires. Surface oxidation of group IV semiconductor nanowires is a naturally occurring process which results in an oxide sheath around the wire core. Although the native oxide can be removed via HF-etching to provide hydrogenated surfaces that are, for example, subsequently functionalized for sensor applications, a residual density of oxygen impurities is expected. In several applications, further treatment of the oxide is instrumental to device operation, as for the case of defect states near the core/oxide interface believed to be responsible for part of the nanowire photoluminescence spectrum. Also, a notable 2 order of magnitude enhancement of the hole mobility has been experimentally demonstrated after chemical modification of the SiOx surface.

In general, oxygen impurities promote the silicon atoms at the outer part of the core to various oxidation states and distort the underlying lattice periodicity, thus, forming scattering centers for charge carriers. To understand the impact of oxygen related defects on electrical conduction, it is instructive to view the defects as a source of roughness on the atomic scale. It is well-known that surface roughness is a key factor for charge carrier transport as it reduces mobility in planar metal-oxide-semiconductor field effect transistors (MOSFETs). More recently, the detrimental effect of surface roughness was shown to be even more pronounced as the body of the silicon is scaled down to a few nanometers in MOSFETs made of films of ultrathin silicon-on-insulator. By comparison, SiNWs are an extreme realization of a thin semiconductor body with confinement in an additional spatial dimension yielding one-dimensional transport.

In this project, we have so far assessed the local surface oxidation-limited carrier transport in silicon nanowires with growth direction along [110] and diameters of a few nanometers.

Contrary to expectations, we find that the mean free paths are comparable to or longer than the wire lengths envisioned for transistor and other nanoelectronics applications1. Transport along [100]-oriented nanowires is less favorable, thus providing an advantage for [110] nanowire orientations, as preferentially produced in some growth methods.

1 G. Fagas, J.C. Greer, ‘Ballistic Conductance in Oxidized Si Nanowires’, Nano Lett. 9, 5 (2009)

PHYSICS ELECTRONIC PROPERTIES OF OXIDISED SILICON NANOWIRES

Dr. Giorgos FagasTyndall National Institute

ABSTRACT

The recent fabrication of semiconductor wires with just a few nanometers in cross-section has upheld their strong position for applications in nanoelectronics and nanophotonics. Unlike their bulk counterparts, the electronic structure of Si/Ge nanowires (NWs) depends on growth orientation, the detailed shape of the wires’ cross-section as well as surface chemistry. For example, we have shown that hydroxylated Si NWs show a notable red shift in the band gap compared to their hydrogenated counterparts at similar diameter. In addition, the oxidation process may induce up to a few percent strain to the nanowire core due to lattice mismatch. Uniaxial strain in reference hydrogenated nanowire structures yields significant variation in the band gap and may increase the electron effective masses while decreasing the hole effective mass. In this project, we will use first principles calculations (Density Functional theory) to address the oxide-induced modifications to the electronic structure of sub-3nm Si nanowires.

PHYSICS NANOSCALE SIMULATORS IN IRELAND (NSI) GRADUATE PROGRAMME PRACTICALS

Prof. Stephen FahyUniversity College Cork

Nanoscale Simulators in Ireland (NSI) is a network of scholars and researchers involved in the theory and computation of molecules and materials at the atomic-scale. This field encompasses much of computational chemistry and physics, but is also relevant to the biosciences, materials science and electronic engineering. A common theme is the investigation of how nanoscale structures, such as electrons, atoms and molecules, lead to observed properties and useful technologies. The simulation methods used are generally based on classical or quantum mechanics (especially, electronic structure theory). In autumn 2008, NSI members have begun developing a series of course modules on nanoscale research methods and background theory for postgraduate students. These courses involve lectures delivered via the web, assignments and practical exercises. The aim of this project is to provide the necessary computational resources for students partaking in NSI courses.

PHYSICS AGGREGATION AND COLLECTIVE MOTION OF SWIMMING ORGANISMS AND CELLS IN FLUID FLOWS

Dr. Zoltan NeufeldUniversity College Dublin

The proposed research investigates the effects of mixing and transport on the distribution of swimming organisms and cells living in a moving fluid environment. Such organisms like oceanic plankton or bacteria play an important role in marine ecosystems and geochemical cycles and are also widely used in industrial biochemical processes. Mathematical modeling and computational approaches will be used to gain insight into the effect of flows on various aspects of their behavior like aggregation, group formation, orientation using chemical signals, foraging strategies etc. with applications to environmental modeling and biotechnology in controlling the dispersion, separation or aggregation of microorganisms.


PHYSICS

ATOMIC SCALE MODEL INTERFACES BETWEEN HIGH-K SILICATES AND GERMANIUM

Dr. James C. Greer, Ms. Sinéad Griffin and Dr. Simon D. Elliott

Tyndall National Institute

Computer chips based on silicon have transformed society over the last fifty years, but the Silicon Age may soon be over. Silicon was chosen as a semiconductor for logic circuits largely because of the properties of its oxide. In the last two years, silicon oxide has been replaced in the logic circuit with materials based on hafnium oxide that are more insulating – as for example in the Xeon processors in ICHEC’s state-of-the-art Stokes computer. With silicon oxide out of the way, chip manufacturers are now looking at semiconductors that perform better than silicon. One example is germanium, since electrical current moves more than twice as easily through germanium as silicon.

At Tyndall we have been simulating the structures that will be formed when hafnium oxide based insulators are interfaced with a germanium semiconductor in the logic element of the chip. The dimensions of these devices will be well below 50 nanometers (you could line up a million of them around the circumference of a human hair) and so the structure of individual atomic layers at this interface will be important for how well the device works.

In recent years, advances in theoretical approaches and computing power have allowed high-level quantum mechanical simulations of ever-increasing complexity: first of pure solids and then of their surfaces. Computing interfaces is the next frontier for materials simulation. Interfaces are in any case of increasing importance for nanotechnology, including the nanoelectronics application outlined above. The interface simulations at Tyndall and ICHEC are some of the most extensive yet carried out anywhere to date, and the first of their kind to consider a this material system. To cover the ensemble of possible interface structures, we have generated a series of atomic models that span the extremes of what is experimentally-realisable. Simulated annealing and geometry optimisation was carried out on each model to ensure a realistic structure, and the quantum mechanical energy was used to assess relative stability. The scale of these simulations required extensive use of ICHEC facilities.

The results reveal for the first time the atomic-scale structure of sub-stoichiometric interfacial layers, and in particular the differences between silicon and germanium. The Ge-Ge dimers that are a feature of the bare surface are also found to play a role in the interface. The simulations show that annealing in oxygen will lead to progressive growth of a germanium sub-oxide interfacial layer, via a cycle of cleavage and formation of dimers. Dimer formation means that defects, such as three-coordinate Ge, occur spontaneously in the interfacial layer, and this will spoil the electrical performance of the nano-device.

Figure 1: Computed model of atomic structure between hafnium oxide based insulators (hafnium=blue, oxygen=red, silicon=yellow) and semiconducting germanium (green). These structures are under investigation for increased processor speed in future computers. However defects will reduce their efficiency – this model features a defective three-coordinated germanium atom (green ball).

PHYSICS

SPIN POLARONS IN P-TYPE MAGNETITE

Aurab Chakrabarty and Charles PattersonTrinity College Dublin

Magnetite is a ferrimagnetic oxide with a Curie temperature in excess of 800C. Magnetic oxides which retain their magnetism above room temperature are not common and magnetite is currently under experimental investigation for magneto-electronic device fabrication. It has a spinel structure in which formal charges on Fe ions are FeA3+FeB2+FeB3+O48-. The Fe A site ions are surrounded by O ion tetrahedra and the B sites are surrounded by O ion octahedra. Below T = 120K, the famous Verwey transition, magnetite undergoes a transition in which the high symmetry, metallic state with Fd3m space group symmetry breaks symmetry and forms a semi-conducting state with P2/c symmetry. Since there is an equal mixture of Fe2+ and Fe3+ ions on B sites, magnetite can form a metallic state in which there is rapid exchange of electrons between Fe2+ and Fe3+ sites, or an insulating state in which these electrons order. Recent work from our group1 showed that density functional theory predicts the metallic ground state with Fd3m symmetry as the ground state while hybrid density functional theory predicts the semi-conducting P2/c state with charge order on the Fe B sites. Charges order in this state in a somewhat unexpected way. There are four distinct B sites (B1,.., B4) in the P2/c structure which form B1 and B2 chains and B3/B4 combined chains of octahedra in the magnetite spinel structure. We found that B1 chains consisted of Fe2+ ions, B2 chains were Fe3+ ions and mixed B3/B4 chains were Fe3+ on B3 sites and Fe2+ on B4 sites.

More recently we have investigated what happens when magnetite is doped with Li and which states form when magnetite has an oxygen or iron vacancy. The compound lithium ferrite has stoichiometry Fe0.5Fe2.5O4. Lithium substitutes for Fe on tetrahedral A sites. Since it exists as Li+ in lithium ferrite, each Li dopant introduced removes two electrons from easily oxidised Fe2+ ions. At the lithium ferrite stoichiometry, all Fe exist as Fe3+ and lithium ferrite is a wide gap semiconductor. When higher concentrations of Li are introduced, electrons are removed from the O 2p band. A naive view would be that holes introduced into an oxide O 2p band in this way would yield a moderate mobility p-type material. However, in recent work on magnetic transition metal oxides, including cuprates2 and manganites3, we have found that holes doped in O 2p bands of oxides form polaronic states in which the hole becomes localised on one or several O ions. All of these calculations were carried out using hybrid density functional hamiltonians and the Crystal package4. Hybrid density functionals are a mixture of the exchange-correlation density functional and Hartree-Fock exchange. The important differences that Hartree-Fock exchange makes in predictions of oxide properties include an improved prediction of the single-particle band gap and localisation of holes on O ions. We show an example of a polaronic O 2p hole state in a lithium ferrite calculation where one extra Li has been introduced into the unit cell above the ferrite stoichiometry. Since the holes introduced also have a net magnetic moment they can be seen in a spin density plot. Figure 1 shows a spin density plot in a plane in a Li5Fe19O32 unit cell. A spin polaron localised on an O ion is labelled P1 and circled. It is adjacent to the extra Li ion which has been introduced. Fe ions (brown), O ions (blue), Li ions (grey).

1A. D. Rowan, C. H. Patterson and L. V. Gasparov, Phys. Rev. B 79, 205103 (2009).2 C. H. Patterson, Phys. Rev. B 77, 94523 (2008).3 C. H. Patterson, Phys. Rev. B 72, 085125 (2005).4 www.crystal.unito.it


PHYSICS

SIMULATING ELECTRONIC TRANSPORT IN SCANNING TUNNELLING MICROSCOPY

N. Baadji, C.D. Pemmaraju, I. Rungger, A. Droghetti, S. SanvitoTrinity College Dublin

Simulations of scanning tunneling microscopy (STM) measurements for molecules on surfaces are traditionally based on a perturbative approach, most typically employing the Tersoff-Hamann method. This assumes that the STM tip is far from the sample so that the two do not interact with each other. However, when the tip gets close to the molecule to perform measurements, its electrostatic interplay with the substrate may generate nontrivial potential distribution, charge transfer, and forces, all of which may alter the electronic and physical structure of the molecule. These effects are investigated with the ab initio quantum transport code SMEAGOL1, 2 combining the nonequilibrium Greens- function formalism with density functional theory. In particular, we investigate alkanethiol molecules terminated with either CH3 or CF3 end-groups on gold surfaces for which recent experimental data are available.

From a technical point of view we have demonstrated that the DFT-NEGF code SMEAGOL can be used to simulate STM-type experiments in the near to contact regime3. This is under the working condition of the tip-to-sample distance being sufficiently small that the basis orbitals have not been artificially cut off, so that the vacuum region between the tip and the surface is still well described. SMEAGOL then allows us to investigate systems in which

the tip interacts with the molecule, and to study effects of finite bias on the electronic structure of the molecule. We also calculate the complex band structure for different basis sets, and show that if we use additional basis orbitals in the vacuum region we can properly describe the exponential decay of the wave function in vacuum even at large separation between tip and molecule.

Calculations for alkanethiol molecules with STM-type arrangements show strong asymmetry in the I-V curves, which can be explained by the asymmetry in the coupling to the two different electrodes (substrate and tip). However, the calculated asymmetry is similar for both CH3- and CF3-terminated decanethiols, in contrast to the experimental measurements, showing a far stronger asymmetry for the case of CF3 termination. We have then investigated the suggestion, which attributed the asymmetry to small configurational changes in the molecule under bias due to electrostatic interaction between the tip and the substrate. Our calculations demonstrate that changes to molecule length are too small and are unlikely to have any major impact on the I-V curves. However, we also show that a rotation of the molecule may cause significant changes to the current, so that this might be at the origin of the large differences in the experiments with CH3 and CF3 termination.

1 A.R. Rocha, V.M. Garcia Suarez, S.W. Bailey, C.J. Lambert, J. Ferrer and S. Sanvito, Phys. Rev. B 73, 085414, (2006), Spin and Molecular Electronics in Atomically-Generated Orbital Landscapes.2 A.R. Rocha, V.M. Garcia Suarez, S.W. Bailey, C.J. Lambert, J. Ferrer and S. Sanvito, Nature Materials 4, 335, (2005), Towards molecular spintronics.3 C. Toher, I. Rungger and S. Sanvito, Phys. Rev. B 79, 205427, (2009), Simulating STM transport in alkanes from first principles.

Figure 1: (a) I-V curves for CH3-terminated decanethiol attached to a gold surface for different distances between the C atom in the CH3 group and the probing tip. Changing this distance by 0.1 °A causes the size of the current to change by approximately one order of magnitude at 1 V. (b) I-V curves for decanethiol molecules terminated with both CH3 and CF3 groups. Note the bias asymmetry, with the conductance at negative bias being two to three times larger than that for positive bias. The conductance is lower for the CF3-terminated molecule than that of the CH3-terminated.

PHYSICS

HOT QCD SIMULATIONS ON ANISOTROPIC LATTICES

Dr. Jonivar SkullerudNational University of Ireland Maynooth

ABSTRACT

At high temperatures, strongly interacting matter is expected to undergo a transition from a low-temperature phase where quarks and gluons are confined within hadrons and chiral symmetry is spontaneously broken (giving rise to the large masses of most hadrons), to a high-temperature phase, the quark–gluon plasma (QGP), where quarks and gluons are the effective degrees of freedom and chiral symmetry is restored. This state of matter is currently being probed in heavy-ion collider experiments at Brookhaven (RHIC) and CERN. This project will investigate the properties of the QGP using numerical simulations of the underlying theory, Quantum Chromodynamics (QCD). In particular, we will study the survival and properties of bound states of quarks, and the bulk thermodynamic and transport properties of the plasma.

PHYSICS

AB INITIO SIMULATIONS OF SPIN-DYNAMICS

Maria Stamenova and Stefano SanvitoTrinity College Dublin

ABSTRACT

The aim of this project is to develop a computationally feasible first-principles-based scheme for simulating the dynamics of the electronic spin degrees of freedom in real magnetic materials: from isolated magnetic molecules, to magnetic impurities on surfaces or in crystalline bulk phases. Such a scheme is within the framework of the time-dependent spin density-functional theory (TD-SDFT) and is currently being implemented for isolated systems in the context of the adiabatic approximation, where the time-dependent exchange-correlation (XC) potential is calculated from the ground-state XC density- functional applied to the instantaneous density distribution. The density-functional core of our numerical algorithm is provided by SIESTA – a standard order-N localized-basis static DFT code. As conceived, the scope of this project will extend to investigating spin-dynamics in magnetic structures under current carrying-conditions and beyond the adiabatic local density approximation (ALDA).


4.3 Class C Projects


ICHEC Class C Projects

Mr. Jakub Baran, Tyndall Institute Theoretical Investigation of Adsorption of Metal-Phthalocyanies (M=Co,Sn,Pb) on Ag(111) surface.

Dr. Pavel Baranov, UCCIdentification of Indel RNA Editing Patterns in the Human Genome

Dr. Volker Braun, DIAS A Prelude to Numerical Calabi-Yau Geometry

Dr. Nicolae-Viorel Buchete, UCDMultiscale Study of Molecular Aggregation Processes

Prof. Ann Burnell, NUI, Maynooth Annotation & Analysis of the Panagrolaimus Superbus Genome.

Mr. Alex Byrne, UCD Particle Acceleration in A Turbulent Magnetic Field

Dr. Maria Chernyakova, DIASFERMI Data Analysis

Dr. Marcus Claesson, UCC Genome Analysis of Gut Microbiota and Microbiomes

Dr. Colin Clarke, DCU Application of Machine Learning and Gene Selection Methods for Whole Genome Microarray Analysis of Breast Cancer.

Dr. Cian Crowley, TCDRadiative Transfer in the Atmospheres of Cool Giant Stars

Dr. Norman Davey, UCD High Throughput Discovery of Short, Linear Motifs

Dr. Sean Delaney, DIASInvestigation of Particle Acceleration in Astrophysical Shocks using Parallel Computing Techniques

Dr. Claude Ederer, TCDExploring New Functionalities in Complex Oxides

Mr. Alin Elena, ICHECTime Dependent Tight Binding+U: Applications to Fluorescent Molecules

Dr. Simon Elliott, Tyndall Institute Interim Time for SIGMUND (#4092)

Dr. Giorgos Fagas, Tyndall Institute Electronic Properties of Oxidised Silicon Nanowires

Ms. Emer Feerick, NUI, Galway Development of Advanced Analysis and Testing Methodologies for the Design of a Novel Device for the Fixation of Proximal Humerus Fractures

Dr. John Gallagher, DCU Molecular Modelling and Ab Initio Calculations of (i) Methyl-N-(Methylbenzoyl)-N-(2-pyridyl)Benzamides and their Halo Analogues and (ii) Pyridinecarboxamides

Dr. Jose Gracia, DIASOpenSESAMe - A Tool for Automated Synthetic Observations

Dr. Jim Greer, Tyndall Institute Monte Carlo Configuration Generation for Electronic Structure Computations

Dr. Maria Elena Grillo, UCC Charge Retention in SONOS Non-Volatile Memory Devices: Oxide-Nitride Interfaces and Deposition Process

Dr. Mark Hannam, UCC Gravitational Waves from Black-Hole Binary Collisions

Mr. Rejwanul Haque, DCU Semantic Information as Source Context in Phrase-Based SMT

Dr. Michael Hartnett, NUI, Galway Development of Coupled Ocean-Atmosphere Climate Model

Ms. Caroline Hopkins, NUI, Galway A Computational Investigation into the Delamination of Biomedical Stent Coatings

Mr. Xiaowei Jiang, TCD Evolution of Bacterial Secretion Systems

Dr. Sarah Kebbell, DCUSynthesis and DFT Calculations of 1,2,3,5-Tetrazocines

Ms. Nicola Kelly, NUI, Galway Computational Analysis of Normal and Postoperative Knee Joint Mechanics

Dr. Patrik Lambert, DCU Baseline Discriminative Word Alignment System

Mr. Cathal Leahy, UCD Study of Conformational Changes in Short Peptides

Ms. Noeleen Loughran, DCU Evolution of Mammalian Reproductive Proteins

Ms. Noeleen Loughran, DCU The Evolution of Genomic Imprinting

Mr. Mark Lynch, DCU Analysis of Sequence Patterns from Surface Layer Proteins Isolated from Different Strains of C. Difficile

Prof. Gary McGuire, UCD Minimum Number of Clues Problem

Dr. James McInerney, NUI, Maynooth Metabolic Network Evolution in Escherichia Coli and Relatives

Mr. Alastair McKinstry, ICHEC Tidally-Driven Oceanic Circulation of Exoplanets

Dr. Pavle Mocilac, DCUMolecular Modelling and Ab Initio Calculations Benzamides and Pyridinecarboxamides.

Mr. Tadhg Morgan, UCCCoherent Control of Atoms on Atom Chips Using STIRAP

Dr. Gareth Murphy, DIASPIC Simulations of Near and Mildly Relativistic Shocks

Mr. Ronan Murphy, UCDMultiscale Study of Protein Folding Processes

Dr. Zoltan Neufeld, UCD Collective Search Strategies in Turbulent Flows

Ms. Riona Ni Ghriallais, NUI, Galway Investigation of Vascular Injury in the Superficial Femoral Artery Due to Stenting

Dr. Lampros Nikolopoulos, DCUTime-dependent Schrodinger Equation of Multielectron Systems in an Electromagnetic Field: Parallel implementations of PTDSE code

Dr. Noel O’Boyle, UCC Computational Design of Molecular Wires for Solar Cells

Dr. Mary O’Connell, DCUEvolutionary Analysis of the Fanconi Anemia Pathway Dr. Mary O’Connell, DCUResolution of the Phylogeny of Hydrogenases

Prof. Noel O’Dowd, ULEffect of Residual Stress/Strain on High Temperature Creep of Polycrystals: A Multiscale Finite Element Analysis

Dr. Kevin O’Kelly, TCD Modeling of Indentation Fracture in Polycrystalline, Multi-Phase Ceramics

Dr. Orla O’Sullivan, Teagasc Metagenomics for Food and Health

Mr. Sergio Penkale, DCU Fragment Rescoring for Data-Oriented Translation

Mr. Luiz Pereira, TCD Charge Transport across Disordered Carbon Nanotube Networks

Dr. Dimitri Perrin, DCULarge-Scale Multi-Agent Simulations of Disease Spread in Large Human Populations

Dr. Andrew Phillips, UCD In Silico Evaluation of NacNac Transition Metal Catalyst Precursors

Mr. Stephen Power, TCDMagnetic and Electronic Properties of Disordered Graphene Systems

Dr. Estelle Roux, DIASJoint Inversion of Magnetotelluric and Surface-wave Data in an Anisotropic Earth

Prof. Heather Ruskin, DCUMulti-Layer Models of Epigenetic Mechanisms

Dr. Michael Scott, DCU ECC2K-130

Dr. Charles Spillane, UCC Identifying Taxon Restricted Genes in Higher Plants

Mr. Cathal Tomas, NUI, Galway The Optimisation of the Straightening Process for Stainless Steel Wires

Ms. Florentina Tofoleanu, UCD Structural and Energetic Properties of Alzheimer’s A-beta Amyloid Fibrils

Dr. Lamia Tounsi, DCUArabic Treebank-Based LFG parsing

Mr. Thomas Walsh, DCU Evolution of Tissue Specific Genes

Mr. Thomas Walsh, DCU The Evolution of a Placental Specific Regulatory Network.

Mr. Thomas Walsh, DCU Selection Analysis on Key interacting Partners of a Placental Regulatory Network


5 Appendices

5.1 Appendix 1 People and Partnerships

Membership of the Oversight Board

Dr. Dennis Jennings, ChairmanDr. Seán Baker, IONA TechnologiesProf. Richard Catlow, University College LondonProf. Luke Drury, Dublin Institute for Advanced StudiesProf. David Fegan, University College DublinProf. Martyn Guest, Cardiff UniversityProf. Terry Smith, National University of Ireland, Galway

Membership of the Science Council

PHYSICAL SCIENCESProf. Giovanni Ciccotti, University College DublinProf. Stephen Fahy, University College CorkDr. Jim Greer, Tyndall National InstituteProf. Miles Turner, Dublin City UniversityDr. Jiri Vala, National University of Ireland, MaynoothProf. Graeme Watson, Trinity College Dublin

LIFE SCIENCESDr. Gianluca Pollastri, University College DublinDr. Davide Pisani, National University of Ireland, MaynoothProf. Heather Ruskin (Chair), Dublin City UniversityProf. Denis Shields, University College Dublin

ENVIRONMENTAL SCIENCESProf. Peter Lynch, University College DublinMr. Ray McGrath, Met ÉireannProf. Chris Bean, University College Dublin

ENGINEERING SCIENCESProf. Peter McHugh, National University of Ireland, GalwayProf. Noel O’Dowd, University of Limerick

EXTERNAL ASSESSORProf. Ken Taylor, Queens University Belfast

5.2 Appendix 2 - Outputs

List of all ICHEC related Publications for the reporting period

Membership of the Users Council

CLASS ADr. Niall English, University College Dublin (Chair)Dr. Jiri Vala, National University of Ireland, Maynooth

CLASS BDr. Simon Elliot, Tyndall National InstituteDr. Giorgos Fagas, Tyndall National InstituteDr. Gemma Kinsella, National University of Ireland, MaynoothDr. Michael Nolan, Tyndall National InsitituteDr. Gareth O’Brien, University College DublinProf. Denis Shields, University College Dublin

CLASS CDr. Andrew Phillips, University College DublinDr. Pavle Mocilac, Dublin City University

e-INISProf. Luke Drury, Dublin Institute for Advanced Studies

Publication title Authors Journal, Volume and Page

Non-stoichiometric oxide and metal interfaces and reactions

Bennett R.A., Mulley J.S., Basham M., Nolan M., Elliott S.D. and Mulheran P.A.

Applied Physics A (2009) In press.

Dispersion analysis and computational efficiency of elastic lattice methods for seismic wave propagation

O’Brien G.S., Bean C.J. and Tapamo H.

Computers & Geosciences (2009) 35:1768-1775

Molecular Adsorption on the Doped Ceria (110) Surface

Nolan M. Journal of Physical Chemistry C (2009) 113: 2425-2432

Quantification of Ink Diffusion in Microcontact Printing with Self-Assembled Monolayers

Gannon G., Larsson J.A., Greer J.C. and Thompson D.

Langmuir (2009) 25: 242-247

Charge Transfer in Cr Adsorption and Reactions at the Rutile TiO2(110) Surface

Nolan M., Mulley J.S. and Bennett R.A.

Physical Chemistry Chemical Physics (2009) 11: 2156-2160

Small polarons in Nb- and Ta-doped rutile and anatase TiO2

Morgan B.J., Scanlon D.O. and Watson G.W.

Journal of Materials Chemistry (2009) 19:5175-5178

OpenDDA: a Novel High-Performance Computational Framework for the Discrete Dipole Approximation

McDonald J., Golden A. and Jennings S.G.

International Journal of High Performance Computing Applications (2009) 23: 42-61

Finite size effects in the Kitaev honeycomb lattice model on a torus

Kells G., Moran N. and Vala J.

Journal of Statistical Mechanics (2009) P03006

Healing of oxygen vacancies on reduced surfaces of gold-doped ceria

Nolan M.

Journal of Chemical Physics (2009) 130: 144702

Ballistic Conductance in Oxidized Si Nanowires

G. Fagas, J.C. Greer

Nano Lett. 9, 5 (2009) pp 1856-1860

Monolayer Packing, Dehydration, and Ink-Binding Dynamics at the Molecular Printboard

Gannon G., Larsson J.A. and Thompson D. The Journal of Physical Chemistry C (2009) 113:7298-7304

Lines of Evidence for Horizontal Gene Transfer of a Phenazine Producing Operon into Multiple Bacterial Species

Fitzpatrick D.A.

Journal of Molecular Evolution (2009) 68:171-185

Codon Size Reduction as the Origin of the Triplet Genetic Code

Baranov P.V., Venin M. and Provan G.

PLoS ONE (2009) 4:e5708

Bilingually Motivated Domain-Adapted Word Segmentation for Statistical Machine Translation

Ma Y. and Way A.

Proceedings of the 12th Conference of the European Chapter of the ACL, Athens, Greece (2009) pp. 549-557

Time reversal imaging of synthetic volcanic tremor sources

Lokmer, I., G. S. O’Brien, D. Stich, and C. J. Bean Geophys. Res. Lett., 36, L12308, doi:10.1029/2009GL038178

Volcano topography, structure and intrinsic attenuation: their relative influences on a simulated 3D visco-elastic wavefield

O’Brien G.S. and Bean C.J.

Journal of Volcanology and Geothermal Research (2009) 183:122-136

Modeling the population dynamics of antibiotic-resistant bacteria: an agent-based approach

Murphy J.T., Walshe R. and Devocelle M.

International Journal of Modern Physics C (2009) 20:435-457

Electrostatically immobilised BOX and PYBOX metal catalysts: application to ene reactions

McDonagh C. and O’Leary P.

Tetrahedron Letters (2009) 50:979-982

Investigation of the mechanical interaction of the trabecular core with an external shell using rapid prototype and finite element models

McDonnell P., Harrison N., Lohfeld S., Kennedy O., Zhang Y. and McHugh P.E.

Journal of the Mechanical Behavior of Biomedical Materials (2010) 3:63-76

Simulation of vertebral trabecular bone loss using voxel finite element analysis

McDonnell P., Harrison N., Liebschner M.A.K. and McHugh P.E.

Journal of Biomechanics, July 2009

Acceptor Levels in p-Type Cu2O: Rationalizing Theory and Experiment

Scanlon D.O., Morgan B.J., Watson G.W. and Walsh A.

Physical Review Letters (2009) 103:096405


Reactivity on the (110) Surface of Ceria: A GGA+U Study of Surface Reduction and the Adsorption of CO and NO2

Scanlon D.O., Galea N.M., Morgan B.J. and Watson G.W.

Journal of Physical Chemistry C (2009) 113:11095-11103

Intrinsic ferromagnetism in CeO2: dispelling the myth of vacancy site localization mediated superexchange

Keating P.R.L., Scanlon D.O. and Watson G.W.

Journal of Physics: Condensed Matter (2009) 21:405502

Ab initio calculation of the bias-dependent transport properties of Mn12 molecules

C.D. Pemmaraju, I. Rungger, and S. Sanvito

Phys. Rev. B, 80, 104422, (2009)

Simulating STM transport in alkanes from first principles

C. Toher, I. Rungger, and S. Sanvito

Phys. Rev. B, 79, 205427, (2009)

Resonant electronic states and I-V curves of Fe/MgO/Fe(100) tunnel junctions

I. Rungger, O. Mryasov, and S. Sanvito

Phys. Rev. B, 79, 094414, (2009)

Exploring the limits of the self-consistent Born approximation for inelastic electronic transport

W. Lee, N. Jean, and S. Sanvito

Phys. Rev. B, 79, 085120, (2009)

Comment on “Theoretical Description of Carrier Mediated Magnetism in Cobalt Doped ZnO”

Stefano Sanvito and C.D. Pemmara ju

Phys. Rev. Lett., 102, 159701, (2009)

Electron doping and magnetic moment formation in N- and C-doped MgO

Droghetti A. and Sanvito S.

Applied Physics Letters (2009) 94:252505

A spin of their own Szulczewski G., Sanvito S. and Coey M. Nature Materials (2009) 8:693-695

Electrostatic spin crossover effect in polar magnetic molecules

Baadji N., Piacenza M., Tugsuz T., Della Sala F., Maruccio G. and Sanvito S.

Nature Materials (2009) 8:813-817

EC-EARTH: A seamless earth system prediction approach in action

Hazeleger, W., Severijns, C., Semmler, T., Stefanescu, S., Yang, S., Wang, X., Wyser, K., Baldasano, J. M., Bintanja, R., Bougeault, P., Caballero, R., Dutra, E., Ekman, A. M. L., Christensen, J. H., van den Hurk, B., Jimenez, P., Jones, C., Kallberg, P., Koenigk, T., McGrath, R., Miranda, P., van Noije, T., Parodi, J. A., Schmith, T., Selten, F., Storelvmo, T., Sterl, A., Tapamo, H., Vancoppenolle, M., Viterbo, P., and Willen, U.

Bulletin of the American Meteorological Society, November 2009

Switching a Single Spin on Metal Surfaces by a STM Tip: Ab Initio Studies

Kun Tao, V. S. Stepanyuk, W. Hergert, I. Rungger, Sanvito and P. Bruno

Phys. Rev. Lett. 103, 057202 (2009)

Upper bound for the conductivity of nanotube networks

Pereira L.F.C., Rocha C.G., Latge A., Coleman J.N. and Ferreira M.S.

Applied Physics Letters (2009) 95:123106

Modeling the polaronic nature of p-type defects in Cu2O: The failure of GGA and GGA+U

Scanlon D.O., Morgan B.J. and Watson G.W.

Journal of Chemical Physics (2009) 131:124703

Understanding the p-Type Conduction Properties of the Transparent Conducting Oxide CuBO2: A Density Functional Theory Analysis

Scanlon D.O., Walsh A. and Watson G.W.

Chemistry of Materials (2009) 21:4568-4576

Tracking Relevant Alignment Characteristics for Machine Translation

Patrik Lambert, Yanjun Ma, Sylwia Ozdowska, Andy Way

Proc. of Machine Translation Summit XII. pp. 268-275. Ottawa, Canada.

A prediction of cell differentiation and proliferation within a collagen-glycosaminoglycan scaffold subjected to mechanical strain and perfusive fluid flow

AJF Stops, KB Heraty, M Browne, FJ O’Brien, PE McHugh

Journal of Biomechanics, November 2009, DOI: 10.1016/j.jbiomech.2009.10.037

Locating volcano-seismic signals in the presence of rough topography: wave simulations on Arenal volcano, Costa Rica

Mataxian J.P., O’Brien G.S., Bean C.J., Valette B. and Mora M.

Geophysical Journal International (2009) Published online in advance of print.

Thermostat Artifacts in Replica Exchange Molecular Dynamics Simulations

Edina Rosta, Nicolae-Viorel Buchete, Gerhard Hummer

Journal of Chemical Theory and Computation, 5:1393-1399 (2009)

Dynamical Models for the Formation of Elephant Trunks in H II Regions

Jonathan Mackey, Andrew J. Lim

Monthly notices of the Royal Astronomical Society

Context-dependent interaction leads to emergent search behavior in social aggregates

Colin Torney, Zoltan Neufeld and Iain D. Couzin

Proceedings of the National Academy of Sciences USA 2009 106:22055-22060 doi:10.1073/pnas.0907929106

Effect of La doping on CO adsorption at ceria surfaces

I. Yeriskin, M. Nolan

Journal of Chemical Physics 2009 vol 131 p. 244702 doi:10.1063/1.3271910

Polaronic trapping of electrons and holes by native defects in anatase TiO2

Benjamin J. Morgan and Graeme W. Watson

Physical Review B 80, 233102 (2009) 10.1103/PhysRevB.80.233102

Influence of turbulent advection on a phytoplankton ecosystem with nonuniform carrying capacity

William J. McKiver and Zoltan Neufeld

Physical Review E, vol. 79, Issue 6, id. 061902 DOI: 10.1103/PhysRevE.79.061902

Positional Information Generated by Spatially Distributed Signaling Cascades

Javier Munoz-Garcia, Zoltan Neufeld, Boris N. Kholodenko

PLoS Comput Biol 5(3): e1000330 doi:10.1371/journal.pcbi.1000330

Plankton bloom controlled by horizontal stirring

W. McKiver, Z. Neufeld, and I. Scheuring

Nonlin. Processes Geophys., 16, 623-630, 2009

Aggregation of chemotactic organisms in a differential flow

Javier Munoz-Garci¬a, Zoltan Neufeld

Physical Review E 80, 061902 (2009) DOI: 10.1103/PhysRevE.80.061902

Source geometry from exceptionally high resolution long period event observations at Mt Etna during the 2008 eruption

De Barros, L., C. J. Bean, I. Lokmer, G. Saccorotti, L. Zuccarello, G. S. O’Brien, J.-P. Mataxian, and D. Patana¨

Geophysical Research Letters 36, L24305 doi:10.1029/2009GL041273

Comparative study of bandwidths in copper delafossites from x-ray emission spectroscopy

Shin D, Foord J.S., Payne D.J., Arnold T., Aston D.J., Egdell R.G., Godinho K.G., Scanlon D.O., Morgan B.J., Watson G.W., Mugnier E., Yaicle C., Rougier A., Colakerol L., Glans P.A., Piper L.F.J. and Smith, K.E.

Physical Review B, vol. 80, Issue 23

(Cu2S2)(Sr3SC2O5)-A Layered, Direct Band Gap, p-Type Transparent Conducting Oxychalcogenide: A Theoretical Analysis.

Scanlon D.O. and Watson, G.W.

Chemistry of Materials, 21, 5435-5442

Carbon nanotube assisted water self-diffusion across lipid membranes in the absence and presence of electric fields

J.-A. Garate, N.J. English, J.M.D. MacElroy

Molecular Simulation, 35, pp. 3-12

First-principles calculation of nitrogen-tungsten codoping effects on the band structure of anatase-titania

R. Long and N.J. English

Applied Physics Letters, 94, 132102

Synergistic Effects of Bi/S Codoping on Visible Light-Activated Anatase TiO2 Photocatalysts from First Principles


The Journal of Physical Chemistry C, 2009, 113, pp 8373-8377

Mechanisms for thermal conduction in methane hydrate

N.J. English and J.S. Tse

Physical Review Letters, 103, 015901

Energetic and electronic properties of P doping at the rutile TiO2 (110) surface from first-principles


The Journal of Physical Chemistry C, 113, pp. 9423-9430

Estimation of zeta potentials of titania nanoparticles by molecular simulation

N.J. English and W.F. Long

Physica A, 388, pp. 4091-4096

Very Different Responses to Electromagnetic Fields in Binary Ionic Liquid-Water Solutions

N.J. English and D.A. Mooney

The Journal of Physical Chemistry B, 113, pp. 10128-10134

Non-equilibrium molecular dynamics study of electric and low-frequency microwave fields on hen egg white lysozyme

N.J. English, G.Y. Solomentsev and P. O’Brien

The Journal of Chemical Physics, 131, 035106

Electromagnetic field effects on binary dimethylimidazolium-based ionic liquid/water solutions

N.J. English and D.A. Mooney

Physical Chemistry Chemical Physics, 11, p. 9370

Band gap engineering of (N, Ta)-codoped TiO2 : A first-principles calculation


Chemical Physics Letters, 478, pp. 175-179

Molecular dynamics study of thermal-driven methane hydrate dissociation

N.J. English and G.M. Phelan

Journal of Chemical Physics, 131, 074704

First-Principles Study of S Doping at the Rutile TiO2 (110) Surface

R. Long, N.J. English and Y. Dai

The Journal of Physical Chemistry C, 113, pp. 17464-17470

Magnetic properties of first-row element-doped ZnS semiconductors: a density functional theory investigation


Physical Review B, 80, 115212

Static and alternating electric field and distance-dependent effects on carbon nanotube-assisted water self-diffusion across lipid membranes

J.-A. Garate, N.J. English and J.M.D. MacElroy

The Journal of Chemical Physics, 131, 114508

Mechanisms for thermal conduction in various polymorphs of methane hydrate

N.J. English, J.S. Tse and D. Carey

Physical Review B, 80, 134306


Media Coverage Recieved

Date Media Title URL

12/10/09 Irish Times The Business of Prediction http://www.irishtimes.com/newspaper/sciencetoday/2009/1210/1224260413004.html

23/04/09 RTE News Space: Irish Scientists make breakthrough

http://www.rte.ie/news/2009/0423/1news_av.html?2531013,null,230

01/03/09 Sunday Business Post

Stokes supercomputer boosts Irish scientists’ work

http://www.thepost.ie/post/pages/p/story.aspx-qqqt=MORE+COMPUTER+NEWS-qqqm=nav-qqqid=39845-qqqx=1.asp

27/02/09 Irish Times Supercomputer Stokes soaks up statistics to predict the weather

http://www.irishtimes.com/newspaper/finance/2009/0227/1224241883824.html

11/02/09 SiliconRepublic.com

Ireland’s fastest supercomputer deployed in just hours

http://www.siliconrepublic.com/news/article/12254/cio/irelands-fastest-supercomputer-deployed-in-just-hours

15/01/09 Irish Times Facing down the rising rivers http://www.irishtimes.com/newspaper/sciencetoday/2009/0115/1231738223219.html

5.3 Appendix 3The Condominium Cluster Model

“Local” HPC compute resources “Condominium” model

CONCEPTThere has hardly ever been a period in the development of IT services in the modern era where data intensive and processor intensive (i.e. high-end computing) computing are serviced from a technology landscape that has been changing so rapidly. New and emerging architectures based on GPUs, as well as the proliferation of services from cloud computing providers make planning for the future more challenging than it has ever been. And all of this without even considering the storing and management of data files with cumulative size growing at near-exponential rates, now well into the terabyte range and set to grow by close to an order of magnitude every 18 months.

How should we address these challenging as well as exciting developments at a Island level to ensure that we keep up with and benefit from these emerging trends while at the same time do it in a declining budgetary environment? We believe that the development of specially adapted condominiums , managed by ICHEC centrally in a co-located data centre, for all of the institutions that need high-end computer cycles should form an important part of the strategy. It is not suggested that these condos should replace local compute resources, but they should be seen as complementary to them. They should replace the need for institutions to procure, buy and manage their own high-end compute resources for small systems of the order of 100+ cores and where it remains essential for access to these facilities to remain under local control.

So what are condominium clusters, so-called by a number of US institutions now rolling out the model in order to reduce escalating hosting costs and ensure optimum use of resources increasingly constrained by tightening financial budgets?

Condominium clusters are clusters composed of resources owned by different institutions and administered and hosted centrally, by ICHEC in this instance, with the provision that spare cycles are made available across condominium boundaries. We have been testing the condo model for the last 18 months in UCD and NUIM. Both institutions are very satisfied with the way they have operated and both have now signed up for bigger condos in the recently commissioned Stokes cluster (a 3840-core SGI ICE cluster using Intel Westmere technology operational as of September 1st). Six condominiums are now in operation and managed by ICHEC for UCD, DCU, NUIM, NUIG, UL and DIAS.

THE WAY THE MODEL WORKS IS THIS:

towards the purchase of a much larger system, buying a share of the total system at a cost pro rata to the size of

the share it wishes to ‘own’ e.g. if ICHEC purchases a 4,000 core cluster, and a 150 core condo was being sought, then the once-off capital cost would be 150/4,000 of the total capital purchasing price. Note that there are likely to be significant cost savings from the purchase of a large system. In addition to the capital cost of cores, there will be a once-off cost for the associated storage.

where the institutional condos are seamless components. If/when problems occur with the hardware on the site (i.e. at the UCD data centre), UCD staff liaise with ICHEC staff to ensure minimum disruption. The complete administration service is provided free to the condo owner.

and a performance of c.40Tflops. ICHEC is offering a limited number of 50k (c.£42k) (capital) shares in Stokes, amounting to c. 96 cores with 8 nodes

a fixed boundary (i.e. a partition) or have access via an allocation model. Jobs in the partition model are limited to the size of the condo and the availability of sufficient cores to run a specific job. Jobs in the allocation model (currently used by all 6 institutions) are scheduled so that the average usage by the institution’s users corresponds approximately to the size of the institution’s condo.

the only additional costs for the institution are the monthly electricity costs charged pro rata to the size of the condo. For the Stokes cluster, this is approximately 0.5k/month or 6k/year for each 50k share (Stokes is housed in the UCD data center and electricity costs are charged at the UCD rates). Note in particular, there are no additional costs for systems support, software licenses, warranties.

BENEFITS

ICHEC can guarantee that the institution will get the desired outcome without the risks of possibly a “difficult” procurement and in the knowledge also that ICHEC’s deal with the vendor would almost certainly have a significantly better outcome in terms of cores per euro.

storage (only fraction of a “drawer” is purchased, not the full infrastructure), as well as on the infrastructure front which has already been purchased (management nodes, login nodes, interconnect, software licenses, etc.)

change in the future and a small pro-rata fee might be considered)

ANNUAL REVIEW 2009

record of 99.5% availability over the last 3 years

application software.

model rather than partition model for access

In summary, therefore, the condominium concept is a very attractive one for institutions and with its impressive saving of costs, it is a particularly compelling one in the present economic climate. PARTITION AND ALLOCATION MODELS FOR RESOURCE USAGE

An institution’s ownership of compute resources within the condominium model can be used in either of two main ways:1. A fixed size partition. The number of cores (N) owned by

the institution is rserved for users of that institution only. Only N cores can be used at any one time and no-one else is allowed access to these cores even if they are not in use.

2. A logical allocation of Core Hours. Each month, the institution gets an allocation (number of cores owned multiplied by number of hours in month) and this is then debited as jobs run. Jobs run in the same pool of resources as National Service projects but with an appropriate fairshare target.

The characteristics of the two options are:

PARTITION

Advantages

institution.

from the owning institution.

Disadvantages

are free outside

production jobs and so further partition the partition.

result in a lot of wasted cycles and longer turnaround times (see Figure 1).

RESOURCES ALLOCATION

Advantages

owned.

than are owned.

Disadvantages

ICHEC Annual Review 2009

Documents

Transcript of ICHEC Annual Review 2009