Situated @: Bath; Birkbeck; Cambridge; CCLRC Daresbury Reading The Royal Institution

Situated @:

Bath;Birkbeck; Cambridge;CCLRC DaresburyReadingThe Royal InstitutionUniversity College London (UCL)

eMinerals one of NERCs eScience testbed projects

The eMinerals team:

Environmental scientists;Chemists; Physicists;Computational andGrid scientists.

PI:Martin Dove ([email protected])

Web:www.eminerals.org

eMinerals: Science Outcomes enabled by new Grid ToolseMinerals: Science Outcomes enabled by new Grid Tools

Maria AlfredssonMaria AlfredssonNottingham 21/9/2005Nottingham 21/9/2005


Research undertaken by:Bath group:A. Marmier, D.J. Cooke, S.C. ParkerBirkbeck group:Z. Du and N.H. de LeeuwCambridge group:K. Trachenko, E. Artacho, J.M Pruneda,M.T. DoveDaresbury group:I. Todorov and W. SmithRI group:M. Blanchard and K. WrightUCL group:M. Alfredsson, J.P. Brodholt andG.D. Price


Environmental ProcessesEnvironmental ProcessesAIM:we use computational modelling to research mineralogical processes at an atomistic level, providing information on transport and immobilisation processes of pollutants, including both toxic elements (.i.e. As, Cd, Pb and organic molecules) as well as radioactive waste.We have also looked alternative energy resources to fossil fuels.

Sources of pollution e.g.:• Acid mine drainage• Land filling sites• Industries and farming• Accidents with toxics• Natural catastrophes or mineralogical properties


Environmental ProcessesEnvironmental Processes

Problem:Relastic models of mineral process are computationally very expensive.

Layout:Layout:• Grid ResourcesGrid Resources• Data ManagementData Management• Science OutcomesScience Outcomes

Solution:GRID COMPUTING

• LLaakekess (Bath, Cambridge, UCL): (Bath, Cambridge, UCL): 4 linux-based clusters4 linux-based clusters 88 nodes in total with 2Gb memory per node88 nodes in total with 2Gb memory per node

• Pond (Cambridge): 1 Pond (Cambridge): 1 Apple Xserve cluster Apple Xserve cluster 8 nodes with 8Gb memory per node8 nodes with 8Gb memory per node

• 24-node IBM cluster (Reading)24-node IBM cluster (Reading)

• 33 CCoonnddoorr--ppoooollss:: UCL > 900 machinesUCL > 900 machines Cambridge (25 machines)Cambridge (25 machines) BathBath

NNGGSS –– CCSSAARR -- HHPPCCxx

Grid Resources:

Resources marked in red suitablefor first principles code green represents resources suitable forinter-atomic potential codes.

• Storage Resource Broker (SRB)Storage Resource Broker (SRB)Bath, Cambridge, Reading and the central MCAT at Bath, Cambridge, Reading and the central MCAT at DaresburyDaresbury

• Chemical Markup Language (CML)Chemical Markup Language (CML)-version of XML adapted for chemical applications-version of XML adapted for chemical applications-All codes developed in eMinerals support CML-All codes developed in eMinerals support CML

• Personal Interface Grid (PIG)Personal Interface Grid (PIG)

• MAST MAST

Data Management

• WIKIWIKI

• RcommandsRcommands

• MetadataMetadata

Job Submission:

• Globus Globus (GSI/X.509-certificaes)(GSI/X.509-certificaes)

• Condor-GCondor-G

• SeagullSeagull

Computer Codes:

Submit jobs from all machines fromour work station.

Maintained and developed with eMinerals:• DL_Poly• Metadise – Monte Carlo implemented• Siesta• Casino Other Codes:

• Gulp • Marvin

• AbInit• Casino• VASP• Crystal

automatic meta-scheduler to submit to the “most appropriate”machine in the mini-grid.

Dagman and Perl scripts


Science Outcome:Science Outcome:• Surface and InterfacesSurface and InterfacesDetermine water exchange and diffusion coefficientDetermine water exchange and diffusion coefficientEffect of impuritesEffect of impurites

• Phase TransitionsPhase Transitionsdue to compositional and pressure effectsdue to compositional and pressure effectsLattice dynamics calculations to determine most stableLattice dynamics calculations to determine most stablepolymorphpolymorph

• Radioactive wasteRadioactive waste

Aim:To fully understand transport and immobilisation processes of contaminants we need an accurate description of the mineral/solvent interfaces.

Solution:We perform Molecular Dynamics simulations using the DL_POLY code.

Snapshot of Goethite/Solvent interface using MD-simulation on the HPCx. A. Marmier, D. Cooke, S. Kerisit and S.C. Parker Bath University.

Mineral/Solvent Interfaces

Computer resources:Condor-pool - distributing many independent calculations over the machines available, using Dagman or Perl scripts good statistical data, which can be used to determine diffusion and water exchange coefficients.NGSHPCx – larger jobs

Mineral/Solvent Interfaces

Result: • Ordering of the water molecules close to mineral surface.• Cl- ions order closer to the mineral surface than Na+ ions

• The classical modelsof the electrical doublelayer do not describecorrectly the ion distribution close to thesurface.

A. Marmier, D.J. Cooke, S. Kerisit and S.C. Parker

Bath University.

Pt/Graphite interfacePt/Graphite interface

• Graphite: Model for Graphite: Model for organic substrateorganic substrate

• Pt/Graphite: Alternative Pt/Graphite: Alternative (renewable) energy (renewable) energy resource to fossil fuels resource to fossil fuels know to generate green know to generate green house gases.house gases.

• Graphite: Model for Graphite: Model for organic substrateorganic substrate

• Pt/Graphite: Alternative Pt/Graphite: Alternative (renewable) energy (renewable) energy resource to fossil fuels resource to fossil fuels know to generate green know to generate green house gases.house gases.A. Marmier and

S.C. Parker at University of Bath


Aim: Derive highly quality empirical potentialsfrom density functional theory (DFT) calcualtions.

Problem: Computational costly

Solution: Grid computing - NGS

Aim: Derive highly quality empirical potentialsfrom density functional theory (DFT) calcualtions.

Problem: Computational costly

Solution: Grid computing - NGS

A. Marmier andS.C. Parker at University of Bath

Conclusions:• Most stable site is located on a

bridge site

• The activation barrier is 0.5 eV

• The adsorption sites and energiesare different for inter-atomicpotential

calculations

Conclusions:• Most stable site is located on a

bridge site

• The activation barrier is 0.5 eV

• The adsorption sites and energiesare different for inter-atomicpotential

calculations


A. Marmier andS.C. Parker at University of Bath

CaO-termimated

TiO2-termimated

{001} surfaces of CaTiO3

Mineral Surfaces

M. Alfredsson, J.P. Brodholt and G.D. PriceUCL

Calculations:• investigate 10-20 surfaces• 2 to 5 surface terminations• 4 to 16 impurity positions• > 4 concentrations

Total number of calculationsper impurity: 120-2440

Computer Resources:• Condor Cluster • SRB

We defined a new methodto calculate surface energies which allow us to determine crystal particle shape. We find particle shapes change with concentration of the impurity and the type of dopant.

Important to understand the reactivity and inter- actions between pollutants and minerals.

Mineral Surfaces

increasing concentration

In all mineral processes we are dealing with impurities, which may changes the crystal structures

Phyllosilicates (layered silicate minerals, including clays) are known to adsorb and store toxic elements.

Here we show how the crystal structure of layeredLi2Si2O5 transforms (‘breaks up’) in the presence of different elements, e.g. Cs.

Z. Du and N. H. de LeeuwBirkbeck College and UCL

Compositional Phase Transitions

LiCs

Na

Computational Resources:• Condor Pools• Eminerals mini-grid• SRB

Z. Du and N. H. de Leeuw: Birkbeck College and UCL

Compositional Phase Transitions

Li

Na

Processes Entalphy

(kJ/mol)

-30.9

-26.2

-37.9

-40.9

2084420844 OSiLiMOSiLiK

)()( 2084320844 aqKOSiNaLiKaqNaOSiLiK

)()( 20842220843 aqKOSiLiNaKaqNaOSiNaLiK

)()( 20843208422 aqKOSiLiKNaaqNaOSiLiNaK

)()( 2084420843 aqKOSiLiNaaqNaOSiLiKNa

Results:Solid solutions of guest ions in silicates are often thermodynamically stable.• Cation exchange from solution is an endothermic process; only K-Na exchange expected to occur

Pyrite (Fools gold): Pyrite (Fools gold): FeSFeS22Fe-bearing minerals active role in the control of acid mine

drainage and transport of heavy metals like As.

Transport and imobilisation process:• Pyrite may contain ca. 10wt% of As• Adsorption of As on Pyrite surface

Aim: understanding electronicstructure and bonding propertiesof pure pyrite. Possible phasetransitions?

Method: linear respons phononcalculations, using DFT

Computational resources:HPCx linking back to the SRBs

M. Blanchard and K. Wrightat the RI

Pyrite (Fools gold): Pyrite (Fools gold): FeSFeS22Results:

• Pyrite is an insulator (in agreement with experiment)• Pyrite is described by S2 molecules interacting with Fe ions

Conclusions: • Calculated frequencies are in good agreement with experiment• All vibrational modes show non-linear pressure dependence• Mode Grüneisen parameters give information about thermodynamical properties

M. Blanchard and K. Wrightat the RI

Pressure Induced Phase diagrams: MgO and FeO

Expt.1) HF-AE* HF-PP** QMC-PP**

a (Å)

B0 (GPa)

4.194.19

157157

4.1954.195

184184

4.0894.089

196196

4.0944.094

178178

1) M. I. McCarthy et al PRB (1994) and ref. therein

*AE=All-electron**PP=Pseudo-potential

Note: The PP used in the HF and QMCNote: The PP used in the HF and QMCcalculations is the same.calculations is the same.

Problem:QMC calculations are ca. 1000 times more computer intensive than traditional first principles calculations.

Solution:HPCx – the CASINO code show excellent scaling

Problem:Traditional DFT techniques often fail in reproducing Fe-bearing minerals

Solution:Quantum Monte Carlo (QMC) calculationsHybrid-DFT calculations

by UCL-team

PPTT calculated from H calculated from HB1B1=H=HB2B2; Birch-Murnaghan 3; Birch-Murnaghan 3rdrd order EOS order EOS

Transition Pressure (PTransition Pressure (PTT) B1 to B2: QMC) B1 to B2: QMC

Result:QMC and LDA (with the same PP) give similar results

PPTT ~ 597GPa ~ 597GPa

B1B1

Method

GGA-PAWGGA-PP(PW)GGA-PP(PW)LDA-LAPWLDA-PP(PW)

QMC-PPQMC-PPLDA-PP (PW)LDA-PP (PW)

5975972020569569

509489664510451

P(GPa)

Oganov et al JPC 2003 and ref. thereinThis work

PP=Pseudo-potential

B2B2

Observeration: We consumed ca. 200.000 Cpu Hrs

by UCL-team

P(GPa)

r-B1(AFM)r-B1(AFM) i-B8(AFM)i-B8(AFM) B8(NM)B8(NM)

83 145insulatorinsulator insulatorinsulator metallicmetallic

Phase Diagram and Crystal StructuresPhase Diagram and Crystal Structures

TNéel =193 K

P~115 GPaat T=0K

Fei & Mao, Science (1994)

To determine phase To determine phase transitions we need transitions we need to:to:• optimise the optimise the geometries for all the geometries for all the possible crystal possible crystal structures at various structures at various pressures. pressures. ~ 240 ~ 240 calculations for FeOcalculations for FeO• for up to 10 for up to 10 computational computational methods methods (Hamiltonians)(Hamiltonians) ~240 x 10 = ~2400 ~240 x 10 = ~2400 calculationscalculations

Solution:Solution:• Condor cluster @UCLCondor cluster @UCL• SRBSRB

Solution:Solution:• Condor cluster @UCLCondor cluster @UCL• SRBSRB

Aim: Find alternative to QMCSolution: Hybrid-DFT

by UCL-team

Radioactive WasteRadioactive WasteNuclear waste disposal – encapsulation in ceramic materials

Aim:Find the best waste form to be used to immobilise surplusPu and high-radiation waste (hrw)

Problem:Most of the currently considered waste forms are damaged(amorphorised) by irradiation from hrw

K. Trachenko, M.T. DoveI. Todorov and W. Smith

Radioactive WasteRadioactive WasteK. Trachenko, M.T. DoveI. Todorov and W. Smith

Observation of amorphisation in Zircon

Radioactive WasteRadioactive WasteNuclear waste disposal – encapsulation in ceramic materials

Aim:Find the best waste form to be used to immobilise surplusPu and high-radiation waste (hrw)

Problem:• Most of the currently considered waste forms are damaged(amorphorised) by irradiation from hrw.• Amorphisation requires large computational system sizes

Code development:DL_Poly5 million atoms using the HPCx


SiO2

GeO2

TiO2

Al2O3

MgO

Radioactive waste

Result: The more ionic properties the ceramics showthe faster healing processes are observed.

Increasing ionicity

Evolution of time


Snapshot of MD-generatedstructures caused by 40 keV U recoil.

Prior the eMinerals: • project the data presented here would take several years, involving many projects.

• many of the calculations on realistic systems were also out of reach, such as the modelling of the electrical double layer at the solvent/mineral interface, and the radiation damage, using more than 5 millions ions in the simulation.

Future:•“team projects”• automatic work flows for job submission and data analysis.

Level of theory

Adsorbing surface

Contaminant

Quantum Monte Carlo

Large empirical models

Linear-scaling quantum mechanics

Organic molecules

HalogensMetallic elements

Cla

ys,

mic

as

Alu

min

osili

cate

s

Nat

ura

l org

anic

mat

ter

Pho

sph

ates

Car

bona

tes

Oxi

des/

hydr

oxid

es

Sul

phi

des

Acknowledgement:Acknowledgement:The “Eminerals team”The “Eminerals team”

NERC for financial supportNERC for financial support


Web:

www.eminerals.org

Documents

Situated @: Bath; Birkbeck; Cambridge; CCLRC Daresbury Reading The Royal Institution