Nanotech2010 High Throughput

30
Materials d high throug G. Fitzgerald, G. Go A. Perlov, Accelrys M. Sarwar, S. Frenc M. Sarwar, S. Frenc D. Thompsett, John TechConnect WORLD Confer discovery with ab initio ghput calculations oldbeck-Wood, J.L. Gavartin, s Software Inc ch, S. Garçia, A. Martinez-Bonastre, ch, S. Garçia, A. Martinez-Bonastre, nson Matthey rence, June 2010, Anaheim, CA

description

Presentation to TechConnect World 2010. High-throughput approaches for fuel cell development

Transcript of Nanotech2010 High Throughput

Page 1: Nanotech2010 High Throughput

Materials discovery with ab initio high throughput calculations

G. Fitzgerald, G. Goldbeck

A. Perlov, Accelrys Software Inc

M. Sarwar, S. French, S. Garçia, A. MartinezM. Sarwar, S. French, S. Garçia, A. Martinez

D. Thompsett, Johnson Matthey

TechConnect WORLD Conference, June 2010, Anaheim, CA

Materials discovery with ab initio high throughput calculations

G. Fitzgerald, G. Goldbeck-Wood, J.L. Gavartin,

A. Perlov, Accelrys Software Inc

M. Sarwar, S. French, S. Garçia, A. Martinez-Bonastre, M. Sarwar, S. French, S. Garçia, A. Martinez-Bonastre,

D. Thompsett, Johnson Matthey

TechConnect WORLD Conference, June 2010, Anaheim, CA

Page 2: Nanotech2010 High Throughput

Statement

• Materials discovery typically involves exploration of very large phase space.

• Quantum and atomistic simulation technology are

sufficiently mature to be used to screen materials and

come up with lead candidates.

Bulk and surface defects Alloying

Defect decoration Surface segregation

© 2009 Accelrys, Inc.

come up with lead candidates.

• Robust, yet flexible software infrastructures make it feasible to

– Screen materials even with compute-heavy simulation methods.

– Provide a knowledge and decision base for research teams.

Materials Studio

Materials discovery typically involves exploration of very large phase space.

Quantum and atomistic simulation technology are

sufficiently mature to be used to screen materials and

Clustering

Skin formation

2

Robust, yet flexible software infrastructures make it feasible to

heavy simulation methods.

Provide a knowledge and decision base for research teams.

Pipeline Pilot Client

Pipeline Pilot

Web Port

Materials Studio Collection

Page 3: Nanotech2010 High Throughput

Fuel Cell Catalyst application

• Market problem: Fuel Cells issues include

– High Cost

– Limited stability

• Technology problems

– Catalytic breakdown of O2 is rate determining

– Catalysis relies on Platinum

– other metals do not to perform well enough.

• R&D problems: Finding alternatives to Pt

– Different alloy combinations

© 2009 Accelrys, Inc.

– Different alloy combinations

– Different alloy microstructures

– i.e. Screening thousands of materials

• Modeling problem: High throughput & relevance

– Complex enumeration of periodic structures

– Long calculation times

– Property analysis relevant to experiment

is rate determining

other metals do not to perform well enough.

3

Modeling problem: High throughput & relevance

Complex enumeration of periodic structures

Page 4: Nanotech2010 High Throughput

Overall:½O2+H2 �

PEM Fuel Cells

© 2009 Accelrys, Inc.

Anode- hydrogen oxidation:

H2 2H+ + 2e-

� H2O+ electricity

4

Cathode- oxygen reduction:

½O2 +2H+ + 2e- H2O

Page 5: Nanotech2010 High Throughput

1

1.2

1.4

Cell Potential [V]

Theoretical Cell Voltage [E° = 1.23 V]

Theoretical cell voltage: 1.23V

Actual much less due to various loss processes

Voltage losses in a PEMFC

© 2009 Accelrys, Inc.

0

0.2

0.4

0.6

0.8

0 200 400 600 800

Current Density [mA/cm2]

Cell Potential [V]

Cathode Activation

Theoretical Cell Voltage [E° = 1.23 V]

Actual much less due to various loss processes

5

1000 1200 1400

Current Density [mA/cm2]

Cell Resistance

Anode Activation

Mass Transport

Cell Performance

Page 6: Nanotech2010 High Throughput

Adsorption of O2 Dissociation of O

Dissociative ORR mechanism in PEMFC

© 2009 Accelrys, Inc.

e-

Combination with a second proton +e

….to form OH

e-

Dissociation of O2 Combination with a proton +e-….

Dissociative ORR mechanism in PEMFC

6

Combination with a second proton +e-….

….to form H2O

Page 7: Nanotech2010 High Throughput

Ideal Catalyst – Sabatier’s Principle

• The optimum catalyst-adsorbate interaction must

– Not be too weak:

• Chemical bonds between surface and adsorbate can be formed

• Internal bonds weakened so intermediates can be generated

– Not be too strong

• Intermediates generated can react further

• Desorbtion takes place and freeing up of adsorption sites

• Pt is close to optimum, but expensive

© 2009 Accelrys, Inc.

Catalytic

Activity

• Pt is close to optimum, but expensive

– Other pure metals perform worse

– What about alloys?

Sabatier’s Principle

adsorbate interaction must

Chemical bonds between surface and adsorbate can be formed

Internal bonds weakened so intermediates can be generated

Intermediates generated can react further

Desorbtion takes place and freeing up of adsorption sites

7

Metal-adsorbate bond strength

Pt

Page 8: Nanotech2010 High Throughput

Phase Space

Activity

© 2009 Accelrys, Inc.

With high throughput modeling,that would otherwise not be covered

0% - Pt 50% - Pt100% - Co 50% - Co

Phase Space

8

modeling, phase space to be exploredcovered experimentally.

100% - Pt0% - Co

Page 9: Nanotech2010 High Throughput

Sampling must include structure and composition

Bulk and surface defects Alloying

© 2009 Accelrys, Inc.

Defect decoration Surface segregation

Sampling must include structure and composition

Alloying Clustering

9

Surface segregation Skin formation

Page 10: Nanotech2010 High Throughput

Overview of approach

• Objective: find alloy with similar profile as Pt, slightly weaker bonding

• Generate representative catalyst surface models

• Calculate key properties and determine descriptors of ORR

activity and bonding:

– Material Stability (Segregation Energies)

© 2009 Accelrys, Inc.

– Material Stability (Segregation Energies)

– O adsorption Energies

– OH adsorption Energies

– d-band centres

– electronic workfunction

– thermodynamics of reaction steps

• Methodology:

– Plane Wave Density Functional Theory – CASTEP

– High throughput strategies, automation

– Data accumulation and storage in database

– Data Management and reporting

Catalytic

Activity

Metal-adsorbate bond strength

Ptfind alloy with similar profile as Pt, slightly weaker bonding

Generate representative catalyst surface models

Calculate key properties and determine descriptors of ORR

10

CASTEP

Data accumulation and storage in database

Work function

Page 11: Nanotech2010 High Throughput

High throughput calculations: Overview

© 2009 Accelrys, Inc. 11

Page 12: Nanotech2010 High Throughput

AxB1-x ensemble generation

Constrained atoms

1st layer `A`

2nd layer `B`

3rd layer `C`

4th layer `D`

A0B2C2D0

© 2009 Accelrys, Inc.

A1

A2

A3

A4

B1

B2

B3

B4

C1

C2

C3

C4

D1

D2

D3

D4

0 0 0 0 1 1 0 0 0 0 1 1 0 0 0 0

)!(!

!

kNk

N

k

N

−=

For a given supercell generate all

configurations with k out of N host

sites substituted by X atomsA0B2C2D0

12

Define a configuration class by the

number of substitutions in each layer

Apply symmetry transformations

and find degeneracy of irreducible

configurations

Map each structure into a unique

binary string

Page 13: Nanotech2010 High Throughput

Example Pt3Co: 6 layers slab setup

N=16, k=4

configurations

133 – non-equivalent configurations (due to symmetry)

Pt not stable on surface of Pt

A0 superclass contains

1820)!416(!4

!16=

© 2009 Accelrys, Inc.

A0 superclass structures

A0 superclass contains

Class Conf. x gi Class Conf. x gi Class

A0B4C0D0 1x1 A0B3C1D0 1x12+1x4 A0B2C2D0

A0B3C0D1 1x12+1x4 A0B2C1D1

A0B2C0D2

configurations

equivalent configurations (due to symmetry)

Pt not stable on surface of Pt3Co -> only consider A0 superclass

A0 superclass contains 42 irreducible configurations

13

A0 superclass structures

A0 superclass contains 42 irreducible configurations

Conf. x gi Class Conf. x gi Class Conf. x gi

2x6+1x24 A0B1C3D0 1x4+1x12 A0B0C4D0 1x1

4x12+2x24 A0B1C2D1 4x12+2x24 A0B0C3D1 1x12+1x4

2x6+1x24 A0B1C1D2 4x12+2x24 A0B0C2D2 2x6+1x24

A0B1C0D3 1x12+1x4 A0B0C1D3 1x12+1x4

A0B0C0D4 1x1

Page 14: Nanotech2010 High Throughput

Pt3Co: most stable configurations

© 2009 Accelrys, Inc.

E = E0

ba

E = E0 + 0.03 eV E = E0+ 0.04 eV

~11 configurations give significant contribution into the TD average.

Entropic factor may be important

999.0

96.0

/)(

11,1

11

/)(

4,1

4

0

0

==

==

−−

=

−−

=

∑TkEE

j

j

TkEE

j

j

Bj

Bj

egZ

egZ

14

c d

E = E0+ 0.04 eV E = E0+0.07 eV

~11 configurations give significant contribution into the TD average.

J. L. Gavartin et al., Transactions of the

Electrochemical Society 25(1) 1335-1344 (2009).

Page 15: Nanotech2010 High Throughput

Surface phase diagram

• stoichiometry near the surface may significantly deviate from its nominal bulk value.

Understanding surface phase diagram is critical for

– assessing catalytic reactivity

– testing electrochemical stability

Pd3Co: The thermodynamic Co fraction in the

second and third layers is

Pt Co : The variation of Co fraction between the

thermodynamic fraction fk of metal X in the layer k

© 2009 Accelrys, Inc.

Pt3Co : The variation of Co fraction between the

layers is somewhat less pronounced.

Shuo Chen et al Am. Chem. Soc./ 2008, 130, 13818

stoichiometry near the surface may significantly deviate from its nominal bulk value.

Understanding surface phase diagram is critical for

Co: The thermodynamic Co fraction in the

second and third layers is overstoichiometric;

Co : The variation of Co fraction between the

15

Co : The variation of Co fraction between the

layers is somewhat less pronounced.

Shuo Chen et al Am. Chem. Soc./ 2008, 130, 13818

Page 16: Nanotech2010 High Throughput

Effect of lattice strain on oxygen adsorption

)(2

1)()( 2

* OEsurfaceEsurfaceOEEad −−+=

Compressive lateral strain leads to

approximately linear decrease of the

Oxygen adsorption energy, while lattice

expansion leads to an increase of Eads(O).

A similar trend was reported earlier for

Cu(111), Ru(0001) and Au(111) surfaces.

© 2009 Accelrys, Inc.

Cu(111), Ru(0001) and Au(111) surfaces.

Y. Xu, Surf. Sci., 494, 131-144 (2001).

M. Mavrikakis, Phys. Rev. Lett., 81, 2819 (1998).

M. Mavrikakis, Catal. Lett., 64, 101 (2000).

O adsorption on most stable (Pt/Pd)3Co

surfaces suggests some additional efects

besides strain, e.g. chemical modification

and surface relaxation.

Effect of lattice strain on oxygen adsorption

expansion leads to an increase of Eads(O).

16

surfaces suggests some additional efects

besides strain, e.g. chemical modification

Page 17: Nanotech2010 High Throughput

High throughput materials calculations

© 2009 Accelrys, Inc. 17

Page 18: Nanotech2010 High Throughput

Creating the discovery database

© 2009 Accelrys, Inc. 18

Page 19: Nanotech2010 High Throughput

Reporting and mining the database

© 2009 Accelrys, Inc. 19

Page 20: Nanotech2010 High Throughput

Calculations database: Portal

Access via a Web Portal:

1. Choose the database name (local or remote)

2. Choose composition of a material of interest

3. Choose report tables to show

© 2009 Accelrys, Inc.

1. Choose the database name (local or remote)

2. Choose composition of a material of interest

20

Page 21: Nanotech2010 High Throughput

Calculations database: systems overview

Statistics for all selected jobs: Run time and CPUs used

Systems categorized as bulk, slab or molecule and shown by chemical composition

Choose the chemical composition of interest

© 2009 Accelrys, Inc.

Calculations database: systems overview

Statistics for all selected jobs: Run time and CPUs used

and shown by chemical composition

21

Page 22: Nanotech2010 High Throughput

Calculations database: Table of runs for a system

Sortable tables display key job information (a single calculation per line):

Choose the specific individual calculation to see all its details

© 2009 Accelrys, Inc.

Calculations database: Table of runs for a system

Sortable tables display key job information (a single calculation per line):

Choose the specific individual calculation to see all its details

22

Page 23: Nanotech2010 High Throughput

Calculations database: Job details

The details are given in different tables, that may be shown or hidden

Jmol visualiser for structure viewing, manipulation and analysis

© 2009 Accelrys, Inc.

The details are given in different tables, that may be shown or hidden

Jmol visualiser for structure viewing, manipulation and analysis

23

Page 24: Nanotech2010 High Throughput

Results details: Workfunction and d-band centre

© 2009 Accelrys, Inc.

band centre

24

Page 25: Nanotech2010 High Throughput

Oxygen Reduction Reaction

© 2009 Accelrys, Inc.

E

E0=E(O2+*)

ETS=E(O*-O*)

25

Reaction coordinate

E1=E(O2*)

E2=2E(O*)

Ediss=E2-E1 Eads1=E1-E0

Ea=ETS-E1 Eads2=E2-E0

Eads1=E1-E0

iCatDesign

Page 26: Nanotech2010 High Throughput

Reaction free energy protocol

•Calculate free energy of main steps of dissociative ORR reaction as

a function of cell potential U

On input:

File with energies

Cell Potential U

Substrate

O adsorbed

OH adsorbed

© 2009 Accelrys, Inc.

1. J.K. Nørskov, J. Rossmeisel, A. Logadottir, et al.J. Phys. Chem. B,108, 17866

OH adsorbed

H2O and H2 gas phase

Correction energies

(-ST, ZPE, solvation)

Reaction free energy protocol

Calculate free energy of main steps of dissociative ORR reaction as

26

1. J.K. Nørskov, J. Rossmeisel, A. Logadottir, et al.J. Phys. Chem. B,108, 17866-17892 (2004).

Page 27: Nanotech2010 High Throughput

Reaction free energy protocol

© 2009 Accelrys, Inc. 27

Page 28: Nanotech2010 High Throughput

Summary

• A series of tools and models have been developed for screening various Pt alloy

combinations for stability and activity.

• Tools developed can be used to screen Pt and non

• ORR activity is improved by lattice strain (compressive for Pt)

• Strain is controlled by alloying with base metals antisegregating away from the

surface (typically metals with smaller atomic radius)

• D-band centre, surface segragation and adsorption energies are found to be useful

descriptors

© 2009 Accelrys, Inc.

descriptors

A series of tools and models have been developed for screening various Pt alloy

Tools developed can be used to screen Pt and non-Pt compositions

ORR activity is improved by lattice strain (compressive for Pt)

Strain is controlled by alloying with base metals antisegregating away from the

surface (typically metals with smaller atomic radius)

band centre, surface segragation and adsorption energies are found to be useful

28

Page 29: Nanotech2010 High Throughput

Conclusions: High Throughput Calculations

• Materials Discovery beyond the trivial case requires a vast number of systems to be studied, each requiring HPC computing itself.

• Successful projects require collaboration of different organizations and modeling with experiment.

• Require new strategies allowing for large multicollaborative projects providing automation in – optimization of calculations – strict quality control

© 2009 Accelrys, Inc.

– strict quality control– streamlined analysis and reporting– Data storage and mining in a flexible, extensible database

• Pipeline Pilot and the Materials Studio Collection have been used to– Build a framework for Materials HTC

• Automated job generation, execution and analysis• Datamining via web portal

– Screen >2000 catalyst leads and identify candidates for development– Provide a source for project teams to explore.

Conclusions: High Throughput Calculations

Materials Discovery beyond the trivial case requires a vast number of systems to be studied, each requiring HPC computing itself.

Successful projects require collaboration of different organizations and

Require new strategies allowing for large multi-platform and multi-user collaborative projects providing automation in

29

Data storage and mining in a flexible, extensible database

Pipeline Pilot and the Materials Studio Collection have been used toBuild a framework for Materials HTC

Automated job generation, execution and analysis

Screen >2000 catalyst leads and identify candidates for developmentProvide a source for project teams to explore.

Page 30: Nanotech2010 High Throughput

Acknowledgements

• Dan Ormsby

• Amity Andersen

• David Gunn

• Arek Krzystala

• Victor Milman

• Patricia Gestoso-Suoto

J. L. Gavartin, M. Sarwar, D. C. Papageorgopoulos, D. Gunn, S. Garcia,

A. Perlov, A. Krzystala, D. L. Ormsby, F. Liu, G. Goldbeck

Andersen, S. French, D. Thompsett. Exploring fuell cell cathode

materials: High throughput calculation approach.

Transactions of the Electrochemical Society

© 2009 Accelrys, Inc.

Funding:

This project is partly funded under Technology Strategy Board Project Number:

/5/MAT/6/I/H0379C. The TSB is a business

established by the government. Its mission is to promote and support research into, and

development and exploitation of technology and innovation for the benefit of UK

business, in order to increase economic growth and improve the quality of life. It is

sponsored by the Department for Innovation, Universities and Skills (DIUS) (22).

J. L. Gavartin, M. Sarwar, D. C. Papageorgopoulos, D. Gunn, S. Garcia,

A. Perlov, A. Krzystala, D. L. Ormsby, F. Liu, G. Goldbeck-Wood, A.

Andersen, S. French, D. Thompsett. Exploring fuell cell cathode

materials: High throughput calculation approach.

Transactions of the Electrochemical Society 25(1) 1335-1344 (2009).

30

This project is partly funded under Technology Strategy Board Project Number:

/5/MAT/6/I/H0379C. The TSB is a business-led executive non-departmental public body,

established by the government. Its mission is to promote and support research into, and

development and exploitation of technology and innovation for the benefit of UK

business, in order to increase economic growth and improve the quality of life. It is

sponsored by the Department for Innovation, Universities and Skills (DIUS) (22).