Understanding catalysts from first-principles:tackling challenges in modeling catalysts

Bryan R. Goldsmith

Fritz Haber Institute of the Max Planck SocietyBerlin, Germany

CH3ReO3 + H2O2 + cyclohexenein aqueous acetonitrile

dispersed metal ionson amorphous SiO2

TM

Si

O

Catalysis has a profound impact on everyday life

Value-added chemicals from biomassFertilizer

Synthesis of ammonia from its elementsFritz Haber, Nobel Lecture (1920)

Commodity chemical production Pharmaceutical industry

Y. Wang , S. De, N. Yan, Chem. Commun., 52, 6210 (2016)

Modeling can provide deep insights into catalysisGold nanoparticle on TiO2 for CO oxidation

[1] J. Saavedra, H. Doan, C. Pursell, L. Grabow, B. Chandler, Science 345 (2014)

ExperimentsElementary step kinetics

In-situ reactivityStructure

Chemisorption

TheoryTransition States

IntermediatesStructureDynamics

[1]

Computationalscreening

First-principles microkinetic modeling

Catalyst designCatalyst characterization

Computationalspectroscopy

First-principlesthermochemistry &kinetic parameters

Reaction mechanism

Mechanistic understanding

Active sites

The dream: computer-aided catalyst design

“Towards the computational design of solid catalysts” J. K. Nørskov et al., Nat. Chem. 1 (2009)

De

scripto

r

Many challenges remain for modeling catalysts

Representative models

Thermodynamics and kinetics with chemical accuracy

Complexity of solution

Liquid/solid interfaces

[1] G. Rupprechter and C. Weiland, Nano Today 2 (2007)[2] B. N. Zope, D. D. Hibbitts, M. Neurock, R. J. Davis, Science 330 (2010)

[2] [1]

length

time

f s p s n s μ s m s s hours years

Continuum Equations,Rate Equations

and Finite Element

Modeling

ElectronicStructureTheory

ab initioMolecularDynamics

ab initio

kinetic Monte Carlo

m

mm

μm

nm

density-

functional

theory

Multi-scale approach for modeling catalysts

Image courtesy of Karsten Reuter, TU Munich

Outline. Three projects that help address the below challenges

Challenge 2: Understanding the importance of solvent during homogeneous catalysis

Mo/SiO2 active site

Challenge 1: Modeling heterogeneity in reactivity of amorphous catalysts

Challenge 3: Finding reliable descriptors of catalysts (and materials)

CH3ReO3

ReO O

CH

HH

Mo/SiO2 active site

Challenge 1: Modeling heterogeneity in reactivity of amorphous catalysts

Outline. Three projects that help address the below challenges

Amorphous catalysts are importantbut not well-understood due to their disorder

H. S. Taylor, Proc. R. Soc. A 108 (1925)C. Yoon and D. Cocke, J. Non-Crystalline Solids 79 (1986)B. Peters and S. L. Scott, J. Chem. Phys. 142 (2015)

Mo(VI) complex on SiO2

Mo

Si

O

Dispersed metal ionson amorphous supports

Lack of well-defined structure

Diversity of catalyst site environments

Preparation dependent properties

Amorphous catalysts are importantbut not well-understood due to their disorder

H. S. Taylor, Proc. R. Soc. A 108 (1925)C. Yoon and D. Cocke, J. Non-Crystalline Solids 79 (1986)B. Peters and S. L. Scott, J. Chem. Phys. 142 (2015)

Challenges

Mo(VI) complex on SiO2

Mo

Si

O

Dispersed metal ionson amorphous supports

Lack of well-defined structure

Diversity of catalyst site environments

Preparation dependent properties

Amorphous catalysts are importantbut not well-understood due to their disorder

H. S. Taylor, Proc. R. Soc. A 108 (1925)C. Yoon and D. Cocke, J. Non-Crystalline Solids 79 (1986)B. Peters and S. L. Scott, J. Chem. Phys. 142 (2015)

Challenges

Open questions

How do typical active and dead sites differ?

How to sample the distribution of sites?

How to build representative models?Mo(VI) complex on SiO2

Mo

Si

O

Dispersed metal ionson amorphous supports

from structure to properties

Slab modelsCluster models

British Zeolite Association webpage http://www.bza.org/K. Reuter and M. Scheffler, Phys. Rev. B 65 (2001)P. E. Sinclair et al., J. Chem. Soc. Faraday Trans. 94 (1998)

Real system Model system

Cut, cap and constrainUnder periodic boundary conditions

Typical modeling protocols for crystalline catalysts

Typical modeling protocols for amorphous catalysts

A. Fong et al., ACS Catal. 5 (2015); J. Handzlik, J. Phys. Chem. 111 (2007)

Typical modeling protocols for amorphous catalysts

Approach 1. Assume one or two cluster models capture main features

A. Fong et al., ACS Catal. 5 (2015); J. Handzlik, J. Phys. Chem. 111 (2007)

Cr

Cr

Typical modeling protocols for amorphous catalysts

Approach 1. Assume one or two cluster models capture main features

Approach 2. Use a crystalline slab as amorphous support surrogate

A. Fong et al., ACS Catal. 5 (2015); J. Handzlik, J. Phys. Chem. 111 (2007)

Cr

Cr

Our approach. Combine small cluster models with a sequential quadratic programming algorithm

Goal: sample heterogeneity in reactivity of amorphous catalysts

Our approach. Combine small cluster models with a sequential quadratic programming algorithm

Mo/SiO2 cluster model

Goal: sample heterogeneity in reactivity of amorphous catalysts

Fluorine = constrained periphery atoms (xp)

Our approach. Combine small cluster models with a sequential quadratic programming algorithm

Mo/SiO2 cluster model

Goal: sample heterogeneity in reactivity of amorphous catalysts

Where is the best place

to constrain the periphery atoms?

Fluorine = constrained periphery atoms (xp)

Our approach. Combine small cluster models with a sequential quadratic programming algorithm

Mo/SiO2 cluster model

Goal: sample heterogeneity in reactivity of amorphous catalysts

Where is the best place

to constrain the periphery atoms?

Fluorine = constrained periphery atoms (xp)

Our approach. Combine small cluster models with a sequential quadratic programming algorithm

Mo/SiO2 cluster model

Goal: sample heterogeneity in reactivity of amorphous catalysts

Esite(xp) subject to E‡(xp) = E‡

Where is the best place

to constrain the periphery atoms?

Fluorine = constrained periphery atoms (xp)

Our approach. Combine small cluster models with a sequential quadratic programming algorithm

Mo/SiO2 cluster model

Goal: sample heterogeneity in reactivity of amorphous catalysts

Esite(xp) subject to E‡(xp) = E‡

Position of periphery atoms

Activation energy some valueEnergy of catalyst site

Esite(xp) subject to E‡(xp) = E‡

Sample the lowest energy catalyst sitefor each value of activation energy

‘Dial’ the activation energy

Inverse design: property to structure

Sequential QuadraticProgramming

New catalyst site structure

Step in activation energy

Esite(xp) subject to E‡(xp) = E‡

Sample the lowest energy catalyst sitefor each value of activation energy

E‡ xp

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

common

Activation energy,

Site

fo

rmat

ion

en

ergy

Example of algorithm on model energy landscape

New catalyst site structure

Step in activation energy

E‡ xp

Sample the lowest energy sitefor each value of E‡

B. R. Goldsmith et al., in Reaction Rate Constant Computations, Royal Society Chemistry (2013)

Ethene metathesis by isolated Mo(VI)/SiO2

Ethene metathesis by isolated Mo(VI)/SiO2

Metallacycle rotation

Trigonal bipyramidal Square pyramidalJ. Handzlik, J. Phys. Chem. 111 (2007)

Ethene metathesis by isolated Mo(VI)/SiO2

Metallacycle rotation

Trigonal bipyramidal Square pyramidalJ. Handzlik, J. Phys. Chem. 111 (2007)

Examined structure-sensitivityusing 10 large models

Examine structure-sensitivity toward metallacycle rotation

Six fluorine atoms = xp = constrained atoms for reactant, transition state, and product

E‡

B. R. Goldsmith, E. D. Sanderson, D. Bean, B. Peters, J. Chem. Phys. 138 (2013)

Metallacycle rotationTrigonal bipyramidal Square pyramidal

Our small cluster model for use with the algorithm

dead sites

active × abundant= important

Rotation barrier, E‡ (kJ/mol)

Site

fo

rmat

ion

ener

gy (

kJ/m

ol)

Wide range in the metallacycle rotation barrier exists(E‡ = 28 to 72 kJ/mol)

E‡

dead sites

active × abundant= important

Rotation barrier, E‡ (kJ/mol)

Site

fo

rmat

ion

ener

gy (

kJ/m

ol)

Wide range in the metallacycle rotation barrier exists(E‡ = 28 to 72 kJ/mol)

E‡

10

sla

b m

od

els dead sites

active × abundant= important

Rotation barrier, E‡ (kJ/mol)

Site

fo

rmat

ion

ener

gy (

kJ/m

ol)

Wide range in the metallacycle rotation barrier exists(E‡ = 28 to 72 kJ/mol)

E‡

“compressed”active

“stretched” inactive

Rotation barrier, E‡ (kJ/mol)

Si-S

i dis

tan

ce

(An

gstr

om

s)

Identify structure-activity descriptors of catalyst sites

E‡

rSi-Si

E‡

Compare one simple cluster model to five large slab models

C. Ewing et al., Ind. Eng. Chem. Res. 55 (2016)

Five large slab models

One small cluster model

E‡

Compare one simple cluster model to five large slab models

Small cluster model captures thetrend and ≈90% of the variance

C. Ewing et al., Ind. Eng. Chem. Res. 55 (2016)

Symbolsslab models

Black line our algorithm

Si-Si Distance (Angstroms)

Ro

tati

on

bar

rier

, E

‡(e

V)

Five large slab models

One small cluster model

Outline. Three projects that help address the below challenges

Challenge 2: Understanding the importance of solvent during homogeneous catalysis

CH3ReO3

ReO O

CH

HH

Homogeneous vs. amorphous catalysts

Homogeneous catalystsAmorphous catalytic solidsSingle active siteStructurally well-definedAmenable to spectroscopySolvent environment

Diversity of active sites Lack of long-range orderLess amenable to spectroscopy

A. Moses J. Am. Chem. Soc. 129 (2007); J. Pelletier and J. Basset, Acc. Chem. Res. 49 (2016)

Homogeneous vs. amorphous catalysts

Homogeneous catalystsAmorphous catalytic solidsSingle active siteStructurally well-definedAmenable to spectroscopySolvent environment

Diversity of active sites Lack of long-range orderLess amenable to spectroscopy

CH3ReO3/SiO2-Al2O3

CH3ReO3

in CH3CN(aq)

A. Moses J. Am. Chem. Soc. 129 (2007); J. Pelletier and J. Basset, Acc. Chem. Res. 49 (2016)

Active alkene metathesis catalyst Not active for alkene metathesis

ReReRe

Al

Si

O O

C

HHH

O

C

Homogeneous vs. amorphous catalysts

Homogeneous catalystsAmorphous catalytic solidsSingle active siteStructurally well-definedAmenable to spectroscopySolvent environment

Diversity of active sites Lack of long-range orderLess amenable to spectroscopy

CH3ReO3/SiO2-Al2O3

CH3ReO3

in CH3CN(aq)

A. Moses J. Am. Chem. Soc. 129 (2007); J. Pelletier and J. Basset, Acc. Chem. Res. 49 (2016)

Active alkene metathesis catalyst Very active for olefin epoxidation

ReReRe

Al

Si

O O

C

HHH

O

C

“Greening” the Ethylene Oxide Process

C2H4 + 1

2O2 C2H4O (+ 2 CO2, 15 %) replace

Ag/a-Al2O3

“Greening” the Ethylene Oxide Process

C2H4 + 1

2O2 C2H4O (+ 2 CO2, 15 %)

C2H4 + H2O2 C2H4O + H2O → C2H4(OH)2

replace

by

Ag/a-Al2O3

“Greening” the Ethylene Oxide Process

M. Ghanta et al. Ind. Eng. Chem. Res., 52 (2013)

C2H4 + 1

2O2 C2H4O (+ 2 CO2, 15 %)

C2H4 + H2O2 C2H4O + H2O → C2H4(OH)2

replace

by

Ag/a-Al2O3

CH3ReO3 (MTO)

Homogeneous catalyst

C2H4O selectivity ≈ 100 %No H2O2 decompositionWorks with higher olefins (e.g., C3H6)

CH3ReO3

J. H. Espenson, Chem. Comm. 479 (1999)W. A. Herrmann, R. W. Fischer, J. D. G. Correia, J. Mol. Catal. 94 (1994)

+

Methyltrioxorhenium activates H2O2

for oxygen atom transfer

J. H. Espenson, Chem. Comm. 479 (1999)W. A. Herrmann, R. W. Fischer, J. D. G. Correia, J. Mol. Catal. 94 (1994)

+

Methyltrioxorhenium activates H2O2

for oxygen atom transfer

ⱡ

J. H. Espenson, Chem. Comm. 479 (1999)W. A. Herrmann, R. W. Fischer, J. D. G. Correia, J. Mol. Catal. 94 (1994)

+

Methyltrioxorhenium activates H2O2

for oxygen atom transfer

ⱡ

J. H. Espenson, Chem. Comm. 479 (1999)W. A. Herrmann, R. W. Fischer, J. D. G. Correia, J. Mol. Catal. 94 (1994)

+

Methyltrioxorhenium activates H2O2

for oxygen atom transfer

ⱡⱡ

J. H. Espenson, Chem. Comm. 479 (1999)W. A. Herrmann, R. W. Fischer, J. D. G. Correia, J. Mol. Catal. 94 (1994)

+

Methyltrioxorhenium activates H2O2

for oxygen atom transfer

ⱡⱡ

J. H. Espenson, Chem. Comm. 479 (1999)W. A. Herrmann, R. W. Fischer, J. D. G. Correia, J. Mol. Catal. 94 (1994)

+

Methyltrioxorhenium activates H2O2

for oxygen atom transfer

The ‘clean’ spectra of MTO makes it amenable to kinetic studies

1H NMR spectra recorded at 3 minute intervals, 23.0 °C

T. Hwang*, B. R. Goldsmith*, B. Peters, S. L. Scott, Inorg. Chem. 52 (2013)

MTO ⇌ A ⇌ B

MTO

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

MTO to A

Reaction parameters: H1, S1, G1

Activation parameters: H1‡,S1

‡, G1‡

A to B

H2, S2, G2

H2‡,S2

‡, G2‡

ThermodynamicsCalculated Experimental

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

MTO to A

Reaction parameters: H1, S1, G1

Activation parameters: H1‡,S1

‡, G1‡

A to B

H2, S2, G2

H2‡,S2

‡, G2‡

ThermodynamicsCalculated ExperimentalH1 > 0 H1 < 0

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

MTO to A

Reaction parameters: H1, S1, G1

Activation parameters: H1‡,S1

‡, G1‡

A to B

H2, S2, G2

H2‡,S2

‡, G2‡

ThermodynamicsCalculated ExperimentalH1 > 0 H1 < 0S1 < 0 S1 > 0

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

MTO to A

Reaction parameters: H1, S1, G1

Activation parameters: H1‡,S1

‡, G1‡

A to B

H2, S2, G2

H2‡,S2

‡, G2‡

ThermodynamicsCalculated ExperimentalH1 > 0 H1 < 0S1 < 0 S1 > 0G1 > 0 G1 < 0

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

MTO to A

Reaction parameters: H1, S1, G1

Activation parameters: H1‡,S1

‡, G1‡

A to B

H2, S2, G2

H2‡,S2

‡, G2‡

ThermodynamicsCalculated ExperimentalH1 > 0 H1 < 0S1 < 0 S1 > 0G1 > 0 G1 < 0G1 > G2 G1 < G2

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

MTO to A

Reaction parameters: H1, S1, G1

Activation parameters: H1‡,S1

‡, G1‡

A to B

H2, S2, G2

H2‡,S2

‡, G2‡

ThermodynamicsCalculated ExperimentalH1 > 0 H1 < 0S1 < 0 S1 > 0G1 > 0 G1 < 0G1 > G2 G1 < G2

Many discrepancies remain between experiment and theory

P. Gisdakis et al., Angew. Chem. Int. Ed. 37 (1998)J. M. Gonzales, et al. J. Am. Chem. Soc., 129 (2007)B. R. Goldsmith et al., J. Am. Chem. Soc. 137 (2015)

KineticsCalculated ExperimentalH1

‡ > 100 kJ mol-1 H1‡ = 24.5 kJ mol-1

MTO to A

Reaction parameters: H1, S1, G1

Activation parameters: H1‡,S1

‡, G1‡

A to B

H2, S2, G2

H2‡,S2

‡, G2‡

+ H2O2 + H2OCH3ReO3

15 °C, CD3CN(aq)

Trace water can be important in many reactions

The water dependence of MTO has not been explained

+ H2O2 + H2OCH3ReO3

15 °C, CD3CN(aq)

X = 1 – exp(-kobst)

Trace water can be important in many reactions

The water dependence of MTO has not been explained

Goals1) Fully characterize the thermodynamics and kinetics 2) Elucidate the importance of water

B. R. Goldsmith, T. Hwang, S. Seritan, B. Peters, S. L. Scott, J. Am. Chem. Soc. 137 (2015)

Methods

Density-functional theory

Microkinetic modeling

Experimental kinetic measurements

cyclohexene cyclohexene

Cyclohexene epoxidation step has only weak water acceleration

0.0

0.2

0.4

0.6

0.8

1.0

0

2

4

6

8

10

0 1 2 3 4 5

10

2 k

3a

pp /

M-1

s-1

10

2 k4

ap

p / M-1 s

-1

[H2O]

0 / M

Cyclohexene epoxidation by B

Cyclohexene epoxidation by A

Experiments

Cyclohexene epoxidation step has only weak water acceleration

0.0

0.2

0.4

0.6

0.8

1.0

0

2

4

6

8

10

0 1 2 3 4 5

10

2 k

3a

pp /

M-1

s-1

10

2 k4

ap

p / M-1 s

-1

[H2O]

0 / M

Cyclohexene epoxidation by B

Cyclohexene epoxidation by A

Gⱡ = 118 kJ/mol Gⱡ = 115 kJ/mol

Gⱡ = 96 kJ/mol Gⱡ = 94 kJ/mol

epoxidation by A

epoxidation by B

Experiments

Theory

Water dramatically accelerates the formation of species A and B, not the epoxidation step

At= A

¥+a(1-e

-kfastt) + be

-kslowt

CH3ReO3 + H2O2 + H2OUV-vis, 25.0 °C, CH3CN

fast formationof mostly A

slower formationof mostly B

Free energy profiles for the formation of A and B

free

en

ergy

, kJ

mo

l-1

reaction coordinate

-14

-12

CH3ReO3

+ 2 H2O2 A + H2O2

+ H2O

experimental

Free energy profiles for the formation of A and B

B + H2O

free

en

ergy

, kJ

mo

l-1

reaction coordinate

-14

-12

7582

CH3ReO3

+ 2 H2O2 A + H2O2

+ H2O

experimental

Free energy profiles for the formation of A and B

B + H2O

free

en

ergy

, kJ

mo

l-1

reaction coordinate

-14

-12

7582

CH3ReO3

+ 2 H2O2 A + H2O2

+ H2O

experimental 110121

reaction coordinate

-9-5CH3ReO3

+ 2 H2O2A + H2O2

+ H2O B + H2O

calculated

0H2O

0H2OFree energy profiles for the formation of A and B

B + H2O

free

en

ergy

, kJ

mo

l-1

reaction coordinate

-14

-12

7582

CH3ReO3

+ 2 H2O2 A + H2O2

+ H2O

experimental 110121

reaction coordinate

-9-5CH3ReO3

+ 2 H2O2A + H2O2

+ H2O B + H2O

calculated

0H2O

0H2OFree energy profiles for the formation of A and B

B + H2O

free

en

ergy

, kJ

mo

l-1

reaction coordinate

-14

-12

7582

CH3ReO3

+ 2 H2O2 A + H2O2

+ H2O

experimental 110121

reaction coordinate

-9-5CH3ReO3

+ 2 H2O2A + H2O2

+ H2O B + H2O

calculated

0H2O

0H2O

91

97

1H2O

1H2O

Free energy profiles for the formation of A and B

B + H2O

free

en

ergy

, kJ

mo

l-1

reaction coordinate

-14

-12

7582

CH3ReO3

+ 2 H2O2 A + H2O2

+ H2O

experimental 110121

reaction coordinate

-9-5CH3ReO3

+ 2 H2O2A + H2O2

+ H2O B + H2O

calculated

0H2O

0H2O

91

97

1H2O

1H2O96

94

2H2O

2H2O

Free energy profiles for the formation of A and B

B + H2O

The full catalytic cycle and the importance of water

B cycle

A cycle

The full catalytic cycle and the importance of water

B cycle

A cycle

experimental kinetic profiles

computed kinetic profiles

Outline. Three projects that help address the below challenges

Challenge 3: Finding reliable descriptors of catalysts (and materials)

Computationalscreening

First-principles microkinetic modeling

Catalyst designCatalyst characterization

Computationalspectroscopy

First-principlesthermochemistry &kinetic parameters

Reaction mechanism

Mechanistic understanding

Active sites

Can we reliably extract descriptors from such catalysts?

De

scripto

r

Amorphous catalysts Homogeneous catalysts

Identifying physically meaningful descriptors can help materials discovery

Identifying physically meaningful descriptors can help materials discovery

Descriptor = function(atomic or material features)

Descriptor → Property

Identifying physically meaningful descriptors can help materials discovery

Predict new materials

Increase understanding

Descriptor = function(atomic or material features)

Descriptor → Property

Identifying physically meaningful descriptors can help materials discovery

J. K. Nørskov et al. Nat. Chem. 1, 37 (2009)

CO + 3H2 ⇌ CH4 + H2O

Predict new materials

Increase understanding

Descriptor = function(atomic or material features)

Descriptor → Property

On (111) facets

Identifying physically meaningful descriptors can help materials discovery

CO + 3H2 ⇌ CH4 + H2O

Predict new materials

Increase understanding

Descriptor = function(atomic or material features)

Descriptor → Property

On (111) facets

The development of data-mining tools can facilitate the discovery of descriptors

J. K. Nørskov et al. Nat. Chem. 1, 37 (2009)

material property 1 (y1)

mat

eria

l pro

per

ty 2

(y 2

)Subgroup discovery: find meaningful local descriptors of a target property in materials-science data

Review: M. Atzmueller, WIREs Data Min. Knowl. Discov. 5 (2015)

material property 1 (y1)

mat

eria

l pro

per

ty 2

(y 2

)Subgroup discovery: find meaningful local descriptors of a target property in materials-science data

Review: M. Atzmueller, WIREs Data Min. Knowl. Discov. 5 (2015)

material property 1 (y1)

mat

eria

l pro

per

ty 2

(y 2

) The periodic table has subgroups

Subgroup discovery: find meaningful local descriptors of a target property in materials-science data

Review: M. Atzmueller, WIREs Data Min. Knowl. Discov. 5 (2015)

Find descriptors that predict crystal structuresfor the 82 octet AB-type materials

Rocksalt Zincblende

vsTarget property

sign(Erocksalt – Ezincblende)

B. R. Goldsmith, M. Boley, J. Vreeken, M. Scheffler, L. Ghiringhelli, New J. Phys. 19, (2017)

Find descriptors that predict crystal structuresfor the 82 octet AB-type materials

Rocksalt Zincblende

vsTarget property

sign(Erocksalt – Ezincblende)

B. R. Goldsmith, M. Boley, J. Vreeken, M. Scheffler, L. Ghiringhelli, New J. Phys. 19, (2017)

Input candidate descriptors into subgroup discovery from DFT calculations• Radii of s, p, d orbitals of free atoms • Electron affinity• Ionization potential….and many others

𝑟pA − 𝑟p

B

𝑟 sA

Subgroup discovery classifies 79 of the 82 compoundsusing a two-dimensional descriptor

𝒓𝐩𝐀 − 𝒓𝐩

𝐁 ≤ 𝟏. 𝟏𝟔 Å

and 𝒓𝐬𝐀 ≤ 𝟏. 𝟐𝟕 Å

𝑟pA − 𝑟p

B

𝑟 sA

Subgroup discovery classifies 79 of the 82 compoundsusing a two-dimensional descriptor

zincblende subgroup

𝒓𝐩𝐀 − 𝒓𝐩

𝐁 ≥ 𝟎. 𝟗𝟏 Å

and 𝒓𝐬𝐀 ≥ 𝟏. 𝟐𝟐 Å

𝒓𝐩𝐀 − 𝒓𝐩

𝐁 ≤ 𝟏. 𝟏𝟔 Å

and 𝒓𝐬𝐀 ≤ 𝟏. 𝟐𝟕 Å

𝑟pA − 𝑟p

B

𝑟 sA

Subgroup discovery classifies 79 of the 82 compoundsusing a two-dimensional descriptor

zincblende subgroup

Rocksalt

subgroup

𝒓𝐩𝐀 − 𝒓𝐩

𝐁 ≥ 𝟎. 𝟗𝟏 Å

and 𝒓𝐬𝐀 ≥ 𝟏. 𝟐𝟐 Å

𝒓𝐩𝐀 − 𝒓𝐩

𝐁 ≤ 𝟏. 𝟏𝟔 Å

and 𝒓𝐬𝐀 ≤ 𝟏. 𝟐𝟕 Å

𝑟pA − 𝑟p

B

𝑟 sA

Subgroup discovery classifies 79 of the 82 compoundsusing a two-dimensional descriptor

zincblende subgroup

Rocksalt

subgroup

MgTe AgI

CuF

Subgroup discovery applied to gas-phase gold clusters

Examine 24,400 gold cluster configurations of sizes Au5-Au14

Gold clusters have interesting chemical and electronic properties

Snapshots of stable gold cluster configurations

B. R. Goldsmith, M. Boley, J. Vreeken, M. Scheffler, L. M. Ghiringhelli, New J. Phys. 19, (2017)M. Boley, B. R. Goldsmith, L. M. Ghiringhelli, J. Vreeken, arXiv:1701.07696, (2017)

Rediscover simple insight about HOMO-LUMO gap

HO

MO

-LU

MO

en

ergy

gap

(e

V)

24,400 gold cluster configurations (Au5-Au14) in the gas phase 2,440 configurations per size

(relative) total energy of gold cluster (eV)

Target property HOMO-LUMO energy gap

Three subgroups regarding HOMO-LUMO gap are found

N = # atoms in gold cluster

(relative) total energy of gold cluster (eV)HO

MO

-LU

MO

en

ergy

gap

(e

V)

𝐞𝐯𝐞𝐧 𝑵 𝐚𝐧𝐝 𝑵 > 𝟕

𝐨𝐝𝐝(𝑵)

𝑵 = 𝟔

Target property HOMO-LUMO energy gap

Three subgroups regarding HOMO-LUMO gap are found

N = # atoms in gold cluster

(relative) total energy of gold cluster (eV)HO

MO

-LU

MO

en

ergy

gap

(e

V)

𝐞𝐯𝐞𝐧 𝑵 𝐚𝐧𝐝 𝑵 > 𝟕

𝐨𝐝𝐝(𝑵)

𝑵 = 𝟔

Metallic

Semiconducting

Target property HOMO-LUMO energy gap

Three subgroups regarding HOMO-LUMO gap are found

N = # atoms in gold cluster

(relative) total energy of gold cluster (eV)HO

MO

-LU

MO

en

ergy

gap

(e

V)

𝐞𝐯𝐞𝐧 𝑵 𝐚𝐧𝐝 𝑵 > 𝟕

𝐨𝐝𝐝(𝑵)

𝑵 = 𝟔

Metallic

Semiconducting

Target property HOMO-LUMO energy gap

Toward accelerating computer-aided catalyst design

Plans to apply machine learning to other systems

Finding descriptors of amorphous catalysts

Perovskite structure prediction and reactivity

Designing templates for organometallic catalysts

[1] C. Büchner et al., J. Non-Cryst. Solids 435 (2016)

[1]

Acknowledgements

Baron Peters

Susannah L. Scott

Matthias Scheffler

Wei-Xue Li

Theory Department,The Fritz Haber InstituteLuca M. Ghiringhelli

Mario Boley

Taeho Hwang

and many others!

Chemical Engineering Department, UC Santa Barbara

Subgroup discovery tutorials and other analytics tools are online

https://www.nomad-coe.eu/