Materials Genome Initiative for Soft Matter: Computational ... … · 1 Materials Genome Initiative for Soft Matter: Computational Challenges Peter T. Cummings 1Department of Chemical

1

Materials Genome Initiative for Soft Matter: Computational Challenges

Peter T. Cummings1Department of Chemical & Biomolecular Engineering, Vanderbilt University2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory

Materials Genome Initiative Grand Challenges Summit: Soft MaterialsInstitute for Bioscience & Biotechnology Research, Rockville, Maryland

September 19-20, 2013Center for Nanophase Materials Sciences!" #!$ %&'() *!"&#*!+ +!,#%!"#%-

Simulation-Based Engineering and Science

Activity 2008-2010 International comparative study

Chaired by Sharon Glotzer– Life sciences and medicine– Materials– Energy and sustainability

Strategic directions workshop Chaired by Peter Cummings

One of key precursor studies for MGI

2

http://www.nsf.gov/mps/ResearchDirectionsWorkshop2010/RWD-color-FINAL-usletter_2010-07-16.pdf

Soft Matter

Wikipedia definition “[S]ubfield of condensed matter comprising a variety of physical states that

are easily deformed by thermal stresses or thermal fluctuations” Examples: Liquids, colloids, polymers, foams, gels, granular materials, some

biomaterials Common feature: Predominant physical behaviors occur at energy scale

comparable with kBT – Quantum aspects generally unimportant– Weak dispersion interactions frequently dominate

Can be difficult to predict properties directly from atomic or molecular constituents

– Soft matter often self-organizes into mesoscopic physical structures that are much larger than microscopic scale yet much smaller than the macroscopic scale

• Macroscopic properties are result of mesoscopic sructure

Contrast with hard materials– Energy scales large by comparison to kBT – Relatively easy to predict properties since atoms/molecules are typically in crystalline

lattice with no mesoscopic structuring

3

MGI for Soft Matter

Characteristics of soft matter make MGI-like screening very difficult To accomplish

computationally, we typically have to perform multiple steps, depending on nature of system Simplest case: relatively simple liquid with no mesoscopic structure

– From chemistry, gather or derive forcefields– Initialize simulation (molecular dynamics or Monte Carlo, as appropriate)– Equilibrate simulation– Perform production runs of simulations– Statistically average to obtain properties

4

Chemistry and conditions Properties

Case in Point

Room-temperature ionic liquids Ionic liquids (i.e., molten salts) made of organic ions whose ability to

crystallize is frustrated by organic groups Relevant in multiple energy technologies

Energy-efficient separations, carbon capture, energy storage as electrolytes for supercapacitors

5

C O NH S F

C4mpip+

Tf2N-

Ionic Liquids: Designer Solvents for a Cleaner World, James F. Wishart, President’s FY 2005 Budget Request to Congress

Case in Point: RTILs RTILs

~ 9 billion possible RTILs Structured on nanoscale but not mesoscale (based on SAXS and SANS)

Atomistic description is appropriate and sufficient

Perfect candidates for screening? E.g., for solubility of carbon dioxide

Problems: Only ~100 forcefields available for RTILs

Hence, to implement MGI-like screening of RTILs needs automated forcefield development capability

Contrast with screening of metal-organic frameworks 137,953 hypothetical MOF structures generated Each MOF evaluated for methane storage at 35 bar

and 298 K using grand canonical Monte Carlo ~300 MOFs had higher methane-storage capacity at

35 bar than current world record

6

Wilmer, et al., Large-scale screening of hypothetical metal–organic frameworks. Nature Chemistry, 4 (2011) 83-89. doi:10.1038/nchem.1192

Self Assembly in Soft Matter

Both RTILs and MOFs share simplifying feature that atomistic description is appropriate No mesoscopic ordering in RTIL, crystalline (macroscopic) structure for

MOFs

What about soft matter systems that self-assemble into mesoscopic structures that determine macroscopic properties? Examples

Amphiphilic molecules/polymers Lipids that assemble into bilayers

Technical challenges Time to self-assemble Scale of mesoscopic structures

Usual approach Coarse-graining

– Systematic, ad hoc, models that exist only at the coarse-grained level

7

Coarse-Graining8

Noid, W. G., “Perspective: Coarse-grained models for biomolecular systems,” JCP, 139 (2013) 090901

NCBL!

DOXY!

OXY!

ALKE!

Superposition of fully atomistic and coar se-gra ined models of n-dodecane. A 3:1 mapping is used, where each CG bead represents 3 CH2 groups or 2 CH2 groups and a CH3 group

Mapping of fully atomistic model of skin lipid to CG model. Courtesy of Clare McCabe

E.g., protein folding

Self-Assembling Soft Matter

Goal

For simple case... E.g., ionic liquids

9

Chemistry Atomistic forcefield(s)

gather

derive

Initialize atomistic simulation

PropertiesEquilibrate atomistic simulation

Production runs

Statistical averaging


Self-Assembling Soft Matter

Goal

For simple case... E.g., ionic liquids

10


gather

derive


PropertiesEquilibrate atomistic simulation

Production runs



Self-Assembling Soft Matter Bottom-up coarse-grained properties estimation

11


gather

derive


Properties

Equilibrate atomistic simulation

Production runs


Inversion to atomistic

CG forcefield


CG production

runs



gather

derive


Properties


Production runs



CG forcefield


CG production

runs



gather

derive


Properties


Production runs



CG forcefield


CG production

runs

GPU computing


14


gather

derive


Properties


Production runs



CG forcefield


CG production

runs

Self-Assembling Soft Matter Within this component of overall workflow...

Specialized methods to accelerate calculations Monte Carlo methods (e.g., configurational bias) Parallel tempering Rare events techniques ....

Specialized methods to accommodate self assembly Constant stress ensemble ....

Specialized property sampling methods E.g., free energy methods


Production runs


16


gather

derive


Properties


Production runs



CG forcefield


CG production

runs

Self-Assembling Soft Matter Multiple methods for obtaining CG forcefield

Structure matching E.g., iterated Boltzmann inversion*

– D. Reith, H. Meyer, and F. Müller-Plathe, Macromolecules 34 (2001) 2335

Force matching S. Izvekov and G. A. Voth, J. Phys. Chem. B 109 (2005)

2469; J. Chem. Phys. 123 (2005) 134105

Choice of CG method strongly coupled to properties to be calculated CG simulations cannot reproduce all properties of atomistic system

Method should preserve properties of interest


CG forcefield

*Chan, E. R.; Striolo, A.; McCabe, C.; Cummings, P. T. Glotzer, S. C., “Coarse-grained force field for simulating polymer-tethered silsesquioxane self-assembly in solution,” JCP 127 (2007) 114102

Challenges

MGI-like computational screening for soft matter relies on complex scientific workflows How to automate? One acceleration scenario, assuming transferable CG forcefields

Chemistry

Properties


CG forcefield


Equilibrate CG simulation

CG production

runs

Initialize CG

simulationgather

Challenges

Coarse-graining often means running many similar atomistic simulations Data management = reusability

Hybrid materials....

The three V’s Validation, validation, validation

Atomic, molecular and nanoscale experimental probes (especially X-ray and neutron sources at DOE and NIST) provide capabilities for validation of computational approaches, particularly to nanoscale structure and dynamics

19

Hard materials Soft materialsHybrid materials

Met

al-o

rgan

ic

fram

ewor

ks

Poly

hedr

al

olig

imer

ic

sils

esqu

ioxa

nes

Challenges

Interfaces

Critically important in multiple applications Energy storage, catalysis, separation processes,... Combines issues of soft and hard materials, often synergistically

– Fluid Interface Reactions Structure and Transport (FIRST) Energy Frontier Research Center

20

Hard materials Soft materials

Nano-porous electrodes Micropores (<2nm) increase specific surface area Anomalous increase of capacitance is observed experimentally

Understanding Supercapacitors21

Largeot, C.; Portet, C.; Chmiola, J.; Taberna, P.-L.; Gogotsi, Y.; Simon, P. J. Am. Chem. Soc. 2008, 130

REVIEW ARTICLE

nature materials | VOL 7 | NOVEMBER 2008 | www.nature.com/naturematerials 849

(zone III) were !tted with equation (5). For micropores (<1 nm), it was assumed that ions enter a cylindrical pore and line up, thus forming the ‘electric wire in cylinder’ model of a capacitor. Capacitance was calculated from43

ln0

r 0 (6)/ = (6)

where a0 is the e"ective size of the ion (desolvated). #is model perfectly matches with the normalized capacitance change versus pore size (zone I in Fig. 4). Calculations using density functional theory gave consistent values for the size, a0, for unsolvated NEt4

+ and BF4

– ions43.#is work suggests that removal of the solvation shell is required

for ions to enter the micropores. Moreover, the ionic radius a0 found by using equation (6) was close to the bare ion size, suggesting that ions could be fully desolvated. A study carried out with CDCs in a

solvent-free electrolyte ([EMI+,TFSI–] ionic liquid at 60 °C), in which both ions have a maximum size of about 0.7 nm, showed the maximum capacitance for samples with the 0.7-nm pore size46, demonstrating that a single ion per pore produces the maximum capacitance (Fig. 5). #is suggests that ions cannot be adsorbed on both pore surfaces, in contrast with traditional supercapacitor models.

MATERIALS BY DESIGN#e recent !ndings of the micropore contribution to the capacitive storage highlight the lack of fundamental understanding of the electrochemical interfaces at the nanoscale and the behaviour of ions con!ned in nanopores. In particular, the results presented above rule out the generally accepted description of the double layer with solvated ions adsorbed on both sides of the pore walls, consistent with the absence of a di"use layer in subnanometre pores. Although recent studies45,46 provide some guidance for developing materials with improved capacitance, such as elimination of macro- and mesopores and matching the pore size with the ion size, further material optimization by Edisonian or combinatorial electrochemistry methods may take a very long time. #e e"ects of many parameters, such as carbon bonding (sp versus sp2 or sp3), pore shape, defects or adatoms, are di$cult to determine experimentally. Clearly, computational tools and atomistic simulation will be needed to help us to understand the charge storage mechanism in subnanometre pores and to propose strategies to design the next generation of high-capacitance materials and material–electrolyte systems47. Recasting the theory of double layers in electrochemistry to take into account solvation and desolvation e"ects could lead to a better understanding of charge storage as well as ion transport in ECs and even open up new opportunities in areas such as biological ion channels and water desalination.

REDOX-BASED ELECTROCHEMICAL CAPACITORS

MECHANISM OF PSEUDO-CAPACITIVE CHARGE STORAGESome ECs use fast, reversible redox reactions at the surface of active materials, thus de!ning what is called the pseudo-capacitive behaviour. Metal oxides such as RuO2, Fe3O4 or MnO2 (refs 48, 49), as well as electronically conducting polymers50, have been extensively studied in the past decades. #e speci!c pseudo-capacitance exceeds

0

5

10

15

1.5 M1.5 M1.5 M1.0 & 1.4 M1.7 M

I IV

D

E

F

CB

A

Average pore size (nm)

0 1 2 3 4 5 15 25 35 45

II III

Norm

alize

d ca

paci

tanc

e (

F cm

–2)

+–

––

–

–

––

–

–

––

–

–

–

–

––

–

–

–

b d

a

–

–

–

–

–

–

Figure 4 Specific capacitance normalized by SSA as a function of pore size for different carbon samples. All samples were tested in the same electrolyte (NEt4

+,BF4–

in acetonitrile; concentrations are shown in the key). Symbols show experimental data for CDCs, templated mesoporous carbons and activated carbons, and lines show model fits43. A huge normalized capacitance increase is observed for microporous carbons with the smallest pore size in zone I, which would not be expected in the traditional view. The partial or complete loss of the solvation shell explains this anomalous behaviour42. As schematics show, zones I and II can be modelled as an electric wire-in-cylinder capacitor, an electric double-cylinder capacitor should be considered for zone III, and the commonly used planar electric double layer capacitor can be considered for larger pores, when the curvature/size effect becomes negligible (zone IV). A mathematical fit in the mesoporous range (zone III) is obtained using equation (5). Equation (6) was used to model the capacitive behaviour in zone I, where confined micropores force ions to desolvate partially or completely44. A, B: templated mesoporous carbons; C: activated mesoporous carbon; D, F: microporous CDC; E: microporous activated carbon. Reproduced with permission from ref. 44. © 2008 Wiley.

Norm

alize

d ca

paci

tanc

e (

F cm

2 )

0.66

7

8

9

10

11

12

13

4.3 Å

7.6 Å

7.9 Å

2.9 Å

EMI cation

TFSI anion

14

0.7 0.8Pore size (nm)

0.9 1

–

+

1.1

FluorineCarbon

NitrogenHydrogen

OxygenSulphur

H

HH

H

H

HH

H

FF

F

FF

F

ON

CO

OSS

C

O

HC C

C C

C CN N

Figure 5 Normalized capacitance change as a function of the pore size of carbon-derived-carbide samples. Samples were prepared at different temperatures in ethyl-methylimidazolium/trifluoro-methane-sulphonylimide (EMI,TFSI) ionic liquid at 60 °C. Inset shows the structure and size of the EMI and TFSI ions. The maximum capacitance is obtained when the pore size is in the same range as the maximum ion dimension. Reproduced with permission from ref. 46. © 2008 ACS.

New results

Traditional view

0 1 2 3 4 5 average pore size (nm)

0

5

10

15

Chmiola, J.; Yushin, G.; Gogotsi, Y.; Portet, C.; Simon, P.; Taberna, P. -

L., Science, 2006, 313, 1760.

1.5 M solution of tetraethylammonium tetrafluoroborate in acetonitrile

MD Simulation

Goal is to reconcile conflicting experiments Simulation setup

Nanoslits filled with room-temperature ionic liquids (RTILs)– Electrolyte: [emim+][Tf2N-] – Electrode: slit-shaped pore – Three layers of graphene sheets– Slit width (d): 0.67 ~ 1.8 nm– Applied potential: ~ 1.41 V – NPT: 333K, 1atm

22

emim+ Tf2N-

8.5×5.5×2.8 Å3 10.9×5.1×4.7 Å3

d

6.1 nm 3.5 nm

2.1 nm

MD Simulation23

Cathode Anode

Simulation setup

Capacitance Dependence on Micropore Size

Normalized capacitance as a function of slit width (C-d curve)

24

• 1st peak agrees will with experiments

• d > 1.0nm: 2nd peak is a new feature

• d = 0.7 ~ 1.0 nm: anomalous increase• d < 0.7 nm: decreases with pore size

Feng, G., & Cummings, P. T. (2011). Supercapacitor Capacitance Exhibits Oscillatory Behavior as a Function of Nanopore Size. The Journal of Physical Chemistry Letters, 2, 2859–2864. doi:10.1021/jz201312e

Documents

Materials Genome Initiative for Soft Matter: Computational ... … · 1 Materials Genome Initiative for Soft Matter: Computational Challenges Peter T. Cummings 1Department of Chemical