Top Banner
Exploring deep Earth minerals with accurate theory K.P. Driver, R.E. Cohen, Z. Wu, B. Militzer, P. Lopez Rios, M. Towler, R. Needs, and J.W. Wilkins Funding: NSF, DOE; Computation: NCAR, TeraGrid, NCSA, NERSC, OSC, CCNI Outline: 1) Probing minerals in deep Earth (with focus on silica, SiO 2 ) 2) Simulation methods predict minerals; First-principles (ab initio) methods are best 3) Introduction to many-body problem in solids; Quantum Monte Carlo succeeds 4) Computation of QMC silica phase diagram and thermodynamic properties crust: (18 mi, P<5 GPa, T<1000 K), mantle: (1800 mi, P<135GPa, T<2700 K), core: (4000 mi, <300 GPa, T~7000 K (>surface of sun)) What materials comprise Earth? Seismic waves directly probe Earth's interior. Diamond anvils and computers simulations infer. Take away from this talk: An appreciation for first-principles simulations and their usefulness for studying Earth minerals. Earth 4.5 billion years old 1 GPa = 145,000 lbs/in 2
28

Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Jun 12, 2019

Download

Documents

vodang
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Exploring deep Earth minerals with accurate theoryK.P. Driver, R.E. Cohen, Z. Wu, B. Militzer, P. Lopez Rios, M. Towler, R. Needs, and J.W. Wilkins  Funding: NSF, DOE; Computation: NCAR, TeraGrid, NCSA, NERSC, OSC, CCNI

Outline:1) Probing minerals in deep Earth (with focus on silica, SiO

2)

2) Simulation methods predict minerals; First-principles (ab initio) methods are best3) Introduction to many-body problem in solids; Quantum Monte Carlo succeeds4) Computation of QMC silica phase diagram and thermodynamic properties

crust: (18 mi, P<5 GPa, T<1000 K), mantle: (1800 mi, P<135GPa, T<2700 K),core: (4000 mi, <300 GPa, T~7000 K (>surface of sun))

What materials comprise Earth?●Seismic waves directly probe Earth's interior.●Diamond anvils and computers simulations infer.

Take away from this talk:An appreciation for first­principles simulationsand their usefulness for studying Earth minerals.

Earth 4.5 billion years old 1 GPa = 145,000 lbs/in2

Page 2: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Nuclear Bombsgo here

“restart core spin” = save the world

What would you do for a klondike bar?

(2003)

Page 3: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Compression of silica

●Simplest of Earth's silicates; ubiquitous component of Earth.●Complex series of phase transitions with increasing pressure.●Quartz to stishovite is a four­fold to six­fold coordination change.●Stishovite undergoes a 2nd order transition to a CaCl

2­structured phase.

●CaCl2­structure transforms to a PbO

2­structure, which is stable to the core.

Page 4: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

1)Ab initio: Quantum Monte Carlo - (nearly) exact many-body method - 100-1000 times more costly than DFT, N2

2)Ab initio: Density Functional Theory (DFT) -nobel prize in Chemistry 1998 – Kohn, Pople -mean field theory, N3 scaling

3)Semi-empirical methods (experimental input) -results biased towards experimental input -compute time scales linearly with N

atoms

4)Classical/Empirical Modeling -ignore quantum mechanics -compute time scales linearly with N

atoms

 Quantum, slow (refine structural data)

Classical, fast (search for structures)

Tools for calculating material properties based on electronic structure:

Zhong, OSU

Hierarchy of Simulation Techniques

Page 5: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

1) Electron Exchange Interactions (Fermi Correlation/Hund #1)●the interaction of electrons via the Pauli exclusion principle●largely responsible for the shape/volume of matter

2) Electron Correlation Interactions (Coulomb correlation/Hund #2)●Coulomb interactions cause electrons to stay out of each others way.

e­ e­In solids, there are Avagadro's number of electrons interacting within essentially continuous bands of quantum states.Is there a clever way to proceed?

DFT cleverly maps the many­body problem onto a single­particle problem while keeping,the theory exact. In practice, we don't know functionals for exchange and correlation exactly.

DFT's long term illness:

Many-body electron interactions required for accuracy

Fermion's obey Pauli

Page 6: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

● DFT works very well in many cases, but can unexpectedly fail.● Predicted properties can be highly dependent on form of the XC­functional.● Quartz/Stishovite: LDA works for structural properties, GGA works for energy.● DFT errors in volume ~5%; errors in elastic constants ~10%.

(Uninterested people can take a nap here and still pass the quiz at the end)

DFT XC-functionals can be unreliable

Page 7: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

What is Quantum Monte Carlo?

What is Quantum Monte Carlo?●A stochastic theory which solves the Schrödinger equation using Monte Carlo integration.●Uses a statistical representation of the wave function explicitly including many­body effects.●100­1000 times more computationally expensive than DFT. 

What is Monte Carlo?● An efficient way of solving many­dimensional integrals (mean value theorem).● Evaluation: Randomly sample the integrand and average the sampled values.Why use Monte Carlo?● Conventional integration methods (e.g. Simpson's rule) use a mesh of points and error in the result falls off increasingly slow with the dimension of the problem.● Statistical error from Monte Carlo is independent of dimension.

DFTVariational

MCDiffusion

MCTrial wave function Optimize wave function

Project out Ground State

=Correlation functionOrbital Determinant

Correlation Exchange

Error ~ 1

N

(Pauli happy!)

Page 8: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

What is Quantum Monte Carlo?

What is Quantum Monte Carlo?●A stochastic theory which solves the Schrödinger equation using Monte Carlo integration.●Uses a statistical representation of the wave function explicitly including many­body effects.●100­1000 times more computationally expensive than DFT. 

What is Monte Carlo?● An efficient way of solving many­dimensional integrals (mean value theorem).● Evaluation: Randomly sample the integrand and average the sampled values.Why use Monte Carlo?● Conventional integration methods (e.g. Simpson's rule) use a mesh of points and error in the result falls off increasingly slow with the dimension of the problem.● Statistical error from Monte Carlo is independent of dimension.

DFTVariational

MCTrial wave function Optimize wave function

Project out Ground StateCorrelation Exchange

Error ~ 1

N

(Pauli happy!)

=Correlation functionOrbital Determinant

Page 9: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Use QMC to compute phase diagram and properties of silica

Goals of this work:●Explore feasibility of using QMC for high pressure/temperature properties of silica.●Compute thermal equations of state and phase diagram.●Compute thermodynamic properties of silica.

Page 10: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

QMC1)Explicit many­body method.2)Use DFT's relaxed crystal structures.3)Optimize DFT wavefunction (fixed nodes).4)Compute energy stochastically.

DFT1)Single­particle theory in effective potential.2)Choose XC­functional and pseudopotential.3)Relax crystal structures.4)Compute energy and wavefunction.

Compute total energies of silica phases at several volumes (pressures)

DFT

QMC Static Energy vs Volume

    T=0 K Transition Pressure (GPa)

CaCl2­aPbO2Experiment 6 to 7  ~90 QMC 6.3(0.14) 88(8)DFT(WC) 2.1 86

Quartz­Stishovite

Page 11: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Add in energy due to thermal vibrations (temperature dependence)

F=E staticV F vibrationV ,T

Compute static lattice energy with QMC●Dominant energy contribution●Most accurate method available for solids●CASINO code

Compute vibrational free energy with DFT●Currently too costly for QMC●Vibrational energy is small●Typically well described in DFT●ABINIT, Linear Response, Quasi­harmonic

●Compute Helmholtz free energy

Dispersion data from Burkel, et al. Physica B, 263­264, pp412­415 (1999).Fr

eque

ncy 

(cm

­1)

Quartz Phonon Dispersion (P=0 GPa)

Page 12: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

For anyone who is currently zoned out, this is a good time to wake up!

The theoretical background part is over ... onto the results, whichwill be on the quiz.

x2

A subtle transition slide to the results section

Page 13: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Thermal Equations of State

● Thermal EoS determines fundamentalthermodynamic parameters and phase relations.

● QMC improves agreement with experiment in each phase: quartz, stishovite, PbO

2.

●Only small number of measurementsfor PbO

2, making QMC most accurate

available.

●QMC gives internal estimate of error. Gray shading indicates one standard deviation of statistical error. 

P=− ∂F∂V

T

Quartz

Stishovite

PbO2

Page 14: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Silica Phase Boundaries

Pressure (GPa)

Tem

pera

ture

 (K)

Tem

pera

ture

 (K) ●Equilibrium phase boundaries 

computed from Gibbs free energies.

●Metastable quartz­stishovitetransition measured with thermocalor shock. QMC agrees well.

●QMC CaCl2­PbO

2 transition lies 

between two measurements.

●DFT(WCGGA) boundary 4 GPa toolow for quartz­stishovite and withinstatistical error of QMC for CaCl

2­PbO

2

G=FPV G1PT ,V 1=G2PT ,V 2 at equilbrium

Page 15: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Thermal Expansivity

Temperature (K)

Ther

mal

 Exp

ansiv

ity (1

0­5K

­1)

●QMC and DFT(WCGGA) temperature and pressure dependence of .

●QMC shows best agreement for quartz.

●Experimentally, ­quartz appears at846 K – we only consider ­quartz.

●QMC and DFT show good agreement with stishovite measurements.

●PbO2 curves are the best available.

=1V ∂V

∂T P

Page 16: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Heat Capacity

Temperature (K)

C p/R

●QMC and DFT(WCGGA) temperature and pressure dependence of C

p.

●QMC and DFT results are nearly identical.

●Good agreement with experiment forquartz and stishovite.

●QMC PbO2 curves are best available.

C p= ∂H∂T

P

Page 17: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Conclusions

●Highly accurate first-principles calculations can be used to compute properties of minerals deep inside of Earth.

●Accurate description of many-body effects known as exchange and correlation are critical for successful prediction.

●QMC is the most accurate method available for computing materials properties, whichexplicitly includes many-body electron interactions.

●QMC has provided highly accurate phase boundaries and thermodynamical properties of silica phases up to the Earth's core.

Page 18: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Backup Slides1)DMC time step convergence2)DMC finite size convergence3)Shear modulus strain­energy technique4)DMC and VMC strain­energy curves5) Shear Modulus vs Pressure6)Statistical error propagation7)Silica enthalpy difference and volume difference8)Silica bulk moduli9)VMC10)DMC

Page 19: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

DMC Time Step Convergence

●Time step of 0.003 a.u. is converged to within 30 meV.

Page 20: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

DMC Finite Size Convergence

Page 21: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

cijkl=1V

∂2 E∂2

EV

=12

cijklijkl

Strain­energy density relation:

● Elastic constants obtained from curvature of energy­strain curve● Double well at 280 Bohr3 indicates elastic instability of stishovite● CaCl

2 becoming more stable that stishovite under pressure

                            Feasibility of Elastic Constants in QMC● Elasticity is a tough problem for QMC: energy differences ~ 0.005 eV● Extremely expensive to get accurate error bars for large (100 atom) systems● Through parallel computation on large supercomputers, it's possible to succeed.

Stishovite DFT(WC)R '=[ I]RStrain the lattice:

For a volume conserving strain:

1500 processor hours

Calculate elastic constants by straining the lattice

Page 22: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

1) Take optimized input structures from DFT(WC) ­ (we can't do forces in QMC yet)2) Run QMC on thousands of processors for a few days until error bars are sufficiently smallfor each structure.

VMC (500,000 hrs) DMC (additional 1.3 million hrs)

● QMC at this accuracy level is 1200 times more expensive than DFT.● QMC error bars must be made much smaller than the strain energy differences.● VMC error bars decrease twice as fast as DMC error bars.● Highest pressure curves are most difficult to fit and require smallest error bars(work on high pressure curves is still in progress).

QMC Energy vs. Strain Curves: The “Brute Force” Method

Page 23: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Stishovite Shear Modulusc 11

­c12

 (GPa

)

Pressure (GPa)

●Stishovite to CaCl2 transition is driven by instability in the elastic shear modulus.

●VMC modulus computed at several pressures and DMC checks at endpoints.●Shear modulus computed from strain­energy relation (brute force 1000 CPU cost of DFT)●All methods roughly agree, with the shear modulus going unstable around 50 GPa.

cij=1V

∂2 E∂2

Page 24: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

F=∑i , j

∂F∂C i

∂F∂C j

Cov [i , j ]

Statistical Error Propagation

1) Propagation of error equation from Taylor expanding a function about the mean values of its parmeters:

P=∑i

P i

E i

E i

Allow random Gaussian fluctuations on QMC energies E

i with stdv 

Ei

Fit Vinet Equation tonew set of energiesand compute property

actual data set

...Fit Vinet Equation tonew set of energiesand compute property

Fit Vinet Equation tonew set of energiesand compute property

standard deviation of synthetic data sets gives uncertainty in property

Allow random Gaussian fluctuations on QMC energies E

i with stdv 

Ei

Allow random Gaussian fluctuations on QMC energies E

i with stdv 

Ei

...

2) Propagation with Monte Carlo

Page 25: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Enthalpy Difference and Volume DifferenceEnthalpy Difference (Ha/SiO

2) % Volume Difference

●Enthalpy differences and errors determine equilibrium phase relations.●CaCl

2­PbO

2 enthalpy change is not measurable; phases are not quenchable to zero pressure.

●CaCl2 &

 PbO

2 enthalpy curves lie very close together compared to quartz­stishovite.

●Volume change in quartz­stishovite is 20 times larger than in CaCl2­PbO

2.

HP

bO2­

CaCl

2 (H

a/Si

O2)

Hsti

sh­q

uartz

(Ha/

SiO

2)

V

CaCl

2­P

bO2

V

quar

tz­s

tish

Pressure (GPa) Pressure (GPa)

Page 26: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

Bulk Modulus

Temperature (K)

Bulk

 Mod

ulus

 (GPa

)

●QMC and DFT(WCGGA) temperature and pressure dependence of K.

●K decreases linearly with T

●K increases linearly with P

●QMC and DFT generally agree.(T dependence comes from DFT)

K=−V ∂P∂V

T

Page 27: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

1) Variational principle evaluated using Monte Carlo integration

E 0⟨trial∣H∣trial⟩

E vmc≈1M∑i=1

M

E L Ri

(assume psi normalized)

● Sample configurations {R} according to the probability density function.● Metropolis algorithm does this efficiently for us in high dimensional spaces.● Evaluate E

L for each sampled configuration and average the values.

Probability density function Local Energy

E vmc=∫∣∣2[ H

]dR=∫∣R∣2 E L RdR

1

MError ~

2) Optimize the wavefunction by minimizing the energy or variance of the energy● Extremely important for high accuracy; See C. J. Umrigar, PRL (1988), (2005) 

(M = samples*cpu's)

Variational Quantum Monte Carlo (VMC)

Page 28: Exploring deep Earth minerals with accurate theorygreif.geo.berkeley.edu/~driver/Talks/Icecream_Driver_2009.pdf · Exploring deep Earth minerals with accurate theory ... transforms

τ

●DMC is a stochastic projector method for solving the full, many­body Schrödinger equation.

● The Schrödinger equation in imaginary timedescribes a combination of diffusionand branching of electron configurations.

● Electron configurations are allowed to propagate in imaginary time until theyare distributed according to the ground­statewavefunction of the system.

● Electron configurations with low potentialenergy proliferate, while those with highpotential energy die.

● After sufficient number of imaginary timesteps (τ), the exact ground­state wavefunction is projected out.

V(x)

0x

t

x

Diffusion Quantum Monte Carlo (DMC)