Forces: Calculating Them, and Using Them Shobhana Narasimhan JNCASR, Bangalore, India

Similar documents
5.76 Lecture #33 5/08/91 Page 1 of 10 pages. Lecture #33: Vibronic Coupling

Chapter 5 Vibrational Motion

HE ATOM & APPROXIMATION METHODS MORE GENERAL VARIATIONAL TREATMENT. Examples:

Lecture #18

Notes on iteration and Newton s method. Iteration

REGRESSION (Physics 1210 Notes, Partial Modified Appendix A)

Chapter 2 The Solution of Numerical Algebraic and Transcendental Equations

The Born-Oppenheimer approximation

1 Adiabatic and diabatic representations

v = -!g(x 0 ) Ûg Ûx 1 Ûx 2 Ú If we work out the details in the partial derivatives, we get a pleasing result. n Ûx k, i x i - 2 b k

Lecture 25 (Dec. 6, 2017)

ENGI 9420 Engineering Analysis Assignment 3 Solutions

PC5215 Numerical Recipes with Applications - Review Problems

5.80 Small-Molecule Spectroscopy and Dynamics

Machine Learning for Data Science (CS 4786)

U8L1: Sec Equations of Lines in R 2

mx bx kx F t. dt IR I LI V t, Q LQ RQ V t,

Physics 7440, Solutions to Problem Set # 8

Chapter 2: Numerical Methods

U8L1: Sec Equations of Lines in R 2

(c) Write, but do not evaluate, an integral expression for the volume of the solid generated when R is

CS537. Numerical Analysis and Computing

Introduction to Optimization Techniques. How to Solve Equations

Statistical Inference Based on Extremum Estimators

Differentiable Convex Functions

Math 5311 Problem Set #5 Solutions

Definitions and Theorems. where x are the decision variables. c, b, and a are constant coefficients.

17 Phonons and conduction electrons in solids (Hiroshi Matsuoka)

Optimization Methods MIT 2.098/6.255/ Final exam

For example suppose we divide the interval [0,2] into 5 equal subintervals of length

There are 7 crystal systems and 14 Bravais lattices in 3 dimensions.

Root Finding COS 323

CS321. Numerical Analysis and Computing

Quantum Annealing for Heisenberg Spin Chains

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 5

Castiel, Supernatural, Season 6, Episode 18

September 2012 C1 Note. C1 Notes (Edexcel) Copyright - For AS, A2 notes and IGCSE / GCSE worksheets 1

THE SOLUTION OF NONLINEAR EQUATIONS f( x ) = 0.

Find a formula for the exponential function whose graph is given , 1 2,16 1, 6

Infinite Sequences and Series

AP Calculus BC Review Applications of Derivatives (Chapter 4) and f,

Orthogonal transformations

Linear Support Vector Machines

Assignment 2 Solutions SOLUTION. ϕ 1 Â = 3 ϕ 1 4i ϕ 2. The other case can be dealt with in a similar way. { ϕ 2 Â} χ = { 4i ϕ 1 3 ϕ 2 } χ.

x x x 2x x N ( ) p NUMERICAL METHODS UNIT-I-SOLUTION OF EQUATIONS AND EIGENVALUE PROBLEMS By Newton-Raphson formula

NUMERICAL METHODS FOR SOLVING EQUATIONS

18.657: Mathematics of Machine Learning

ENGI 9420 Lecture Notes 3 - Numerical Methods Page 3.01

4755 Mark Scheme June Question Answer Marks Guidance M1* Attempt to find M or 108M -1 M 108 M1 A1 [6] M1 A1

Numerical Methods in Fourier Series Applications

Problem Cosider the curve give parametrically as x = si t ad y = + cos t for» t» ß: (a) Describe the path this traverses: Where does it start (whe t =

Machine Learning for Data Science (CS 4786)

Solutions to Final Exam Review Problems

Physics Supplement to my class. Kinetic Theory

PHY4905: Nearly-Free Electron Model (NFE)

Lecture 6 Chi Square Distribution (χ 2 ) and Least Squares Fitting

For example suppose we divide the interval [0,2] into 5 equal subintervals of length

METHOD OF FUNDAMENTAL SOLUTIONS FOR HELMHOLTZ EIGENVALUE PROBLEMS IN ELLIPTICAL DOMAINS

Linear Regression Demystified

MATH 10550, EXAM 3 SOLUTIONS

Appendix A. Nabla and Friends

Matsubara-Green s Functions

Lecture 6 Chi Square Distribution (χ 2 ) and Least Squares Fitting

MTH Assignment 1 : Real Numbers, Sequences

IP Reference guide for integer programming formulations.

Physics 324, Fall Dirac Notation. These notes were produced by David Kaplan for Phys. 324 in Autumn 2001.

Principle Of Superposition


Machine Learning Regression I Hamid R. Rabiee [Slides are based on Bishop Book] Spring

n 3 ln n n ln n is convergent by p-series for p = 2 > 1. n2 Therefore we can apply Limit Comparison Test to determine lutely convergent.

Root Finding COS 323

Systems of Particles: Angular Momentum and Work Energy Principle

Linear regression. Daniel Hsu (COMS 4771) (y i x T i β)2 2πσ. 2 2σ 2. 1 n. (x T i β y i ) 2. 1 ˆβ arg min. β R n d

Hydrogen (atoms, molecules) in external fields. Static electric and magnetic fields Oscyllating electromagnetic fields

Apply change-of-basis formula to rewrite x as a linear combination of eigenvectors v j.

A widely used display of protein shapes is based on the coordinates of the alpha carbons - - C α

Chimica Inorganica 3

Series: Infinite Sums

EXPERIMENT OF SIMPLE VIBRATION

PAPER : IIT-JAM 2010

Math 312 Lecture Notes One Dimensional Maps

Kinetics of Complex Reactions

Lecture 2 October 11

Markov Decision Processes

1. Szabo & Ostlund: 2.1, 2.2, 2.4, 2.5, 2.7. These problems are fairly straightforward and I will not discuss them here.

SECTION 2 Electrostatics

Complex Analysis Spring 2001 Homework I Solution

Assignment 1 : Real Numbers, Sequences. for n 1. Show that (x n ) converges. Further, by observing that x n+2 + x n+1

1. Hydrogen Atom: 3p State

Most text will write ordinary derivatives using either Leibniz notation 2 3. y + 5y= e and y y. xx tt t

Lyapunov Stability Analysis for Feedback Control Design

State Space Representation

Using Spreadsheets as a Computational Tool in Teaching Mechanical. Engineering

We will conclude the chapter with the study a few methods and techniques which are useful

Lecture 6: Integration and the Mean Value Theorem

Axis Aligned Ellipsoid

Ray-triangle intersection

Machine Learning Theory Tübingen University, WS 2016/2017 Lecture 12

Math 257: Finite difference methods

is also known as the general term of the sequence

Transcription:

Forces: Calculatig Them, ad Usig Them Shobhaa Narasimha JNCASR, Bagalore, Idia shobhaa@jcasr.ac.i Shobhaa Narasimha, JNCASR 1

Outlie Forces ad the Hellma-Feyma Theorem Stress Techiques for miimizig a fuctio Geometric optimizatio usig forces Froze Phoo Calculatios Molecular Dyamics with forces Shobhaa Narasimha, JNCASR 2

The Bor-Oppeheimer Approimatio Ma Bor (1882-1970 R. Oppeheimer (1904-1967 Shobhaa Narasimha, JNCASR 3

The Bor-Oppeheimer Approimatio The may-particle wavefuctio ψ is a fuctio of both uclear ad electroic coordiates. Separate out the Hamiltoia ito a uclear ad electroic part. The wavefuctio ca the be writte as: Ψ tot =ψ ψ uc elec OK because electros are much lighter tha uclei ad therefore respod istatly o time-scale of uclear motio. Shobhaa Narasimha, JNCASR 4

Forces Need for geometry optimizatio ad molecular dyamics. Ca also use to get phoos. Could get as fiite differeces of total eergy - too epesive! Use force (Hellma-Feyma theorem. Richard Feyma s Seior Thesis! (whe he was 21 Shobhaa Narasimha, JNCASR 5

Hellma-Feyma Theorem Wat to calculate force o io I: Get three terms: Whe is a eigestate, -Substitute this Shobhaa Narasimha, JNCASR 6

Hellma-Feyma Theorem Wat to calculate force o io I: Get three terms: Whe is a eigestate, -Substitute this Shobhaa Narasimha, JNCASR 7

Hellma-Feyma Theorem Wat to calculate force o io I: Get three terms: Whe is a eigestate, -Substitute this Shobhaa Narasimha, JNCASR 8

Hellma-Feyma Theorem (cotd. The force is ow give by Note that we ca ow calculate the force from a calculatio at ONE cofiguratio aloe huge savigs i time. If the basis depeds o ioic positios (ot true for plae waves, would have etra terms = Pulay forces. If is ot a eact eigestate (electroic calculatio ot well coverged, may get big errors i forces calculated usig this prescriptio! Shobhaa Narasimha, JNCASR 9

Hellma-Feyma Theorem (cotd. The force is ow give by Note that we ca ow calculate the force from a calculatio at ONE cofiguratio aloe huge savigs i time. If the basis depeds o ioic positios (ot true for plae waves, would have etra terms = Pulay forces. If is ot a eact eigestate (electroic calculatio ot well coverged, may get big errors i forces calculated usig this prescriptio! Shobhaa Narasimha, JNCASR 10

Hellma-Feyma Theorem (cotd. The force is ow give by Note that we ca ow calculate the force from a calculatio at ONE cofiguratio aloe huge savigs i time. If the basis depeds o ioic positios (ot true for plae waves, would have etra terms = Pulay forces. If is ot a eact eigestate (electroic calculatio ot well coverged, may get big errors i forces calculated usig this prescriptio! Shobhaa Narasimha, JNCASR 11

Usig H-F Theorem i a (plae-wave DFT calculatio Force o io I give by: where =(pseudopotetial due to io cores ad = iteractio of ios with each other. Shobhaa Narasimha, JNCASR 12

Stress Strai: Stress: Stress Theorem (Nielse & Marti, 1985 as for forces, ca calculate at a sigle cofiguratio. What if the primitive lattice vectors (specifyig uit cell are ot optimal? - Forces o atoms may = 0 (e.g., a FCC crystal with wrog lattice costat - Stress will ot be zero, however. < 0 cell would like to epad. > 0 cell would like to cotract. Shobhaa Narasimha, JNCASR 13

Geometric Optimizatio Wat to move the atomic positios aroud util the lowest-eergy equilibrium cofiguratio is obtaied. At equilibrium, for all I. We are searchig for a miimum i a 3N I -dim space. Fidig global miimum is hard! Will usually ed up i earest miimum Shobhaa Narasimha, JNCASR 14

Miimizatio Relevat to may parts of DFT calculatios: - Optimizig ioic positios - Miimizig eergy fuctioal - Diagoalizig Hamiltoia - Achievig self-cosistecy (miig A brief foray ito umerical methods. Covetio: subscripts coordiates, superscripts iteratios Shobhaa Narasimha, JNCASR 15

Miimizatio i 1-D usig gradiets Cosider a fuctio f(; we wat to fid 0, the value of where the fuctio has its miimum value. Iterative methods: successive approimatios 1, 2, 3, 0 Ca take several small dowhill steps, i the directio opposite the gradiet f (. = β '( + 1 ( f Might take a log time to coverge. Note: eed first derivatives Shobhaa Narasimha, JNCASR 16

Fidig zeroes i 1-D: the Newto-Raphso Method Use iformatio from first derivatives to make a series of guesses: f f ( 0 '( = + 1 f( + 1 = f ( f '( Note: will coverge i oe step if f is liear +1 Shobhaa Narasimha, JNCASR 17

Fidig miimum i 1-D: the Newto-Raphso Method Lookig for a miimum i f is equivalet to lookig for a zero of f. So replace f by f ad f by f to get: or sometimes: + 1 + 1 = = f '( f "( f '( β f "( Note: eed first derivatives ad secod derivatives Note: will coverge i oe step if f is liear. So will coverge i oe step if f is quadratic. Shobhaa Narasimha, JNCASR 18

Miimizatio i a N-d space Cosider a fuctio f( of N variables = 1, 2,, N The gradiet is f(. Note that at a give, this poits i directio of maimum icrease of f(. Wat to fid 0 s.t. f( has its miimum value at 0, i.e., f( 0 = 0. Will fid iteratively, through a sequece of poits i the N-dimesioal space that are steppig stoes to fidig 0. Covetio: subscripts coordiates, superscripts iteratios. Shobhaa Narasimha, JNCASR 19

Steepest Descet Keep goig dowhill i directio opposite local gradiet. Could try takig lots of little dowhill steps: +1 = β f( = + βg Always, g perpedicular to g +1. Better: Oce directio g idetified, do lie miimizatio alog it. Ca be slow (may ot reach miimum! Problem: whe movig alog ew directio, lose some miimizatio alog old oe(s. Shobhaa Narasimha, JNCASR 20

Secod derivatives i a N-d space: The Hessia is a N N matri, whose elemets are give by: H ij = 2 f ( / i j It gives iformatio about the curvature of the fuctio. Shobhaa Narasimha, JNCASR 21

Newto-Raphso Miimizatio i N-d 1-d N-d f f( 1/f H -1 + 1 = f '( β f "( = β [ H( ] f ( + 1 1 Need to compute gradiets ad iverse Hessias Shobhaa Narasimha, JNCASR 22

Quasi-Newto-Raphso Methods We had: = β [ H( ] f ( + 1 1 But H -1 may be hard/epesive to compute directly. Istead, build up estimates for H -1 by aalyzig successive gradiet vectors. Various prescriptios for doig this, e.g., Broyde method, BFGS method. Shobhaa Narasimha, JNCASR 23

Shobhaa Narasimha, JNCASR 24 BFGS Miimizatio Broyde, Fletcher, Goldfarb, Shao (1970. ( ] [ ( ] [ 1 1 1 f f H H = + ρ [ ] [ ] [ ] [ ] [ ] + + + = + T T T T T T T s s H H s s s s s H H H ( ( ( ( ( ( ( 1 1 1 1 1 1 1 Trust radius Iverse Hessia approimated as: where ( ( 1 f f ad ( ] ( [ 1 f H s (No doubt you will agree that this is a good poit at which to ed our foray ito umerical methods Newto-Raphso

Back to the Problem of Ioic Relaatio Fuctio f to be miimized is total eergy E tot. Poits i 3N I -d space correspod to set of ioic coordiates ( 1, y 1, z 1, 2, y 2, z 2, NI, y NI, z NI. Gradiets f( correspod to set of 3 compoets of forces o the N I ios. Forces ca be computed usig Hellma-Feyma theorem. Now use a miimizatio scheme to fid the ioic positios that give the lowest value of E tot, which is also whe the forces o all ios are (close to zero. Shobhaa Narasimha, JNCASR 25

The Vibratioal Frequecy of a Diatomic Molecule Ca obtai from eergies or from forces: e.g., for CO: 1 3 2 E = k + O( 2 F = k + O( 2 Saada Biswas Shobhaa Narasimha, JNCASR 26

Froze-Phoo Calculatios Take sapshots of a crystal as it is vibratig i a particular mode. Use Hellma-Feyma theorem to compute forces as a fuctio of displacemet. Force vs. displacemet gives phoo frequecy. Note: Ca istead use desity fuctioal perturbatio theory (later today, Stefao Baroi s lecture! Shobhaa Narasimha, JNCASR 27

The Force-Costat Tesor Φ α,β (R i -R j force iduced o atom j i directio β, upo movig atom i i directio α Ca be obtaied easily if forces ca be computed. Shobhaa Narasimha, JNCASR 28

Gettig Phoo Frequecies Dyamical Matri: ~ D( k = i k R e R Φ( R Diagoalize to get phoo frequecies: Det ~ ~ 2 [ D( k Mω ( k ] = 0 Shobhaa Narasimha, JNCASR 29

What else ca you do with forces? Newto s equatios of motio: F = m a = I I I m I 2 d R 2 dt I If forces ca be computed, ca itegrate these to get ioic positios as a fuctio of time. Molecular Dyamics (tomorrow Shobhaa Narasimha, JNCASR 30

Computig Forces & Stress with PWscf tprfor =.TRUE. (Set automatically if calculatio = rela or md or vc-md tstress =.TRUE. (Set automatically if calculatio = vc-md Shobhaa Narasimha, JNCASR 31

Forces Obtaied usig PWscf e.g., for a (urelaed poit defect (vacacy i graphee: Kacha Ulma Shobhaa Narasimha, JNCASR 32

Ioic Relaatio i PWscf Tell the program to carry out ioic relaatio, ad say which method to use to fid miimum &cotrol calculatio = rela. io_dyamics = bfgs &cotrol calculatio = md. io_dyamics = damp Shobhaa Narasimha, JNCASR 33

Ioic Relaatio i PWscf (cotd. Say which atoms are to be moved & i which directios. e.g., for a four-layer Al(001 slab: ATOMIC_POSITIONS Al 0.000 0.000 2.828 0 0 1 Al 0.500 0.500 2.121 0 0 1 Al 0.000 0.000 1.414 0 0 0 Al 0.500 0.500 0.707 0 0 0 Allow these atoms to move oly alog z Fi these atoms Shobhaa Narasimha, JNCASR 34

The Ed! May the Force Be With You. Shobhaa Narasimha, JNCASR 35