You are on page 1of 587

Phase Transitions in Materials

Offering a fresh viewpoint on phase changes and the thermodynamics of materials, this
textbook covers the thermodynamics and kinetics of the most important phase transitions
in materials science, spanning classical metallurgy through to nanoscience and quantum
phase transitions.
Clear, concise, and complete explanations rigorously address transitions from the atomic
scale up, providing the quantitative concepts, analytical tools, and methods needed to
understand modern research in materials science. Topics are grouped according to complexity, ensuring that students have a solid grounding in core topics before they begin
to tackle more advanced material, and are accompanied by numerous end-of-chapter
problems.
With explanations firmly rooted in the context of modern advances in electronic structure
and statistical mechanics, and developed from classroom teaching, this book is the ideal
companion for graduate students and researchers in materials science, condensed matter
physics, solid state science, and physical chemistry.
Brent Fultz is the Barbara and Stanley R. Rawn, Jr., Professor of Materials Science and
Applied Physics at the California Institute of Technology. He has been awarded a Presidential Young Investigator Award, the EMPMD Distinguished Scientist Award (2010), and
has led large projects such as the state-of-the-art neutron scattering instrument, ARCS, and
data analysis for neutron scattering experiments, DANSE.

16:44:32

16:44:32

Phase Transitions in Materials
BRENT FULTZ
California Institute of Technology

16:44:32

University Printing House, Cambridge CB2 8BS, United Kingdom
Cambridge University Press is part of the University of Cambridge.
It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781107067240
© B. Fultz 2014
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2014
Printed in the United Kingdom by TJ International Ltd, Padstow, Cornwall
A catalog record for this publication is available from the British Library
Library of Congress Cataloging in Publication data
Fultz, B. (Brent), author.
Phase transitions in materials / Brent Fultz, California Institute of Technology.
pages cm
Includes bibliographical references and index.
ISBN 978-1-107-06724-0
1. Phase transformations (Statistical physics)–Textbooks. 2. Thermodynamics–Textbooks.
3. Materials–Thermal properties–Textbooks. 4. Statistical mechanics–Textbooks. I. Title.
QC175.16.P5F86 2014
530.4 74–dc23
2013046223
ISBN 978-1-107-06724-0 Hardback
Additional resources for this publication at www.cambridge.org/fultz
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.

16:44:32

This book is dedicated to Emily, Eric, and Elissa

16:44:46

16:44:46

Contents

Preface
Acknowledgments
Notation

page xiii
xvii
xix

Part I Basic thermodynamics and kinetics of phase transformations
1

Introduction
1.1
1.2
1.3
1.4
1.5
1.6

2

Essentials of T–c phase diagrams
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11

3

What is a phase transition?
Atoms and materials
Pure elements
Alloys: unmixing and ordering
Transitions and transformations
Brief review of thermodynamics and kinetics
Problems

Overview of the approach
Intuition and expectations about alloy thermodynamics
Free energy curves, solute conservation, and the lever rule
Common tangent construction
Continuous solid solubility phase diagram
Solid solutions
Unmixing phase diagrams
Eutectic and peritectic phase diagrams
Ternary phase diagrams
Long-range order in the point approximation
Alloy phase diagrams
Problems

Diffusion
3.1
3.2
3.3
3.4

The diffusion equation
Gaussian and error function solutions to the 1D diffusion
equation
Fourier series solutions to the diffusion equation
Bessel functions and other special function solutions to the
diffusion equation

16:53:48

1
3
3
4
6
9
11
14
18
20
20
22
26
28
30
31
37
39
42
44
48
49
52
52
56
61
66

Contents

viii

3.5
3.6

4

Nonequilibrium processing of materials
Alloy solidification with solute partitioning
Alloy solidification: suppressed diffusion in the solid phase
Alloy solidification: suppressed diffusion in the solid and liquid
Practical issues for alloy solidification and evaporation
Heat flow and kinetics
Nucleation kinetics
Glass formation
Solid-state amorphization and suppressed diffusion in a solid phase
Reactions at surfaces
The glass transition
Problems

96
96
99
100
105
106
109
111
112
114
115
120
121

Part II The atomic origins of thermodynamics and kinetics

125

Energy
6.1
6.2
6.3
6.4
6.5
6.6

7

74
75
76
79
81
85
92
94

Terminology and issues
Critical nucleus
Heterogeneous nucleation
Free energy curves and elastic energy
The nucleation rate
Time-dependent nucleation
Problems

Effects of diffusion and nucleation on phase transformations
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
5.9
5.10
5.11

6

69
71
72

Nucleation
4.1
4.2
4.3
4.4
4.5
4.6

5

Kinetic master equation and equilibrium
Linear kinetic response
Problems

Molecular orbital theory of diatomic molecules
Electronic energy bands in solids
Elastic constants and the interatomic potential
Linear elasticity
Misfitting particle
Surface energy
Problems

Entropy
7.1
7.2
7.3
7.4
7.5

Static and dynamic sources of entropy
Short-range order and the pair approximation
Local atomic structures described by clusters
Thermodynamics with cluster approximations
Concept of vibrational entropy

16:53:48

127
127
134
145
149
153
158
161
163
163
165
168
170
172

Contents

ix

7.6
7.7
7.8
7.9

8

175
176
179
187
190

Pressure
8.1
8.2
8.3
8.4
8.5
8.6

9

Phonon statistics
Lattice dynamics and vibrational entropy
Bond proportion model
Bond-stiffness-versus-bond-length model
Problems

Materials under pressure at low temperatures
Thermal pressure, a step beyond the harmonic model
Free energies and phase boundaries under pressure
Chemical bonding and antibonding under pressure
Two-level system under pressure
Activation volume
Problems

Atom movements with the vacancy mechanism
9.1
9.2
9.3
9.4
9.5

Random walk and correlations
Phenomena in alloy diffusion
Diffusion in a potential gradient
Nonthermodynamic equilibrium in driven systems
Vineyard’s theory of diffusion
Problems

Part III Types of phase transformations
10

Melting
10.1
10.2
10.3
10.4
10.5
10.6
10.7

11

211
211
220
227
232
235
240
245
247
247
248
250
257
259
261
264
266

Free energy and latent heat
Chemical trends of melting
Free energy of a solid
Entropy of a liquid
Thermodynamic condition for Tm
Glass transition
Two dimensions
Problems

Transformations involving precipitates and interfaces
11.1
11.2
11.3
11.4
11.5
11.6
11.7
11.8

194
194
199
200
202
205
208
209

Guinier–Preston zones
Surface structure and thermodynamics
Surface structure and kinetics
Chemical energy of a precipitate interface
Elastic energy and shape of growing precipitates
Precipitation at grain boundaries and defects
The eutectoid reaction and ferrous metallurgy
The Kolmogorov–Johnson–Mehl–Avrami growth equation

16:53:48

268
268
270
276
278
280
282
285
291

Contents

x

12

11.9 Coarsening
Problems

293
296

Spinodal decomposition

298
298
300
304
309
311
313

12.1
12.2
12.3
12.4
12.5

13

Phase field theory
13.1
13.2
13.3
13.4

14

Concentration fluctuations and the free energy of solution
Adding a square gradient term to the free energy F(c)
Constrained minimization of the free energy
The diffusion equation
Effects of elastic energy
Problems

Spatial distribution of phases and interfaces
Solidification
Ginzburg–Landau equation and order parameter
Interfaces between domains
Problems

Method of concentration waves and chemical ordering
14.1
14.2
14.3
14.4

Structure in real space and reciprocal space
Symmetry and the star
The free energy in k-space with concentration waves
Symmetry invariance of free energy and Landau–Lifshitz rule for
second-order phase transitions
14.5 Thermodynamics of ordering in the mean field approximation with
long-range interactions
Problems

15

Diffusionless transformations
15.1
15.2
15.3
15.4
15.5
15.6
15.7

16

Dislocations and mechanisms
Twinning
Martensite
The crystallographic theory of martensite
Landau theory of displacive phase transitions
First-order Landau theory
Crystal instabilities and phonons
Problems

Thermodynamics of nanomaterials
16.1
16.2
16.3
16.4
16.5

Grain boundary structure
Grain boundary energy
Gibbs–Thomson effect
Energies of free electrons confined to nanostructures
Configurational entropy of nanomaterials

16:53:48

315
315
318
319
322
330
332
332
338
341
344
349
353
355
356
360
362
368
370
374
377
381
383
384
386
387
390
392

Contents

xi

16.6
16.7
16.8
16.9

17

Magnetic and electronic phase transitions
17.1
17.2
17.3
17.4
17.5
17.6
17.7
17.8

18

Vibrational entropy
Gas adsorption
Characteristics of magnetic nanoparticles
Elastic energy of anisotropic microstructures
Problems

Overview of magnetic and electronic phase transitions
Exchange interactions
Correlated electrons
Thermodynamics of ferromagnetism
Spin waves
Thermodynamics of antiferromagnetism
Ferroelectric transition
Domains
Problems

Phase transitions in quantum materials
18.1
18.2
18.3
18.4
18.5

Low-temperature analysis of phase boundaries
19.1
19.2
19.3
19.4
19.5

20

Ground-state analysis for T = 0
Richards, Allen, Cahn ground-state maps
Low but finite temperatures
Analysis of equiatomic bcc alloys
High-temperature expansion of the partition function
Problems

Cooperative behavior near a critical temperature
20.1
20.2
20.3
20.4
20.5
20.6
20.7

404
405
410
414
417
420
423
426
428
430
432
432
435
437
440
448
452

Bose–Einstein condensation
Superfluidity
Condensate wavefunction
Superconductivity
Quantum critical behavior
Problems

Part IV Advanced topics
19

394
397
399
401
402

Critical exponents
Critical slowing down
The Rushbrooke inequality
Scaling theory
Scaling and decimation
Partition function for one-dimensional chain
Partition function for two-dimensional lattice
Problems

16:53:48

453
455
456
457
458
463
465
466
468
468
469
471
472
474
475
479
482

4 24 Ordering transformations with vacancies B2 ordering with vacancies in the point approximation Vacancy ordering Kinetic paths Problems 508 508 512 515 518 521 Atomic processes and diffusion Dissipation and fluctuations Inelastic scattering Phonons and quantum mechanics Problems Vibrational thermodynamics of materials at high temperatures 24.2 Real space approach 21.1 23.3 24.3 k-space approach Problems 22 Statistical kinetics of ordering transformations 22.6 492 493 495 499 500 506 Lattice dynamics Harmonic thermodynamics Quasiharmonic thermodynamics Thermal effects beyond quasiharmonic theory Anharmonicity and phonon–phonon interactions Electron–phonon interactions and temperature Problems Further reading References Index 522 522 527 528 531 532 536 538 540 545 554 16:53:48 .2 22.Contents xii 21 Elastic energy of solid precipitates 483 483 485 488 491 21.3 22.4 24.1 24.2 24. dissipation.3 23.1 Transformation strains and elastic energy 21.4 23 Diffusion. and inelastic scattering 23.5 24.2 23.1 22.

It is a textbook. and from the statistical kinetics of the atom–vacancy interchange. The level and detail are appropriate for upper division undergraduate students and graduate students in materials science and materials physics. phenomena in nanomaterials. candidacy examinations. The approach uses a minimum amount of information about interatomic interactions. and reasonably complete. phase transformations by nucleation and growth. are presented in the context of phase transitions. however. so the emphasis is on explanations of phenomena rather than a scholarly assessment of their origins. and quantum phase transitions. Essentials of diffusion. The goal is explanations that are concise. avoiding detailed issues at the level of electrons.D. phase transitions involving electrons or spins. and should be approached differently. and their effects on kinetics are covered in Part I. Part II addresses the origins of materials thermodynamics and kinetics at the level of atoms and electrons.Preface Content This book explains the thermodynamics and kinetics of most of the important phase transitions in materials science. 16:44:34 . spinodal decomposition. with the concepts used to understand them. the statistical mechanics of atom arrangements on a lattice is developed. This approach illustrates key concepts of equilibrium T–c phase diagrams.1 After a general overview of phase transitions. freezing and phase fields. Electronic and elastic energy are covered. The kinetics of atom movements are developed for diffusion in solids. These different phase transitions in materials are covered at different breadths and depths based on their richness or importance. martensitic transformations. although this reflects my own bias. and is extended to explain some kinetic processes. Effects of pressure. although the connection between processing and 1 The author asks graduate students to explain some of the key concepts at a blackboard during their Ph. are explained with a few concepts of chemical bonding. Part I presents topics that all graduate students in materials science must know. especially configurational and vibrational. continuous ordering. Statistical mechanics on an Ising lattice is used to understand alloy phase stability for basic behaviors of chemical unmixing and ordering transitions. combined with temperature. Part III is the largest. It describes many of the important phase transformations in materials. nucleation. and the different types of entropy. Topics include melting. The book is organized for approximately linear coverage in a graduate-level course. clear. The book should also be useful for researchers who are not specialists in these fields. The four parts of the book serve different purposes. Many topics from metallurgy and ceramic engineering are covered.

This text is a snapshot of phase transitions in materials in the year 2013. Part IV presents topics that are more modern. the topic of domains. for conciseness in Part IV there are some omissions of methods and steps. and dynamics with the kinetic master equation. composed from the angle of the author. and density functional theory with extensions to phenomena at finite temperatures. but have proved their importance. and it is assumed that the reader has some familiarity with the terms in this expression. Part III includes a number of topics from condensed matter physics that were selected in part because they give new insights into materials phenomena. Teaching I use this text for a graduate-level course taken by Ph. so much of the text focuses on formulating free energies for different systems. Phase transitions are thermodynamic phenomena based on free energy alone. phase transitions in fluid systems including phenomena near the critical temperature. Impressively. These include: other mechanisms of atom movements (and their effects on kinetics). allowing for a more concise presentation than in traditional texts. that many core topics will remain the same – the free energy of materials will remain the central concept. and massive transformations. Many topics in phase transitions and related phenomena are not covered in this text. and parts of it will look different in a decade or so. but the kinetic master equation is also used throughout the text. surrounded by issues of kinetics. there is some traditional presentation of diffusion and nucleation. the renormalization group. although the classical equation G = E − TS + PV is used widely. a k-space formulation of elastic energy. students in both materials science and in applied physics at the California Institute of Technology. The free energy is often formulated with models based on statistical mechanics. fluctuations and dissipation. Sometimes the thermodynamics of phase transitions is developed with the partition function. molecular dynamics methods (classical and quantum). but unlike Parts I through III. whereas phase transformations include kinetic processes that alter the life cycle of the phase change. Phase transitions originate from discontinuities in free energy functions.D. The topics in Part IV are explained at a fundamental level. the square gradient energy. The Ising model proves a reliable workhorse. polymer flow and dynamics including reptation. this field continues to offer a rich source of new ideas and results for both fundamental and applied research. and some complexities of high-temperature thermodynamics are presented. For the kinetics of phase transformations. nonequilibrium states in crystalline alloys. The book draws a distinction between phase transformations and phase transitions.xiv Preface properties is less emphasized. however. Also beyond the scope of the book are computational methods that are increasingly important for studies of phase transformations in materials. the effect of curvature on nucleation. I expect. Lowand high-temperature treatments of the partition function. Other topics that recur in the text are Landau theory in various forms. The field of phase transitions is huge. scaling theory. which 16:44:34 . The 10-week course. offering methods and results that are useful for many different phase transitions in materials. including: Monte Carlo methods. and continues to grow.

but these problems usually include longer explanations and hints that may be worth reading even without working the problem. 16:44:34 . None of the problems is intended to be particularly difficult. as is prior exposure to diffusion and transport. Others develop new concepts not described in the chapter. this year I covered Chapters 10. or extend analyses. so the students are familiar with the use of the partition function to obtain thermodynamic quantities. 24. For example. and this experience has helped to improve their wording and content. is offered in the third academic quarter as part of a one-year sequence. 19. The majority of these problems make use of concepts explained in the text. Familiarity with some concepts from solid-state physics and chemistry is certainly helpful. Please ask me for further information. Time limitations force a selection of topics from Parts III and IV.xv Preface includes approximately 30 hours of classroom lectures. An instructor can certainly exercise discretion in selecting topics for the second half of his or her course. It may be unrealistic to cover all the content in the book in a 15-week semester with 45 hours of lectures. An expanding online solutions manual is available to course instructors whose identity can be verified. but I typically cover more of Part III than Part IV. I cover all topics in Parts I and II. The first two quarters in this sequence cover thermodynamics and statistical mechanics. and some can be answered quickly once the main idea emerges. Most of the problems at the end of each chapter have been used for weekly student assignments. fill in the explanations of concepts. 11. 12. 16. 15. and have seen basic concepts from quantum statistical mechanics such as the Fermi–Dirac distribution. but the text develops many of the important concepts as needed. 22. I usually assign five or six problems every week during the term. moving in sequence through these chapters. In the one-quarter graduate-level course at Caltech. parts of 13. and selections from 20.

16:44:34 .

C. Asta. M. Nagel.-Q. van de Walle. Manley. P. L. T. Johnson. O. A. Lucas.G.N. Papandrew. 16:46:52 . K. L.J.D. Samwer.L.Y.Acknowledgments I thank J. Tracy. D. Johnson. The development of the topic of vibrational entropy would not have been possible without the contributions of my junior collaborators at Caltech. Chen. M. H. Swan-Wood. M. Bogdanoff. Mauger. J.E. Ceder. Hoyt for collaborating with me on a book chapter about phase equilibria and phase transformations that prompted me to get started on this book.G. and S. Winterrose. Lin. T. Anthony. D.L. Zunger.D. A. Today several of them are taking this field into new directions.S. Rez. M. Lan.F.W. especially L.J. and W. This work was supported by the NSF under award DMR-0520547. L. Li. Kresch. Important ideas have come from stimulating conversations over the years with A. Delaire.L. P.B.Y. Jeff has since published a fine book on phase transformations in materials that is available at low cost from McMaster Innovation Press. Frase. Ozolins.J. G. A. Purewal. Khachaturyan. J. M. A. de Fontaine. V. Yue.

16:46:52 .

Cijlm chemical composition (atomic fraction) speed of sound or light concentration of A-atoms weight of atomic wavefunction on atom A in a molecular wave function electronic heat capacity heat capacity at constant pressure heat capacity at constant volume elastic constant D D D0 diffusion coefficient deformation potential prefactor for exponential form of diffusion coefficient .Notation a A  A A-atom APDB α α α-phase α-sublattice αi α2 lattice parameter area vector potential of magnetic field generic chemical element antiphase domain boundary coefficient of linear thermal expansion critical exponent for heat capacity generic phase a lattice of like atoms within an ordered structure root of Bessel function electron–phonon coupling factor b bA  b(k) B  B B-atom ) B(R β β β-phase β-sublattice Burgers vector of dislocation coherent neutron scattering length of isotope A Fourier transform of pairwise energy for two concentration waves bulk modulus magnetic field generic chemical element pairwise energy between atoms coefficient of volume thermal expansion critical exponent for density generic phase a lattice of like atoms within an ordered structure c c cA cA Cel CP (T) CV (T) Cij .

Notation xx ˜ D(c)  Dij (k)  D(k). eW  eκj (k) erf(z) E  E Eel   ε F j . energy of phonon Fermi energy strain f fα fj f (c) F F Fξ (c.and B-atom energy of two atoms. T) correlation factor (atomic) fraction of α-phase interaction free energy free energy per unit volume Helmholtz free energy force free energy for phase ξ with composition c at temperature T g(ε) g γ γ γj γxy phonon density of states reciprocal lattice vector coefficient for linear electronic heat capacity vs. t) gradient (of concentration) Gibbs free energy Van Hove space-time correlation function atomic jump frequency point at origin of reciprocal lattice h  H bond integral Planck’s constant divided by 2π Hamiltonian . T Grüneisen parameter Grüneisen parameter for phonon mode j shear strain −→ grad(c) or ∇c G G(r. δ GV G∗ (r) interdiffusion coefficient dynamical matrix. element of relative change in radius (of misfitting sphere) change in Gibbs free energy per unit volume activation barrier for nucleation static wave of chemical concentration e eA eAB eR . on their right or wrong sublattices polarization for atom of basis index κ in phonon of k in branch j error function energy. thermodynamic energy electric field elastic energy energy. ij charge of electron energy of an A-atom on a crystal site energy of a pair (bond) between an A. A and B. energy of electron fractional difference in T from Tc energy.

Notation xxi j J0 (x). κ l κ κ partitioning ratio k = cs /cl wavevector Boltzmann’s constant thermal conductivity of solid and liquid coefficient for square gradient energy Ginzburg–Landau parameter L L LHS λ λ latent heat long-range order parameter left-hand side wavelength electron–phonon coupling parameter m M M μ μ μ  mass mobility Mendeleev number chemical potential shear modulus magnetic moment n(εi . J1 (x) Jn Jss Jhs .and first-order number of clusters per unit time that change from n to n + 1 steady-state flux in number-space of cluster sizes heat flux in solid and liquid (1D) flux of A-atoms magnetic exchange energy k k kB κs . T) N NAα αβ NAB N(k) N (t) Planck distribution number (of atoms) number of A-atoms on α-sublattice (point variable) number of A–B pairs with A on α and B on β (pair variable) number of quantum states with wavevector less than k vector of number occupancies of states at time t ν ν ν η η frequency Poisson ratio critical exponent for correlation length fractional change of lattice parameter with composition order parameter pi p pA probability of a state momentum partial pressure of vapor of element A ∼ . Jhl JA J(r1 − rj ) flux Bessel functions of zero.

e. L−J (r) 0 probability of A-atom on α-sublattice (point variable) probability of A–B pair with A on α and B on β (pair variable) pressure thermal pressure (from expansion against a bulk modulus) Péclet number interatomic.Notation xxii pαA αβ pAB P Pth P (r) M (r). Lennard–Jones potential quantum of magnetic flux hc/2e Q  Q Q θ (r) θ (r. electronic spin at site i entropy overlap integral configurational entropy vibrational entropy harmonic entropy entropy contribution from quasiharmonicity entropy contribution from anharmonicity electronic entropy entropy contribution from electron–phonon interaction magnetic entropy scattering function . t) D compositional wavevector 2π/λ momentum transfer in scattering quality factor of damped harmonic oscillator Heaviside function.g. [atoms cm−3 ] electronic density of states electronic density of states at the Fermi energy si S S Sconf Svib Sh Sqh Sanh Sel Sepi Smag  ω) S(Q. central-force potential Morse potential. 1 in the region. 0 outside phase of wavefunction in space and time Debye temperature rB rWS rl rk R R(Q) R∗  R n R R RHS ρ ρ() ρ(F ) Bohr radius rB = 2 /(me e2 ) Wigner–Seitz radius position of unit cell basis vector within unit cell number of right atoms on a sublattice of an ordered structure growth rate for compositional wavevector Q critical radius for nucleation position of atom center displacement after n jumps number of atoms in unit cell right-hand side density.

related to cube of de Broglie wavelength W Wij W↑ W (t) the number of wrong atoms on a sublattice of an ordered structure transition rate from state j to state i rate of increase of LRO parameter by jump of A from β to α-sublattice transition matrix for time interval t ω   j angular frequency number of states accessible to the system atomic volume configurations of a system with energy j ξ ξ {χi } χ correlation function length reaction coordinates susceptibility U βAα ≈ . y.Notation xxiii σ σ σ σgb σij surface energy per unit area electrical conductivity spin number (±1) energy per unit area of grain boundary stress t T Tc TC Tm TN T1 . for diffusion) u(x. ..and B-atoms U = (eAA − eBB )/4V Coulomb energy penalty for placing a second electron on a site in Hubbard model Grüneisen parameter for energy of electronic state j v V V V(r) VQ velocity interchange energy V = (eAA + eBB − 2eAB )/4 volume potential energy quantum volume. T τ time temperature critical temperature Curie temperature melting temperature Néel temperature sequence of temperatures such that T2 > T1 translation vector of real space lattice characteristic time (e. . . T2 .g. z) U ϒj displacement vector difference in chemical preferences of A.

Notation xxiv Y ψ(r) Young’s modulus wavefunction z z Z Z coordination number of lattice partition function of subsystem partition function Zeldovich factor .

and makes it easier to incorporate a higher level of detail from Part II into descriptions of phase transitions in Parts III and IV. Part I of the book covers essential topics of free energy. Likewise. the topics in Part I are the thermodynamics of how atoms prefer to be arranged when brought together at various temperatures.PART I BAS IC THERMODYNAMICS AND KINETICS OF P HASE TR ANSFORMATIONS The field of phase transitions is rich and vast. In essence. and a few classic phase transformations that have been part of the historical backbone of materials science. new methods from the larger field are now being applied to studies of materials. The topics in Part I are largely traditional ones. but many concepts and tools of phase transitions in materials are used elsewhere in the larger field of phase transitions. nucleation. but formulating the development in terms of statistical mechanics and in terms of the kinetic master equation allows more rigor for some topics. phase diagrams. 16:47:08 . diffusion. and continues to grow. This text covers parts of the field relevant to materials physics. and how the processes of atom movements control the rates and even the structures that are formed during phase transformations.

16:47:08 .

electrons that occupy states around the nuclei. which have charges that define the chemical elements. pressure. At the critical temperature of the ordering transition there is a switch from one curve to another. The free energy. Thermodynamic phase transitions in materials and condensed matter. but the functional form of this curve of Ford (T) is not the same as Fdis (T) for the disordered state at higher temperature. or perhaps the second derivative d2 F/dT 2 has a kink.1 Accompanying a phase transition are changes in some physical properties and structure of the material.1 What is a phase transition? A phase transition is an abrupt change in a system that occurs over a small range in a control variable. For many systems it is a challenge to find reliable models of the free energy. For thermodynamic phase transitions. extending from atomic nuclei to traffic flow or politics. typical control variables are the “intensive variables” of temperature. Order–disorder phase transitions are enlightening.6. which may have preferred orientations with respect to other spins. 1. we often find it useful to consider separately: • • • • nuclei. changes with temperature when the nuclear spins are aligned. and changes in properties or structure are the usual way that a phase transition is discovered. Temperature disrupts these delicate alignments. so thermodynamic analyses are not available. 16:48:40 . How and why do these groups of atoms undergo changes in their structures with temperature and pressure? In more detail. An ordered state of aligned nuclear spins may have the lowest energy. F. the weak energy of interaction between nuclear spins can lead to nuclear spin alignments. Our focus is on thermodynamic phase transitions in assemblages of many atoms. and 1 A brief review of free energy is given in Sect. or in one of the derivatives of the free energy function. the subject of this book. however. or magnetic field.1 Introduction 1.2. and may be favored thermodynamically at low temperatures. There is a very broad range of systems that can exhibit phase transitions. however. nuclear spins and their orientations. and thermodynamics favors a disordered nuclear magnetic structure at modest temperatures. Sometimes a phase transition involves only one of these entities. For example. and electron spins. occur when there is a singularity in the free energy function of the material. at low temperatures (microkelvin).

While these are indeed important subjects for study. Much of the fascination with phase transitions such as superconductivity is with the insight they give into the interactions between the electrons and phonons. Atoms interact in interesting ways when they are brought together. In condensed matter there are liquids of varying density. once we know the electronic structure of a material. or the electron charges and spins.2 Atoms and materials An interaction between atoms is a precondition for a phase transition in a material (and. it is often possible to understand many properties of a material. and entropy depend on temperature. Nevertheless. tend to position themselves to allow the lowest energy of the material. for example. and optical properties. electronic. Perhaps surprisingly. which are usually assumed to adapt continuously (“adiabatically”) to the positions of the nuclei. magnetic. These Cooper pairs can condense into a superconducting state at low temperatures. presents some concepts from this field that need not be grounded in materials phenomena. delving deeper into the first example of nuclear spin alignments at low temperatures reveals that the information about the alignment of one nucleus is carried to a nearby nucleus by the conduction electrons. Sometimes phase transitions involve multiple physical entities. Magnetic moments form structures of their own. In general. in turn. Other times the discrepancies prove interesting in themselves. it is easy to lose track of the forest if we focus on the interesting trees within it. and the electron density can show spatial modulations. Likewise. for having a material in the first place). although the mechanism is not fully understood today. which means that nuclei move around to let the electrons find low-energy states. In a study of phase transitions. Throughout much of this text. 16:48:40 . especially its chemical. and numerous types of crystal structures. and Chapter 20. chemical bonds are formed when atoms are brought together. in fact. 1. Perhaps electron charge or spin fluctuations couple the electrons in high-temperature superconductors. they are to some extent diversions from the main topic of phase transitions. energy. and these hyperfine interactions between nuclei and electrons are an interesting topic in their own right. and these large-scale features can occur in systems with very different microscopic interactions. Electrons of opposite spin can be coupled together by a wave of nuclear vibration (a phonon). the detailed interactions between the entities of matter are replaced with simplifying assumptions that facilitate mathematical modeling. Sometimes the essence of the phase transition is captured well with such a simple model.” meaning that it displays features that emerge from interactions between numerous individual entities.4 Introduction have spawned several creative methods to understand how an order parameter. A phase transition is an “emergent phenomenon. The nuclei. the same mathematical model reappears in explanations of phase transitions involving very different types of matter. The study of phase transitions has become a respected field of science in its own right. The energy of interatomic interactions is dominated by the energy of the electrons.

. Careful selections of variables to describe states of a material are critical for developing predictive theories of phase transitions. (Fortunately.1 Properties (mechanical.. and external fields. Some aspects of electronic energy are explained in Chapter 6.. each with its own characteristics. The first viewpoint considers only matter. or microstructure to the description of a material. defects such as dislocations. the second viewpoint adds energy to the definition of a material. nuclei.) Entropy is the other thermodynamic function essential for understanding most phase transitions in materials.” which is another way to find materials with appropriate properties. Electronic structure calculations are an important but large topic that extends beyond the scope of this text. and adjusting the intrinsic variables of temperature. Control of microstructure is the means for controlling properties of a material – this is the central paradigm of materials science and engineering (see Fig. but the structural arrangements of atoms are usually described by a hierarchy of features. electronic. Edisonian testing ignores the essence of a material. with interfaces between them and defects inside them.) Processing t Fig. as Edison pursued when finding filament materials for incandescent light bulbs. optical. Designing a microstructure is distinctly different from “Edisonian testing. Some types of accurate calculations are routine today. interfaces. One viewpoint is that a complete description of the structural features of microstructure is a full definition of the material. and other aspects are developed as needed to explain specific phase transformations in Part III. however. reliable calculations of electronic structure have arrived. is not materials science. Thermodynamic control tends to be the most reliable. Temperature is usually the most 16:48:40 .1 Box 1. there are many excellent references on electronic energy and chemical bonding of materials. Microstructure Materials are made of atoms. at least when the atom motions are fast enough so that equilibrium can be approached in a reasonable time.) Paradigm of materials science. For many materials. disorder.. Mixtures of crystals and phases. are the “microstructure” of a material.1.2 Atoms and materials 5 Composition Microstructure (phases. A second viewpoint adds excitations involving electrons. Thermodynamic control involves selecting the chemical composition. Both the energy and entropy of materials depend on the types of atoms and their mutual arrangements. 1. A direct processing-to-properties relationship. 1. and more will be commonplace soon. pressure. There are opportunities to control the states of matter through both thermodynamics and kinetics.1). and is not materials science.

Introduction

6

accessible means for controlling the state of equilibrium, and has served us well through
the bronze, iron, and silicon ages of humankind. Most of this book is concerned with phase
transitions that are driven by temperature.

1.3 Pure elements
1.3.1 Melting: a discontinuous phase transition
A liquid and a crystalline solid are fundamentally different owing to the symmetry of their
atom arrangements. In Chapter 14 when the Landau–Lifshitz criterion for second-order
phase transitions is developed, it is proved that melting must involve a discontinuity in the
first derivative of the free energy dF/dT at the melting temperature. For now, please accept
that it is not appropriate to use the same free energy function
F(T) = E − TS

(1.1)

for both the liquid and the solid phases. Instead, Fig. 1.2 shows two separate curves, Fs (T)
and Fl (T), for the solid and liquid phases of a pure element. The curve Fs (T) for the solid
lies below that of the liquid at T = 0 because the energy of the solid is lower than that of
the liquid.2 As shown on the y-axis, Es < El , and at T = 0 there is no entropy contribution
to the free energy. The free energy of the liquid decreases faster with temperature than
that of the solid because Ss < Sl .3 The two curves Fs (T) and Fl (T) cross at the melting
temperature, Tm .
In equilibrium, an elemental material follows the solid curve of Fig. 1.2 at low temperatures, and switches to the liquid curve above Tm . At Tm there is a discontinuity in the first

Free energy

E1

Liquid

Es

Solid

0

t

Fig. 1.2

Tm
Temperature

Free energy curves of a pure element when its atoms are arranged as a crystalline solid, Fs (T), or as a liquid, Fl (T).
2 Perhaps this is intuitive – the atoms in a crystalline solid have optimized their positions with respect to their

neighbors, and all atoms are in such optimal positions. The liquid has bond distances and angles that are not at
all uniform, meaning that some atoms are in configurations that are less favorable energetically.
3 Again, perhaps this is intuitive – there are more equivalent ways  of arranging the atoms in the liquid than in
a crystalline solid, so the entropy, S = kB ln , is larger for the liquid.

16:48:40

7

1.3 Pure elements

derivative of the free energy. A “first-order” phase transition occurs. This is rather catastrophic behavior, with the material changing from all liquid to all solid over an infinitesimal
range of temperature across Tm .
The free energies are equal at the melting temperature, Fs (Tm ) = Fl (Tm ). For constant
E and S, a consequence is obtained quickly
Fs (Tm ) = Fl (Tm ) ,

(1.2)

Es − Tm Ss = El − Tm Sl ,
El − Es
L
Sl − Ss =
=
,
Tm
Tm
where the latent heat, L, is defined with the difference in entropy at Tm  

L ≡ Sl (Tm ) − Ss (Tm ) Tm .

(1.3)
(1.4)

(1.5)

The latent heat is absorbed at Tm during melting, and released during freezing.
The expression for melting in Eq. 1.4 has ignored the temperature dependence of E and
also S. There are many effects to consider, and any quantitative thermodynamic calculation
of Tm might require a detailed understanding of Es , El , Ss , and Sl .4 Nevertheless, pure
elements do have well-defined melting temperatures. Another useful result is obtained by
working at temperatures very close to Tm . Over a small range of T, (1) the changes in E(T)
and S(T) for both solid and liquid phases are small and can be ignored, and (2) the two
curves in Fig. 1.2 can be approximated as straight lines. At temperatures very close to Tm ,
the difference in free energy of the two phases is
Fl (T) − Fs (T) =

L (Tm − T)
LT
=
.
Tm
Tm

(1.6)

Equation 1.6 is a linear relation between the difference in free energy between the solid
and liquid, and the difference in temperature from Tm . It turns out that the “driving force”
for freezing is this difference in free energy of the solid and liquid, so Eq. 1.6 tells us that
this driving force increases linearly with the undercooling below Tm . This seems correct
intuitively. Incidentally, the sign is correct in Eq. 1.6 – when T > Tm , the Fl (T) is more
negative (favorable) than Fs (T). The thermodynamics of melting (or freezing) illustrates
some general truths:
• The low-energy phase is favored at low temperatures.
• The high-entropy phase is favored at high temperatures.
• If the low-energy phase has a low entropy compared with the other phase, there will be
a phase transition at a finite temperature.

1.3.2 Structural symmetry and continuous phase transitions
When the symmetries of the high-temperature phase and low-temperature phase have special relationships, the phase transition can be continuous. Sometimes a discontinuous phase
4 For example, as atoms in a crystal vibrate from their periodic positions, the electron energy levels may change

somewhat, and there is an increasing entropy of vibrations. Liquids are even more complicated – Chapter 10
attempts to discuss their structure and degrees of freedom.

16:48:40

Introduction

8

1200

1200
Liquid

Liquid
1000

bcc (δ )

800
600

fcc
γ

400

α

Temperature [K]

Temperature [K]

1000

200

800
600

γ

400

α

200
dhcp (β)

0

t

Fig. 1.3

a

0

2
4
Pressure [GPa]

0
b

0

2
4
Pressure [GPa]

(a) Low-pressure region of the cerium T–P phase diagram. The solid phases at zero pressure are: fcc α, double hcp β,
fcc γ , bcc δ. (b) Two paths through the phase diagram of part a. One path has a discontinuous expansion of the fcc
unit cell, the other a continuous expansion.

transition can be changed to a continuous one by adding another thermodynamic variable
such as pressure. Figure 1.3 is a map of the phases of cerium metal, charted in a space
spanned by T and P, known as a “T–P phase diagram.”
Upon heating cerium at atmospheric pressure, it transforms between four different crystalline phases before melting. Our present interest is in two of them, the α- and γ -phases.5
Curiously, both have the fcc crystal structure, but they differ in volume by about 17% at
ambient pressure. Let us start at a pressure of 1 GPa (to avoid the β-phase), and follow a
vertical path in Fig. 1.3b that starts at low temperature, with cerium in the α-phase. Upon
heating to a temperature near room temperature, the α-phase undergoes a sudden expansion of its fcc unit cell as it transforms into the γ -phase. The how and why of this phase
transition is not fully understood today, but its existence is not in doubt.6
Obviously the crystallographic symmetries of two fcc phases are the same, even if they
differ in the sizes of their unit cells. With such a special relationship, we might ask if it is
possible to go from one to the other in a continuous way, without a discontinuous change
in volume. It turns out that this is indeed possible for cerium beyond a pressure of 2 GPa
and a temperature of 500 K. The T–P phase diagram of cerium metal has a “critical point,”
beyond which the two fcc phases are indistinguishable. It is possible to change from a firstorder discontinuous phase transition at lower pressures (or temperatures) to a second-order
continuous phase transition by taking a path around the critical point in the phase diagram.
If we start at T = 0 K, P = 1 GPa, and go up in temperature and pressure along a curved
path that goes to the right of the critical point in Fig. 1.3b, the volume of the fcc unit cell
will expand continuously.
5 Solid phases, typically with different crystal structures, are designated by Greek letters. The sequence of letters

tends to follow their appearance in a phase diagram, or their sequence of discovery.
6 Cerium is the first element on the periodic table with an f -electron, and its electronic structure is a challenge to

understand today. Its fcc–fcc transformation has attracted considerable attention, in part because of suggestions
that its outer electrons become delocalized when the γ -phase collapses into the high-density α-phase.

16:48:40

1.4 Alloys: unmixing and ordering

9

It is perhaps better known that the T–P phase diagram of water has a critical point,
beyond which the liquid and gas phases become indistinguishable.7 Evidently there can be
a symmetry relationship between atom arrangements in gases and liquids that allows such
continuous transitions.

1.4 Alloys: unmixing and ordering
“Alloying,” the mixing of two or more chemical species of atoms, brings additional degrees
of freedom to arrange the atoms, and more possibilities for energies and entropies. For millennia there has been interest in using temperature to alter the states of alloys. Today the
equilibrium states are mapped on a chart of temperature versus composition called a “T–c
phase diagram.” Transitions between phases occur at specific temperatures and compositions, and knowing the boundaries of these phase transitions is of both fundamental and
practical interest. Explaining T–c phase diagrams is an important goal of this text.
Figure 1.4 shows two basic types of transitions from a high-temperature phase (at top) to
a low-temperature phase or phases (at bottom). Note that all three atom configurations in
Fig. 1.4 are based on an underlying square lattice, simplifying the geometry. The problem
is reduced to one of atom positions on an “Ising lattice,” where an Ising lattice is a graph
of nodes connected by first-neighbor links. Each node (or site) is occupied by one atom of
Solid solution, disordered

kT > {–e}, e<
e<0
c = 0.5
L=0
High T
Unmixed (two phases)
eAA+ eBB < 2eAB

t

Fig. 1.4

c=1

c=0

Ordered
2eAB < eAA+ eBB

Low T

c = 0.5
L=1

Three alloy phases. At top is the expected high-temperature phase, a disordered solid solution. This phase is unstable
at low temperatures when chemical bond energies overpower the disorder caused by thermal energy. Two
low-temperature states are expected – an unmixed state of two phases (left), or an ordered compound (right). Only
one of these states is expected, owing to chemical preferences. (L is the long-range order parameter, defined later.)
7 Carbon dioxide also has a well-known critical point of its liquid and gas phases. When pushed beyond the

critical point, “supercritical” carbon dioxide is an effective agent for dry cleaning clothes.

16:48:40

Introduction

10

either species, with a probability dependent on the concentration. The links between nodes
are chemical bonds, which have energies that depend on the specific atoms in each pair.
The high-temperature phase in Fig. 1.4 is a random (or nearly random) distribution of
atoms called a “solid solution.” The two species of atoms are mixed together, or dissolved
in each other. Some A–B alloys mix better than others, much like “liquid solutions” of
ethanol and water, as opposed to oil and water. With many equivalent configurations, a solid
solution phase has a high configurational entropy, Sconf , favoring it at high temperatures.
At low temperatures, two phases of low Sconf are possible:
• The lower left shows an unmixed alloy, where the B-atoms separate from the A-atoms.
Unmixing is expected if atoms prefer their same species as neighbors more than they
prefer unlike atoms as neighbors. Unmixing may lead to small regions, rich in solute,
called “precipitates.”8
• The lower right shows the case of an ordered structure. Ordering is expected if unlike
atoms are preferred as neighbors because this ordered structure maximizes the numbers
of unlike neighbor pairs.
These unmixing and ordering transitions are usually more complicated than the extremes
depicted in Fig. 1.4. At intermediate temperatures there is usually partial unmixing, or
partial ordering.9 It is also possible, for example, for a solid solution to be in equilibrium
with a partially ordered phase. Nevertheless, the two phase transitions from low to high
temperature indicated in Fig. 1.4 are essential ones, and we consider them first.
It turns out that a simplified problem of a two-dimensional square lattice with fixed
first-neighbor interactions is the only problem with an exact solution for the free energy
through the critical temperature. The solution was reported by Onsager in 1944, and is an
impressive achievement, well beyond the scope of this text. Unfortunately, the Onsager
solution seems to be a special case, and efforts to extend it beyond the square lattice have
not been successful. The difficulty is that the problem is a multibody one, where the energies of all atoms depend on the configurations of all others. Consider an alloy that develops
chemical order. Although the free energy of the fully ordered state and the fully disordered
solid solutions are reasonably straightforward to understand, the intermediate states near
the critical temperature are not. As disorder begins to develop in an ordered structure, the
neighborhoods of individual atoms differ. The energy of each atom depends on its bonds
to the surrounding atoms, but the types of these atoms change with disorder. Sadly, it is
not enough to consider only an average neighborhood about a typical atom. Likewise,
the neighbors of the neighbors (some of which are the original atom and its neighbors)
influence the types of atoms in neighboring positions.
The interaction of an atom with its neighbors, whose identities depend on how they
interact with their own neighbors, leads to cooperative behavior in a phase transition.
In essence, with increasing temperature the alloy tends to hold its order locally, but the
quality of order over longer distances deteriorates more rapidly. At the critical temperature the long-range order is lost, but the short-range order is not, and persists to higher
temperatures. Short-range order is often best treated with its own, independent order
8 This is much like a precipitation reaction in a liquid solution, but of course the precipitates in solids cannot fall

to the bottom of a test tube under gravity.
9 An “order parameter” is usually necessary to describe how the quality of order changes with temperature.

16:48:40

1.5 Transitions and transformations

11

parameters. A consistent treatment of short-, intermediate-, and long-range order parameters is a daunting task, however. Approximation methods have been developed, and their
strengths and weaknesses have become important topics to understand. The more abstract
methods of Chapter 20 treat cooperative behavior more properly, but atoms may move too
sluggishly to achieve equilibrium across the large spatial distances that become relevant
near a critical temperature.

1.5 Transitions and transformations
We use the words “phase transformation” to mean the life cycle of a “phase transition,” but
the literature is less consistent with usage. A phase transition originates with the singular
character of the free energy, but the free energy may not be the only important issue in
a phase transformation. The effects of kinetics and mechanisms of atom movements are
important for understanding phase transformations, often giving them unique character. For
studies of phase transitions, on the other hand, kinetic phenomena are either unimportant,
or are considered impediments to understanding. In materials there are many types of phase
transformations, and major ones are discussed in Part III of this book. Section 1.5 presents
a few concepts pertinent to phase transformations.

1.5.1 Diffusional and diffusionless
One distinction between phase transformations in solids is if diffusive atom motions are
involved as in Fig. 1.5a, or if there is a change in structure by a cooperative displacement
of all atoms, such as by the shear shown in Fig. 1.5b. Most chemical phase transformations

t

Fig. 1.5

a

b

c

d

Illustrations of four processes in time. In each pair the configuration before is to the left, and after is to the right.
(a) Atom motion by diffusion, in this case by one atom exchanging sites with a vacancy. The vacancy remains to jump
again, likely in a different direction. (b) Atom motion by cooperative shear, changing a square crystal into a
rhombohedral one. (c) Phase transformation by nucleation and growth, where a new phase forms discontinuously
from the parent phase. (d) Phase transformation by continuous unmixing (ordering is possible too), where a gradual
modulation grows to high amplitude. This is typical of spinodal decomposition.

16:48:40

Introduction

12

in solids involve diffusive motions of atoms, and the chemical unmixing or ordering transformations of Fig. 1.4 require diffusion because individual atoms must move to optimize
their local chemical environments. In contrast, “martensitic transformations” described in
Chapter 15 occur by cooperative shear motions of all atoms.
On the mesoscale, or on the microstructural scale, diffusional phase transformations can
proceed in two essential ways. Figure 1.5c shows the nucleation of a new phase (on left),
where embryos appear and then grow in size (Chapter 11). As they grow, the regions of
new phase have essentially the same chemical composition and same crystal structure at
all sizes. Figure 1.5d depicts an alternative, continuous process where the unmixing occurs
gradually, as in “spinodal decomposition,” for example (Chapter 12). Phase transformations that occur by nucleation and growth seem much more common than those that occur
in a continuous process, in part because they are easier to observe.

1.5.2 Examples of continuous and discontinuous transformations
Differences between continuous and discontinuous unmixing transformations are seen in
the experimental images of Fig. 1.6, obtained by position-sensitive atom probe microscopy.
Each dot marks the detection of one atom of a specific element. The atoms in Fig. 1.6a are
cobalt atoms in a material originally prepared as a solid solution of Cu–1%Co. Cobalt is
largely immiscible in Cu, and the unmixing transformation is seen in the four images of

10 min 5 nm

120 min 5 nm

30 min 5 nm

1440 min 5 nm

a
b

t

Fig. 1.6

5 nm

5 nm

Results from position-sensitive atom probe microscopy measurements on unmixing in two alloys. Both alloys were
prepared as random solid solutions by rapidly cooling from a high temperature, and heated at low temperatures
where the atoms had some diffusive mobility. Each image is from a different part of the material. (a) Unmixing of Co
atoms from a Cu-rich phase during heating at 723 K for times as indicated. (b) Unmixing of Cr atoms in an alloy of
Fe–Cr–Ni after heating at 673 K for 10,000 hours. Image at right was constructed from the three-dimensional data by
sampling the atoms in neighboring roxels (i.e., 3D cubic pixels). Reprinted, with permission, from [1].

16:48:40

13

1.5 Transitions and transformations

Fig. 1.6a, obtained after different times of the process. Even in the earliest stages there is
clustering of cobalt atoms into distinct zones – cobalt unmixes from copper by a discontinuous “nucleation and growth” transformation. There is an energy associated with the surface
around the cobalt atoms, and the minimization of the surface energy causes the cobalt-rich
precipitates to have approximately spherical shapes. On the other hand, Fig. 1.6b shows a
continuous unmixing process called “spinodal decomposition.” Chromium atoms have partially separated from the surrounding iron-rich phase in a stainless steel. Time snapshots
are not shown, but notice how the chromium atoms are in diffuse zones without sharp
boundaries. The image at right in Fig. 1.6b was obtained by defining a local composition
of 30% to help identify the shapes of these zones. The zones are sinuous with a periodicity
of 10 nm or so. They are often described as wavelike composition modulations through the
alloy. In Chapter 12, the diffuse boundaries of the chromium-rich zones will be characterized by an energy that is not a surface energy, but an energy proportional to the square of
the composition gradient.

1.5.3 Vacancies and interstitials
Two or more atoms cannot occupy the same site on a crystal lattice.10 A site is usually
occupied by one atom, but if all sites are occupied it is difficult to understand how atoms
can move around. Atoms do not move by exchanging positions because the energy barrier
is too high. As illustrated in Fig. 1.5a, sometimes an atom is missing from a site. This is
called a “vacancy.” A vacancy can exchange sites with a neighboring atom, usually a firstnearest neighbor, of which there are four in Fig. 1.5a, and in Fig. 1.7, which depicts a few
more steps in the motion of a vacancy. It is important to understand that if a neighboring
atom jumps into a vacancy, the vacancy is not annihilated, unless it moves to a surface or
other suitable location such as a grain boundary. Within the bulk of a crystal, the vacancy
moves from site to site, each time exchanging sites with an atom. The vacancy mechanism
is the usual way that atoms move in metals, and it is a common mechanism for covalently
and ionically bonded materials. The concentrations of vacancies and the activation energies
for moving them are important parameters in atomistic theories of diffusion.
Interstitial atoms are usually small atoms that fit in the gaps or “interstices” between
larger atoms on a crystal structure. For example, small carbon atoms fit into the gaps
between iron atoms on bcc or fcc crystal structures. Usually these interstitial atoms move
rapidly from site to site, in part because the interstitial sites are mostly vacant, but also

t

Fig. 1.7

Series of vacancy jumps in a square lattice, from left to right. All atoms are the same species, but the darkness of
shading indicates how recently the atom has interchanged sites with the vacancy.
10 There are cases of “dumbbells” of two atoms on a site, but these are relatively rare.

16:48:40

Introduction

14

because the energy barriers are usually low. Hydrogen atoms move particularly rapidly in
metals, for example.
Sometimes an atom can be an interstitial species in its own crystal lattice. The diamondcubic lattice of silicon is fairly open, and silicon atoms can move between crystal sites and
interstitial sites. Diffusive atom movements by such an “interstitialcy mechanism” occur
in a number of materials. When an interstitialcy mechanism is active, diffusion is often
“anomalously” fast.

1.6 Brief review of thermodynamics and kinetics
1.6.1 Partition function
Throughout the text a physical system is modeled with “state variables” that parameterize
essential features of materials structure. These state variables (simple ones are chemical
composition or an order parameter) are selected so the energies of the system {εi } can be
calculated. With these energies, a partition function, Z, can be constructed as
Z≡ 

e−εi /kB T .

(1.7)

i

For a particular temperature, the partition function depends on how probability is distributed (“partitioned”) between states of different energy. All states of the same energy
make the same contribution to the partition function, and if there are more states of a particular energy, this energy carries more total weight. Importantly, it is true that the partition
function contains all thermodynamic information about the state variables of our model
system.
Since each state with the same energy εj has the same Boltzmann factor, e−εj /kB T , it is
often convenient to rewrite the partition function as
Z=  

j e−εj /kB T ,

(1.8)

j

where j is the number of states having the energy εj . For a system with a fixed energy εj ,
the probability of each state with this energy is pj = 1/ j and the entropy is11
11 This is consistent with the expression for the total entropy 

Sj = −kB

j 

j

pj ln pj

(1.9)

when each pj is equal to 1/ j . Equal probability for states of equal energy is a central tenet of statistical
mechanics, and may be expected for systems in equilibrium.

16:48:40

15

1.6 Brief review of thermodynamics and kinetics
Sj = kB ln j .

(1.10)

We will usually consider situations where the number of atoms is conserved, and our
physical system is in thermal contact with a large heat bath of temperature T. Consider
the probability of our system being in state i with energy εi , or in state i with energy εi .
The entropy of the system plus heat bath is dominated by the entropy of the heat bath. The
entropy of the heat bath, designated by the states j, changes by the amount
∂S
∂E
∂S
.
Sj (E − εi ) = S(E) − εi
∂E
Sj (E − εi ) = S(E) − εi

(1.11)
(1.12)

With the entropy expression Eq. 1.10, recognizing ∂S/∂E = 1/T, and using the fact that
the probability of a state pi is proportional to the number j for the dominating heat bath 
j
pi
=
, 

pi 
j

(1.13)

pi
eSj /kB
= S  /k ,
pi
ej B
pi
eS/kB e−εi /kB T
e−εi /kB T
= S/k −ε  /k T = −ε  /k T .
pi
e B e i B
e i B

(1.14)
(1.15)

The Boltzmann factor for an individual state i can be normalized by the sum of all other
Boltzmann factors (Eq. 1.7) to give the probability
e−εi /kB T
.
(1.16)
Z
With the thermodynamic probabilities of all states, it is straightforward to calculate
thermodynamic averages, such as for the energy 

ε

=
pi εi ,
(1.17)
pi =

i

ε

(1. = kB T 2 ∂ ln Z . one part will do work on the other until the total free energy is a 16:48:40 . The Helmholtz free energy.6.2 Free energy The historical. When two parts of a system are brought together and interact at the same temperature.19) 1. viewpoint of “free energy” is the amount of energy available in a system to do external work such as pushing a piston in a steam engine. F = E − TS. the Helmholtz free energy. can be obtained from the partition function as F = −kB T ln Z . ∂T (1. is the internal energy minus the entropic contribution that cannot perform work. F ≡ E − TS. but still correct.18) Finally.

24) Helmholtz or Gibbs? An expression for E(S. as explained in Chapter 8.000 atmospheres in P. however.N ∂T N.N .23 provides these valuable relationships       ∂F ∂F ∂F μ= . (1.22 and 1. • The thermodynamic potentials are in the four boxes. 1. (∂S/∂P)T = −(∂V/∂T)P . V. and it has other uses. (∂E/∂N)S.T ∂V T. S are equilibrium values. 1.20 Here μ. F(T. 1. P.22) Substituting Eq.V A comparison of Eqs.22 uses differentials of its natural variables (with positive signs). The term μdN can be added later. it is simpler to ignore effects of pressure and focus on the role of T by using the Helmholtz free energy F. i.21) dF = μdN − PdV − SdT . the Gibbs free energy is the most convenient thermodynamic potential if we have control over T and P.23) dF = ∂N V. If G does not depend significantly on P. (1. The thermodynamic square of Fig.N = T. P=− . • Maxwell relations are formed by going down the sides with a − sign. V. is dF = d(E − TS) = dE − TdS − SdT . S=− . (1.2 (1. N) or G(T. Minimizing E with respect to S is a way to find conditions of thermal equilibrium. and recognizing (or defining) (∂E/∂S)V.8 can guide the selection of a thermodynamic potential. or across the rows with a + sign. T = (∂E/∂S)V.. Minimizing free energy functions will be our primary method for finding states of thermodynamic equilibrium.g. 12 Obtained by differentiating the fundamental relationship for E(S. 16:48:40 . dF = −SdT − PdV. but the role of temperature appears through the functional relationship between E and S.T ∂V T. V.Introduction 16 minimum.N = −P.V = μ. T) are expressed with variations of its independent variables N. which is less common in practice. e.e. (∂P/∂T)V = (∂S/∂V)T . • Computing total differentials such as dF of Eq. N) is the fundamental equation of thermodynamics. and T is a natural variable for the Helmholtz or Gibbs free energies. for most solids a change of 100 K in T causes approximately the same change in G as a change of 5. P.21 into 1.N ∂T N. N). and quantities in the opposite corner. N). V.20) The thermodynamic identity12 is dE = μdN − PdV + TdS . (∂E/∂V)S. The total differential of the Helmholtz free energy. It is usually easier to work with T directly. We will usually use F because. F = E − TS. In general. with their natural variables to their sides. 1. and T       ∂F ∂F ∂F dN + dV + dT . ∂N V.V Box 1. (1. The variations of F(N.. V.

. n−1. but Eq.25. . . . we allow the jumping of atoms between adjacent bins. and the equation is worthy of respect. .9. .9b. For kinetics. but it is quite natural to understand and often easy to use. Two types of jumps. numbered at bottom. a Nn–2 Nn–1 n–2 n–1 Nn n Nn+1 n+1 Nn+2 n+2 Wn+1. n+2.n+1 t Fig. and the third expression is handy for obtaining the entropy from Eq.6 Brief review of thermodynamics and kinetics 17 t Fig.1 the kinetic master equation is used to obtain the diffusion equation. n. with definitions in Fig.9 b n n+1 (a) A row of bins for atoms. the bin will gain Nn atoms as given by Eq. 1. left and right. Figure 1. In Sect. Nn+1 . 1. and it will prove its worth in many analyses of kinetic phenomena. 1. in our case atoms. 1. and the probability that an individual atom leaves the bin in the time interval t. and with contents at top.8 Thermodynamic square with mnemonics. the inflow of atoms 13 This is true if F scales with N. The ideas of the kinetic master equation are illustrated with Fig. 16:48:40 . are depicted between bins n and n + 1. Beneath them are container numbers . . . The outflow of atoms from the nth bin is the product of the number of atoms in the bin at the time.6. .19. 1.24 must be used more carefully if surface energy is important.13 the second is used in Chapter 8 to study pressure. Likewise. We will often use the first expression to treat the chemical potential as the free energy per atom. The kinetic master equation states that if more atoms enter a bin than leave it in a time t. Nn−1 . The name is grand. . Nn . n Wn.9a shows a row of containers for entities. Their contents are listed above as .3 Kinetic master equation The “kinetic master equation” is the most general kinetic equation. . 1. (b) Flow of atoms between two adjacent bins. 1. for example.1. with definitions of rates. 1. 3. .

5 + 0.2 Suppose the internal energies of the solid and liquid phases vary with temperature linearly as E(T) = E0 + T dE/dT.n (t) is proportional to t when t is small. where E0 is the internal energy at T = 0.3 (a) Derive Eq. 1.5 Suppose an alloy of composition 0. why is this inconsistent with an unmixing tendency? (c) Suggest a correction to the assumption that F = F(c). 1.e. and dE/dT is a constant for a particular phase.9. 1.26) where the wavelength of the modulation is λ. i. (1. Using the definitions in Fig.Introduction 18 into the bin n is the product of the number of atoms in the surrounding bins times the probability that each atom goes from the bin n to n. and explain how the difference in entropy of the two phases affects the driving potential. F = F(c). 1. 1.n (t) Nn (t) . and develops concentration modulations of the form c(x) = 0. does the total free energy depend on λ? (b) Suppose λ becomes quite short. 1.e.4 is recovered if dE/dT is equal for the solid and liquid phases. L = El − Es ? (b) Show that Eq. (a) How does this modify Eq. where a is the interatomic spacing.n (t) Nn (t) − Wn . λ = 2a. i.5 sin(kx) .6 in the text. i.) If the result is an ordered alloy.6 Draw several frames of a cartoon showing how a vacancy mechanism with firstneighbor jumps can cause neighboring atoms 1 and 2 on a simple cubic lattice to 16:48:40 .25) + Wn.4. Explain why the driving potential for a firstorder phase transition would be proportional to the difference in temperature from the critical temperature. (b) Assume that the driving potential for a phase transition is the difference in free energy between the two phases. do your best to explain why it is difficult to parameterize the state of order in an alloy with partial order. 1.1 Explain the basic paradigm of materials science. Problems 1. (Credit will be given generously. 1.e.. Fl (T) − Fs (T) = L (TTmm−T) . but please think about this problem before you write.. Assume that λ is much smaller than the size of the alloy.) 1. (a) If the local free energy depends only on composition. the kinetic master equation is   Nn (t) = (1. n where the Wn.5 undergoes chemical unmixing by a continuous process.4 In one paragraph. and k = 2π/λ. Is the alloy better described as an ordered structure or a chemically unmixed structure? (Hint: Draw it.

8 Prove Eq.e..19.7 (a) Is it possible for chemical ordering to occur by a shear transformation (i. 1. 1.19 Problems change from an orientation with 1 on the left of 2 to an orientation with 2 on the left of 1. 16:48:40 . can the order parameter change during a simple shear transformation)? (b) Does the configurational entropy change during a shear transformation? 1.

4). or negligible. 2. but celebrated as a way to identify phenomena common to many phase transitions.2 Essentials of T–c phase diagrams Chapter 2 explains the concepts behind T–c phase diagrams. and with chemical compositions that minimize the total free energy of the alloy.g. and these phases are present in the amounts f . e.1 Overview of the approach Temperature promotes disorder in a material. The emphasis in this chapter is on deriving T–c phase diagrams from free energy functions F(c. can show similar phase transitions. 16:48:42 . The opposite case of a preference for unlike atom neighbors leads to chemical ordering at low temperatures.2 The constraint of solute conservation is expressed easily as the “lever rule. 2 The usage of phase diagrams is deferred to Chapter 5. the individual atoms or molecules prefer their like species as neighbors (see Fig. Because the key features of phase diagrams can be obtained with general types of interactions between atoms. both oil in water and iron in copper. T).1 A T–c phase diagram displays the phases in thermodynamic equilibrium. we must be wise enough to know the predictive power available at different levels of generalization. which selects the equilibrium phases at T from the F(c) curves of the different phases. For binary alloys. Such “emergent” behavior can be missed if there is too much emphasis on the electronics of chemical bonding and atom vibrations. which are maps of the phases that exist in an alloy of chemical composition c at temperature T.. Nevertheless. In these unmixing cases. and continuous solid solubility phase diagrams. peritectic. systems with very different types of chemical bonding. free energy functions can be calculated with a minimum set of assumptions about how different atoms interact when they are brought together. 1. which requires the definition of an order parameter. Some features of ternary alloy phase diagrams are also discussed. but the chemical bond energy favors crystals at low temperatures. This information 1 Pressure is assumed constant. These generalizations of chemical interactions should not be lamented for their loss of rigor. If atoms occupy sites on a lattice throughout the phase transformation. the shapes of F(c) curves and their dependence on temperature are used to deduce eutectic. L. favoring higher-entropy phases such as liquids.” The minimization of the total free energy leads to the more subtle “common tangent construction”.

Sometimes the pairwise atomic bond energies can be tuned as free parameters to account for phase transitions in a specific material. A good parameterization of atom positions may require more detail than is possible with a simple model. a minimal atomistic model is constructed to calculate thermodynamic functions and predict the equilibrium phases in alloys at different combinations of T and chemical composition c. This is a large topic. For alloy thermodynamics we need expressions for E(c) and S(c). for which the Helmholtz free energy is   (2. E and S. The thermodynamic functions of the alloy.4 Nevertheless. depend on the spatial arrangement of the atoms in the alloy. but we must be aware of the two types of risks it entails: 1. and will grow in importance to materials science. can be 3 This restates the paradigm that the atom arrangements in a material determine its properties. the configurations of atoms). however. 4 These methods are powerful. Part II of this book addresses these two issues in more detail.1 Overview of the approach 21 alone is sufficient to predict a phase change with temperature (melting). as is the way configurations are counted to obtain the configurational entropy Sconf (c). E. Convenience for calculation is a virtue and a priority for this chapter. such as density functional theory with planewave pseudopotentials. a good parameterization of atom configurations can accommodate more advanced analyses. typically as a sum of energies of chemical bonds between the nearest-neighbor pairs of atoms. often local ones. T) . If E(c) and Sconf (c) depend only on local atom configurations in a crystal.3 In what follows. and sometimes the alloy thermodynamics can be extrapolated successfully into regions where no experimental data exist. In a first approximation. calculating the energy or entropy from these parameterizations may require more sophisticated methods. too. 2.2.. 16:48:42 .1) F(c. 5 Or. analogously for an order parameter. Nevertheless. The entropy of an alloy also depends on atom configurations. the electronic energy E(c) is assumed to be independent of temperature. because the energies of the electrons in a material are set by the positions of the nuclei (i. Here we obtain E from local atomic arrangements.5 Two important T–c phase diagrams. For solid solution phases. the sublattice occupancies are assumed random. and this is the essence of the present chapter. but adding a bit more detail about specific atoms rewards us with considerably more information about alloy phase diagrams. Even if the parameterizations of atom positions are excellent. it is also convenient to assume statistical randomness of atom occupancies of the sites on the lattice. This book will not emphasize methods to calculate the energy of the phases. unmixing and ordering. it is efficient to use a fixed lattice and pairwise chemical bonds. as is done today by electronic structure calculations. including thermodynamic properties. This minimal model successfully predicts the main features of T–c phase diagrams for binary A–B alloys (A and B are the two species of atoms). the simple models developed in this chapter are often useful semiquantitatively. T) = E(c) − T Sconf (c) + Svib (c. and are useful benchmarks for assessing the value added by more sophisticated treatments.e. and deserves a course of its own.

so this approach has an excellent record of success. i. T). Usually the energy and entropy of materials originate from local characteristics of atom configurations. At modest temperatures. Sel (c. T). F(c) = E(c) − TS(c).” One approach is to use probabilities of local “clusters” of atoms. (There are cases of chemical unmixing in the liquid phase. the equilibrium phases and their chemical compositions depend in detail on the free energy versus composition curves Fξ (c) for each phase ξ . Smag (c.2. and the entropy from disorder in electron state occupancies. then tetrahedra. the electronic energy.e. For most materials this is a gas of isolated atoms. With these assumptions. In this chapter we ignore the entropy caused by the thermal vibrations of atoms. Usually there are chemical unmixings. the “point approximation. there is competition between multiple phases that cannot be placed naturally on the same lattice. Svib (c. 2. How accurately can we account for the energy and entropy of the material? For example. so these contributions will have to be assessed later. for which the thermodynamics of Eq. but at lower temperatures most alloys form a liquid phase with continuous solubility of A.22 Essentials of T–c phase diagrams obtained readily by minimizing such an F(c. and the different chemical compositions frequently prefer different crystal structures. listed from high to low temperature. need not be confined to neighboring pairs of atoms.) • At low-to-intermediate temperatures. with considerations of the Helmholtz free energy.1 Free energies of alloy phases Alloys often transform to different phases at different temperatures. The chemical unmixings may not require 16:48:42 .. Here.19 is again relevant.and B-atoms. Minimizing the total free energy by varying the compositions and fractions of the phases requires the “common tangent” algorithm that allows us to complete the set of the five main types of T–c phase diagrams for binary alloys. so each phase has a different F(c.2 Intuition and expectations about alloy thermodynamics 2. to numbers of atom pairs (pair). 2. More generally. T). T). The energy of delocalized electrons in a box or in a periodic crystal is introduced in Chapter 6. the two general risks listed above can be stated more specifically: 1. however. are some typical phases that are found for an alloy system (“system” meaning a range of compositions for specified chemical elements). T) on an Ising lattice. typically Svib > Sconf . for each phase: • The phase of maximum entropy dominates at the highest temperatures. the entropy from disorder in magnetic spins. working up systematically from chemical composition (point). 2. Chapter 7 goes beyond the assumed random environment of an average atom. which provides the thermodynamic E. We have already mentioned the liquid at high temperatures and ordered structures at low temperature.

a precise composition of A2 B3 ) because some spread of compositions may provide entropy to make off-stoichiometric compositions favorable (e. 16:48:42 .9a to that in Fig. 2. and likewise few A-atoms when c is near 1. F = E − TS. giving the  for Eq. • When c is near 0. so the statistical mechanics of harmonic vibrations needs significant modifications (Chapter 24).g. so different values of F at c = 0 and c = 1. the equilibrium state for a general chemical composition is a combination of crystalline phases. Svib is typically an order-of-magnitude larger than Sconfig (compare the maximum entropy in Fig. and normal modes of vibration interact with each other. procedures to count states explored with thermal excitations. however. 1. 7. and this is developed for independent harmonic oscillations in Sect. such as atom vibrations or electronic excitations. so each solute atom is in the same chemical environment – it is fully surrounded by solvent atoms.g. The reader should know that at a temperature of 1. Nevertheless.10. atoms usually do not have enough mobility to form these precise structures. and count the configurations of a system with equivalent macrostates. however. usually with precise stoichiometries.000 K or so. • At the lowest temperatures. vibrations change their frequencies.3). If atom mobility becomes extremely rapid while some vibrations become very slow. At high temperatures. and a high degree of long-range order.2 Intuition and expectations about alloy thermodynamics Box 2. In such dilute alloys. 6 At low temperatures. so the difference in entropy caused by vibrations is not always dominant in a phase transition.1 Configurational and dynamical sources of entropy We take the statistical mechanics approach to entropy. precise stoichiometries (e. Fig..2 Thermodynamics of solutions: qualitative Assuming good atomic mixing between two chemical elements in an alloy. The pure elements A and B have different melting temperatures. This separation works well for the thermodynamics of crystals at modest temperatures.6 2. Svib usually does depend on atom configurations. 10..23 2. there are only a few B-atoms in the alloy.1 shows a typical curve of Helmholtz free energy. • On the other hand. It is possible for Svib to be similar for different phases. do depend on temperature. for example. as when a liquid changes its viscosity with temperature. its reliability may be in doubt. so some chemical disorder is often observed. A2−δ B3+δ ). 2. Some salient points are: • There are generally different bonding energies for the pure elements A and B. These often correspond to crystal structures that have unique sites for the different atom species.8. For counting: • Procedures to count atom arrangements (or spin arrangements) are independent of temperature.2. Each solute atom therefore has the same bonding energy. versus composition c. we expect the solute atoms to be isolated from each other.

Likewise. we expect the entropy to be highly sensitive to composition when c is near 0 or 1 because in a nearly pure material. 6. with the first two factors being the most important.8 Concentration (c) 1.4 0.2. the number of solvent atoms with solute neighbors increases linearly with c.Essentials of T–c phase diagrams Free energy 24 0. 1. atom mass.e.2 0. The Hume–Rothery rules of alloy phase formation are based on these properties of individual atoms: (1) electronegativity. the configurational entropy per atom changes as Sconfig (c) −kB c ln c. We therefore cannot expect too much accuracy from rules based on atomic properties when predicting different alloy phases. and on the entropy. (2) metallic radius.0 t Fig. Another reason is that at c = 0. in more detail these atomic properties are: 7 In the case of nonlocal “metallic bonds. and were used as predictive tools for metallic alloys in the mid.7 • From the entropy expression S = kB ln . (For vibrational entropy. F(c) should no longer change with composition (i.6 0.3. dF/dc = 0).3 Hume–Rothery rules Hume–Rothery rules give conditions for solid solubility and compound formation. and the overall bonding energy increases linearly with the solute concentration. Nevertheless. there are numerous possible sites for the few solute atoms. and this number decreases smoothly away from the equiatomic composition. 2.to late-twentieth century [2]–[4].5 there is a maximum number of unlike pairs. the donation of an electron to the band structure also affects the energy linearly with solute concentration in the dilute limit. with features described in the text. 2.2.0 B Typical F(c) curve for a solution. These rules are intuitively reasonable. which dominates over the proportionality to c that is followed by the energy (and vibrational entropy). At small c. the outer bonding electrons have different energies and spatial distributions in different crystals. • The value of F near the middle of the composition range depends on the energy.1 A 0. The energy dependence can make positive or negative contributions to F at intermediate compositions. leading to the two structures at the bottom of Fig. 16:48:42 . should be added to this list. There is always a downwards slope of F(c) near c = 0 and c = 1 owing to Sconfig (c).” described in Sect.) Although atoms retain some of their identity when assembled into a solid. especially at low T. One reason is that F(c) must turn around to accommodate its opposite slopes near c = 0 and c = 1.. a fourth.4. especially at high T. • Somewhere in the middle of the composition range. and (3) electron-to-atom ratio.

16:48:42 . as shown in Fig.4 Metallic radius [Å] 1. and could be used to predict changes in screening.6). Darken and Gurry suggested drawing an ellipse with an extent of 0. 6. The real t Fig.9 • Electron-to-atom ratio. 6. 10 There are Hume–Rothery rules for compound formation. For their ease of use. and have much more reliable information about solid solubility. and tend to have a higher solubility than the elements that fall outside the circle.2. because conduction electrons are highly effective in screening the charge disturbances that occur with electron transfer.6). The elements within the circle satisfy the Hume– Rothery rules. • The difference in metallic radius of the atoms shall not differ by more than 15%. but these tend to be less reliable and less general than for solid solubility.0 Si 1. for example. the Hume–Rothery rules for solid solubility are a best buy.2 for the solubility of some main group elements in iron.1.” meaning something like 5% solubility at 1. which provides information on the strength of the bond and its stiffness (see Sects.10 Of course. 9 A local compression of electron wavefunctions raises their kinetic energy.2 2.6 + Mg 1. Electronegativity (Pauling) These rules can be used conveniently by plotting the elements of interest on a map of electronegativity versus metallic radius [4]. The e/a ratio is expected to be a useful parameter if the rigid band model (Sect. 6.2 As 2. This is a measure of the tendency for electron transfer to an atom during chemical bonding.2. 8 In metals this is only a part of the picture.6 1. but changes in the screening may scale with differences in electronegativity.4 Sb Ge In Zn 1.2 Intuition and expectations about alloy thermodynamics 25 • Electronegativity.8 Sn Fe Ga 1. Inserting a larger solute atom into a crystal of smaller atoms causes a local compression in the host crystal. 2.1. The Hume–Rothery rules for solid solutions are stated as conditions for having “extended solid solubility.8 “Darken–Gurry plot” for extended solid solubility of some main group elements in iron.2.3 and 6. too.2 Al 1. Instead of a rectangle of fixed width in electronegativity and metallic radius. phase diagrams of most binary alloys are now known. 2. This accounts for nonlocal Fermi surface effects in metals (Sect.4 in electronegativity and 15% in radius.0 1.8 • Metallic radius. Interatomic forces generally become stiffer under compression. so alloying with large atoms tends to reduce the vibrational entropy.000 ◦ C: • The Pauling electronegativity of the elements shall not differ by more than 0.4. Magnesium has a notoriously low solubility in iron. including its temperature dependence.7) is reasonably reliable. and Hume–Rothery and others treated this effect as an elastic strain energy.

the sequence runs up and down the columns of the periodic table. There are a number of semiempirical approaches to estimate heats of formation of compounds from their chemical elements. One in widespread use is that of A. F(c). the Miedema heats of formation are now fairly reliable. Figure 2.3 Free energy curves. 2. If M is plotted on the x-axis.26 Essentials of T–c phase diagrams Box 2.” M. Miedema et al. after a number of years of tuning the parameters in the model. 2. solute conservation. who boldly proposed a new classification scheme for all elements in the periodic table. Furthermore.1 Free energy versus composition Familiarity with free energy versus composition curves. such as the NaCl structure for an equiatomic AB alloy. The ordering of the elements begins with He at 1. with isoelectronic elements placed near each other in the sequence. and one crystalline solid phase of continuous solid solubility. In general.. there are no ordered compounds for AB alloys if both elements have M < 50. Nevertheless. jumping to the alkali elements with 7 at Fr through Li at 12. is essential for understanding alloy T–c phase diagrams. and information on the atomic size to handle the overlaps of atoms in a solid [6]. a work function to parameterize how easily the outer electrons could be removed from the atom. The interest in this approach is from the “structure maps” that it generates naturally. There is a tendency for specific types of compounds. FS (c) and FL (c). are similar to the curve of Fig. They also show barren regions – for example. the more electronegative elements (F at 102) are higher in this sequence of “Mendeleev numbers. There was hope that these structure maps would assist efforts to design alloys by chemical modifications. The theoretical justification for this approach is not rigorous. 16:48:42 . constrained by kinetic processes. and can be useful for estimating solubility tendencies and the stability of compounds.3. who formulated a heat of formation in terms of an electron density at the outer parts of an atom in a solid. A simple but important phase diagram is for a binary alloy with only two phases: a liquid phase of continuous solubility. Tabulations of Miedema heats of formation of different alloy compositions are available. value of the Hume–Rothery rules is that they give us intuition about the importance of electronegativity and metallic radius in the thermodynamics of alloying. modifications of these analyses can be useful for systems out of equilibrium.R. Approximately. The two free energy versus composition curves.3 shows a typical case.2 Other rules of thumb and empirical trends An interesting idea for understanding compound formation was developed by David Pettifor [5].1. but this has proved only partly successful. to be grouped in specific regions on this grid. and the other inert gases through 6 at Rn. Nevertheless. but they differ in detail. and M is also plotted on the y-axis. and the lever rule 2. all binary combinations of elements are available as points on an x–y grid. the structure maps for AB and A3 B alloys do offer an interesting framework to consider chemical effects on crystal structures and bonding. Pettifor’s arrangement assigns similar values of M to elements with similar chemical properties.

8 1.6 0.3 0. because SL is large. The −TS part of the free energy for the liquid. At the highest temperature. 2.3 require further analysis. for any composition. T1 . Consider the material depicted in Fig. 2.3 Free energy curves.2 S 0.4a. which has an overall composition of c0 = 0. T4 .3. If it forms a mixture of two phases (here 16:48:42 . the liquid curve in Fig. as expected because the liquid has the higher entropy. The alloy is a liquid at all compositions. solute conservation.27 2.2 Conservation of solute and the lever rule The conservation of solute provides a constraint between the fractional amounts of phases and the chemical compositions of the phases. F. It is not simply a matter of picking the lower free energy curve at a particular composition because the alloy can unmix chemically. 2.0 Concentration (c) Free energy versus composition curves for solid and liquid phases at four temperatures T4 > T3 > T2 > T1 . the crystalline solid curve has the lowest free energy.3 has the lowest free energy for any composition. 2. consistent with its lower energy. The intermediate temperatures T2 and T3 of Fig. E. At the temperature T1 . FL (c). At the lowest temperature. the alloy is a solid at all compositions. however. and the lever rule T4 S L Free energy T3 S L T2 L S L T1 0. The free energy is minimized by selecting an optimal fraction of liquid and solid phases with different chemical compositions. the free energy curve for the liquid rises relative to the solid curve (and both may flatten somewhat).3. With decreasing temperature. changes quickly with temperature.0 t Fig. forming a liquid enriched in B-atoms and a solid depleted in B-atoms as explained below.4 0. 2.

4 = −0. where the heavier mass (larger f ) placed on the shorter arm of the lever (smaller c) balances a smaller mass on the longer lever arm. or with T–c phase diagrams themselves.4a.6 0.4. The condition is (cα − c0 )fα + (cβ − c0 )fβ = 0 .0 t Fig.7 − 0. Because fα + fβ = 1. indicated by the gray zones in Fig.Essentials of T–c phase diagrams 28 1.3. must average to zero.and β-phases.2 0. The composition deviations from c0 . Equation 2.3) The fα and fβ are the mole fractions of the α.2) (2. cβ − c0 c0 − cα = . denoted α and β) having different compositions.3) × 0.8 0. 2.4 0. with weights and lever arms as labeled. fβ fα (2.” The lever rule is a consequence only of solute conservation.2 − 0.6 0.and β-phases 16:48:42 . The area of the upper gray band is (0. and the other less.4 Common tangent construction One more tool is needed to relate curves of free energy versus composition to phase diagrams – the “common tangent construction.4b shows the analogy to balancing a lever. cβ − c0 .2 cα 0. whereas the areas of the two lower gray bands are each (0.8. fβ = cβ − cα fα = (2.2 = 0.” It is based on the reasonable assumption that the total free energy.3 is therefore known as the “lever rule. (b) Analogy to a balanced lever.8 fβ c0 cα c0 cβ c0 cβ 1. 2. 2. conservation of solute requires that one of the phases has more solute than c0 . FT . is the sum of contributions from the fractions of the α.0 0.0 Position (x) b (a) Composition profile of a two-phase alloy with overall composition c0 = 0.4 fα 0.4 a 0.3) × 0.5) Figure 2. and is useful when working with either curves of free energy versus composition. cβ − cα c0 − cα .4) (2.0 α α β Composition (c) 0.

10 ensures equal differences in the chemical potentials of the two atom species in the two phases μAα (T) − μBα (T) = μAβ (T) − μBβ (T) . we minimize the total free energy FT by moving the endpoints of the sloping lines in Fig. these two slopes equal the slope of the line between these two special compositions: Fβ (cβ ) − Fα (cα ) ∂Fβ (c) ∂Fα (c) .9. 2. This linear relationship of Eq. cβ . 2. (2.10 is required to ensure the minimum (so the chemical potential of an atom is the same in either phase).5. (2. fβ ) = (1 − fβ )Fα (cα ) + fβ Fβ (cβ ) . 2.3 (for temperature T2 ) is redrawn in Fig. 2. 2. the slopes at Fα (cα ) and Fβ (cβ ) are equal. Between cα and cβ . cα .8) Equation 2. cβ ) = fα Fα (cα ) + fβ Fβ (cβ ) .11) but not necessarily the minimum of FT – equal slopes can be found at two points on the right or left sides of the minima of the two F(c) curves. 16:48:42 . demanding that they stay on the curves of the two allowable phases. FT (cα . and second. A chemical composition of c0 = 0. cβ . (2. For thermodynamic equilibrium. One of the panels of Fig. 2. fβ .29 2. we used the lever rule to transform the free energy from a dependence on phase fraction to a dependence on composition.6) where the fractions are normalized to 1 as fα + fβ = 1. It can be shown that the first equality in Eq. 2. fβ ) = Fα (cα ) + fβ Fβ (cβ ) − Fα (cα ) . Eq.9. From the linear relationship of Eq. The common tangent construction gives the minimum FT .7) (2.7 is picked for illustration. 2. cβ . Using fα = 1 − fβ .9) In essence. the FT is on a straight line between its values at the endpoints. “×.5 is useful to transform from phase fraction to chemical composition  FT (cα . The common tangent selects special compositions cα and cβ that satisfy two equalities: first.4 Common tangent construction FT (fα . The result. We see that the lowest possible position for the “×” is found on the “common tangent” between the two curves of free energy versus composition.  =  = ∂c cα ∂c cβ cβ − cα (2. 2.5. c0 ) = Fα (cα ) + c0 − cα cβ − cα  Fβ (cβ ) − Fα (cα ) .5. (2.” in Fig.5. Fα (cα ) and Fβ (cβ ). The second equality in Eq. FT (cα . we know that FT will be at the three intersections marked by the crosses. is the equation of a straight line of free energy versus the composition c0 in the range cα < c0 < cβ . This figure shows three sloping straight lines between pairs of endpoints on the curves FL and FS .9 can be used to optimize the chemical compositions of a two-phase mixture. 2.10) The equilibrium compositions of the solid and liquid phases obtained from the common tangent construction are labeled “cL ” and “cS ” in Fig.

2. the total free energy is obtained on the straight line with endpoints on the two free energy curves.6 Concentration (c) 0.0 t Fig.4 0.5 Continuous solid solubility phase diagram We now construct the phase diagram for the alloy with the free energy versus composition curves of Fig.3.6 Solid. For a selected c0 .5 that the compositions labeled “cL ” and “cS ” from the common tangent construction serve to minimize the free energy of a mixture of solid and liquid phases for any intermediate composition c0 between 0.6 Concentration (c) 0. 2. 16:48:42 . First note from Fig.0 Free energy versus composition curves for solid (S) and liquid (L) phases at the temperatures T2 .Essentials of T–c phase diagrams 30 2.2 0.5 0. L Temperature T3 L+S T2 T1 0.8 1. The minimum free energy (lowest “×”) is found on the tangent line common to the two curves.4 0.0 t Fig.0 Continuous solid solubility phase diagram derived from curves of Fig. Dots mark compositions obtained by common tangent constructions on the panels of Fig. 2. T4 Liquid.3.3. 2.8 1.2 cS cL 0. 2.4 T2 c0 Free energy L S on mm nt o C nge ta 0. 2. S 0.

Fig. Results are presented in Fig. The focus of Sect. 2.2. The dots on the T–c phase diagram of Fig. 2. and by using mole fractions (or atomic fractions) the concentration c varies from 0 to 1. 2. 2. perhaps halfway between them. the equilibrium phases are obtained as follows • Draw a horizontal line across the diagram at the temperature of interest.5.1) gives an example of using this analysis for predicting the fractions of phases and their compositions during cooling or heating. is on why diffusion (Chapter 3) and nucleation (Chapter 4) cause deviations from the predictions of the equilibrium phase diagram in cases where the cooling is rapid.2. When c = 0 the alloy is pure element A.3. it is a straightforward exercise to construct common tangents and identify the compositions of the phase boundaries. read off the compositions cS and cL . 2. • To obtain the fractions of the solid and liquid phases. the alloy tends to form a little liquid that is enriched in the element B. when c = 1 the alloy is pure element B. however. the alloy is mostly liquid with a small amount of solid phase enriched in the element A.6 Solid solutions 2. These are the equilibrium compositions of the solid and liquid phase at the temperature of interest.4 and 2. Between the temperatures T3 and T4 . and this is also true for the element B.31 2. With F(c) curves at different temperatures as in the four panels of Fig.3 and 2. 2.5. The pure element A has a single melting temperature.84.6 were obtained directly from the common tangents of the F(c) curves of Figs. 2. From the intersections with the solidus and liquidus curves.6. 5. however. Some discretion was exercised in interpolating between the points. so the two curves must come together at c = 0 and c = 1.” marked “L+S” in the phase diagram of Fig. 2.6 Solid solutions and 0. The phase diagram shows that the A–B alloy does not have a single melting temperature.6. Just below the “liquidus” (the upper curve in this phase diagram). which is a T–c phase diagram.6 shows that the last of the A-rich solid phase has melted. Section 5. • The crystal has N sites.1 Parameters for atom configurations and bonds Consider a binary alloy with atom species A and B placed at random on the sites of a crystal. For any point within the “two-phase region. • The concentration of B-atoms is c. and therefore a number cN of B-atoms and (1 − c)N of A-atoms. of course. • The concentration of A-atoms is 1 − c. 5.2 (see Fig. select a composition of interest c0 . and use the lever rule with the compositions cS < c0 < cL in Eqs. • Each site is surrounded by z sites as first-nearest neighbors (1nn). In equilibrium at temperatures above the “solidus” (the lower curve in this phase diagram).6. 16:48:42 .

Z1site Z1site = e−βeA + e−βeB .7b. z/2.14) c (a) Square lattice with coordination number z = 4. (c) Double-counting of bonds with z bonds per site. simply count the number of pairs with each energy.1).7 a b (2. • All the energy of atom configurations depends on 1nn (first-nearest-neighbor) pairs of atoms11 (see point 2 of Sect. to avoid overcounting the bonds. Its energy depends on the number of its A-atom neighbors. • The alloy is assumed to be completely random. 2 2 eA = (2. and four neighbors about each atom. A pictorial justification for using z/2 bonds per atom is shown in Fig. 2. and the number of its B-atom neighbors. however. Sometimes.12) (2. – Each A–B pair has energy eAB . zc. (b) Tiling of a square lattice with one atom per site. and its other (1 − c)z neighbors are A-atoms. 16:48:42 . and two bonds per site. 2. and likewise for the average B-atom. on average each A-atom is surrounded by zc B-atoms. – To obtain the total energy of an alloy configuration. – Each A–A pair has energy eAA .Essentials of T–c phase diagrams 32 • For a random solid solution.13) The division by 2 corrects for a double-counting of bonds when performing the average by summing over all atoms in the crystal – the bond from each neighbor atom back to the original A-atom gets counted a second time. 11 One way to justify this assumption is to argue that most of the chemical bonding energy is associated with 1nn pairs. – Each B–B pair has energy eBB . 2. t Fig. The energy for this average A-atom. so the species of atom on any lattice site does not depend on the species of atoms occupying its neighboring sites (see point 1 of Sect. which is often approximately correct. a 1nn model can average successfully over other interactions.2 Partition function and free energy We begin by writing the partition function for a single site. 2 2 z z eB = (1 − c) eAB + c eBB . z(1−c).1). is z z (1 − c) eAA + c eAB . Consider an A-atom on a lattice site. 2. 2.6.

(2.17) e A (N − n)! n! Obtaining the Helmholtz free energy from the expression F = −kB T ln Z. and using the Stirling approximation ln(x! ) x ln x − x F(c) = [(N − n)eA + neB ] −kB T [N ln N − n ln n − (N − n) ln(N − n)] . e A (N − n)! n! (2. For a random solid solution.6 Solid solutions 33 where β ≡ 1/(kB T).16) For a fixed concentration c (one value of n). (2. all site occupancies are independent.15) We can evaluate ZN with the binomial expansion. ZN = N   n=0  −βe N−n −βe n N! e B . so the total partition function of the alloy with N sites is N ZN = Z1site . . we need only one term from the sum12 over n   −βe N−n −βe n N! ZN (c) e B .2.

F(c) = eAA + 2c(eAB − eAA ) + c2 (4V) (zN/2)

−kB T ln N − c ln n − (1 − c) ln(N − n) N ,

(2.18)

(2.19)

which can be written as
F(c) = Econf (c) − TSconf (c) ,

(2.20)

Econf (c) = eAA + 2c(eAB − eAA ) + c2 (4V) (zN/2) .

(2.21)

by using the definitions

Econf (c) = [(1 − c) eAA + c eBB − c(1 − c) 4V] (zN/2) .

Sconf (c) = kB − c ln c − (1 − c) ln(1 − c) N .

(2.22)
(2.23)

Here Econf (c) and Sconf (c) are the configurational energy and configurational entropy of the
alloy.13
In Eq. 2.19 we defined implicitly an important combination of pair potentials
V ≡ (eAA + eBB − 2eAB )/4 .

(2.24)

This parameter V is sometimes called the “interchange energy.” Its physical meaning is
made clear with the change in configuration of an A- and B-atom pair in Fig. 2.8. For
the initial configuration of the central A- and B-atoms in Fig. 2.8a, the figure shows a
total of five A–B pairs, one A–A pair, and one B–B pair. After interchange of only the
12 Also, when N is very large, if the equilibrium concentration is c, Z becomes a sharply peaked function with
N

a maximum at n = cN. It is then standard practice in statistical mechanics to replace ZN with its maximum
value (see [7]).
13 The word “configurational” refers to arrangements of atoms on the crystal sites.

16:48:42

34

Essentials of T–c phase diagrams

B

t

B

A

B

A

A

A

A

B

A

B

B

B

B

a

A

b

A

Fig. 2.8

(a) Initial configuration around central A- and B-atoms. (b) New configuration after exchange of the central A–B pair.

Box 2.3

Pairwise energy in the point approximation
Here are some details of calculating E(c) by counting atom energies as the number of each species times
its energy
E(n) = (N − n)eA + neB

(2.25)

With Eqs. 2.12 and 2.13, and noting that n = Nc and N − n = N(1 − c)  

zN  

zN
E(c) =
(1 − c) (1 − c)eAA + c eAB +
c (1 − c)eAB + c eBB , (2.26)
2
2 

zN 
(2.27)
(1 − c)2 eAA + c2 eBB + 2c(1 − c) eAB .
E(c) =
2
Equation 2.27 shows that all the energy from A–B pairs scales as 2c(1 − c), which is the probability that a
1nn pair on the lattice is an A–B or B–A.
We seek terms linear in c that come from replacing A-atoms with B-atoms, helping to isolate the linear
interpolation of the energy from pure A to pure B. Deviations from this straight line will show the chemical
preferences of the unlike A–B pairs, which are most plentiful at intermediate compositions.
zN 
eAA − c eAA − c eAA + c2 eAA + (c eBB − c eBB ) + c2 eBB
E(c) =
2 

+2c(1 − c)eAB ,
(2.28)  

zN
eAA + c(eBB − eAA ) − c(1 − c)(eAA + eBB ) + 2c(1 − c)eAB , (2.29)
E(c) =
2 

zN 
E(c) =
(2.30)
eAA + c(eBB − eAA ) − c(1 − c)(eAA + eBB − 2eAB ) ,
2 

zN 
eAA + c(eBB − eAA ) − c(1 − c) 4V ,
(2.31)
E(c) =
2
which is the same as Eq. 2.22.

central A- and B-atoms,14 there are three A–B pairs, two A–A pairs, and two B–B pairs
(Fig. 2.8b). After the interchange in the figure, the change in pair energy is precisely 4V =
eAA + eBB − 2eAB . For other interchanges, or for multiple interchanges, the change in pair
energy will always be an integer multiple of ±4V.
14 Note that this interchange conserves the total number of atoms of each type. Also note that an interchange of

like pairs, A–A or B–B, has no effect on the total number of like or unlike pairs.

16:48:42

35

2.6 Solid solutions

Box 2.4

Configurational entropy in the point approximation
The partition function of Eq. 2.17 has a combinatorial prefactor that becomes the configurational entropy
when the free energy is calculated as F = −kB T ln Z  

N!
.
(2.32)
Sconf = kB ln
(N − n)! n!
Using the Stirling approximation, ln x! = x ln x − x
Sconf = kB [N ln N − N − (N − n) ln(N − n) + (N − n) − n ln n + n] . (2.33)
There is a cancellation of the terms −N + (N − n) + n, and we use the expressions n = Nc and
N − n = N(1 − c)    


Sconf = kB N ln N − (1 − c) ln N(1 − c) − c ln Nc .
(2.34)
There is a cancellation of the terms ln N − (1 − c) ln N − c ln N giving
Sconf = −kB N [(1 − c) ln(1 − c) + c ln c] ,

(2.35)

which is the same as Eq. 2.23. Because c and 1 − c are less than 1, the two logarithm functions are negative,
and Sconf is positive. It has a maximum value of 0.69 kB N at c = 0.5, and is zero at c = 0 and c = 1.
Logarithm functions have singularities as c → 0 or c → 1. These are slow divergences, and the factors
c and 1 − c have stronger dependences on c, so the functions c ln c and (1 − c) ln(1 − c) are both well
behaved in Eq. 2.35 at the concentration limits. (Sconf (c) does have singularities in its derivatives as c → 0
or c → 1.) Nevertheless, as c → 0 the function −c ln c becomes larger than c ε, where ε is a constant,
such as the energy of an isolated defect or impurity atom. The configurational entropy therefore dominates as
c → 0 and c → 1, so mixing a small amount of impurity is always favorable in the dilute limits at finite
T, even if chemical bonding does not favor it.

2.6.3 Functional forms of E(c) and S(c)
The first two terms in Eq. 2.22 for Econf (c) have coefficients 1 − c and c, and are therefore linear in composition. These terms prove uninteresting for unmixing, but if 4V = 0,
we would have only these two terms for Econf (c). When 4V = 0, the pairwise energy is
insensitive to atom interchanges as in Fig. 2.8, and there can be no change in Econf (c) if we
make half the material pure A and half pure B, compared with making all the material an
equiatomic solid solution of A and B. With no energetic tendency for unmixing, but with
an entropy that favors mixing, the alloy will remain a solid solution at all temperatures.
Such an alloy with no energetic preference for unmixing (or ordering) is called an “ideal
solution.”
Unmixing is controlled by the term c(1 − c) 4V in Eq. 2.22, which has the same
sign as V. Consider first the case where V < 0, meaning that the average of A–A and
B–B pairs is more favorable than A–B pairs. This gives the alloy a tendency to unmix

16:48:42

Essentials of T–c phase diagrams

36

1.2

2.0

zV/(kT ) = –2
Free energy [kT/atom]

1.0

1.5

1.0
S(c)

F = E – TS

Entropy [k/atom], Energy [kT/atom]

E(c)

0.5

0.8
0.6
0.4
zV/(kT ) = –1
0.2
0.0
–0.2

zV/(kT ) = –0.667

–0.4

t

Fig. 2.9

0.0
0.0
a Pure A

0.2

0.4
0.6
0.8
Concentration (c)

1.0
Pure B

b

0.0
Pure A

0.2

0.4
0.6
0.8
Concentration (c)

1.0
Pure B

(a) Configurational energy and entropy for V < 0 (Eqs. 2.21 and 2.23). (b) Free energy for different values of zV/(kT)
as labeled.
chemically.15 The energy and entropy terms of Eqs. 2.22 and 2.23 are graphed in Fig. 2.9a.
When Econf (c) is added to −TSconf (c) to obtain the free energy curves in Fig. 2.9b, the
logarithmic singularities in composition derivatives of the entropy always force a reduction in F from its values at c = 0 and c = 1. The energy term, positive in this case
with V < 0, can dominate at intermediate concentrations, however, especially at low
temperatures.
For small magnitudes of zV/kB T (high temperatures), a random solid solution offers the
minimum in free energy for any composition. For the small zV/kB T = −0.667 in Fig. 2.9b,
Fconf (c) has a simple shape. On the other hand, the free energy Fconf (c) has an interesting
curvature in c when the ratio is more negative than −1, or equivalently for temperatures
below
Tc = −zV/kB .

(2.36)

This condition for Tc is obtained by setting the second derivative of Fconf (c) equal to zero,
as in Problem 2.5. This Tc is the “critical temperature” for unmixing at a concentration of
c = 0.5.16 Further interpretation of these curves, the subject of the next section, leads to
the unmixing phase diagram of Fig. 2.10.
15 The opposite sign of V > 0 means that at low temperatures the alloy will tend to maximize the number of its

A–B pairs. It does so by developing chemical order, defined by order parameters presented in Sect. 2.10.
16 This unmixing transformation differs from unmixing in the eutectic transformation of Sect. 2.8 in that it does

not require nucleation of a distinct phase.

16:48:42

2.7 Unmixing phase diagrams

37

1.4

Temperature [zV/k]

1.2
α

1.0
0.8

α'

α"
α' + α"

0.6
0.4
0.2
0.0
0.0

t

Fig. 2.10

Box 2.5

0.2

0.4
0.6
0.8
Concentration (c)

1.0

Unmixing phase diagram derived from the curves of Fig. 2.9. Arrow marks composition limits obtained by common
tangent construction from top curve of Fig. 2.9b.

Deeper issue
Our Econf (c) and Sconf (c) were derived for a random arrangement of atoms. If this is true, and it is at
sufficiently high temperature, the probabilities for an A- or B-atom on a site ((1 − c) and c) are accurate,
as are the probabilities for the pairs of the different species (zc2 /2 for a B–B pair, for example). At lower
temperatures the equilibrium site occupancies are not random, however, owing to the energetics of chemical preferences. Both Econf (c) and Sconf (c) need to be modified when there is a tendency for A-atoms to
prefer A-atom neighbors at lower temperatures. We ignore this in the next sections, but here is an important
consequence.
At temperatures somewhat above the critical temperature for unmixing, the alloy is not random. Over short
distances there is a tendency for more A–A and B–B pairs because thermal energy does not entirely disrupt the
chemical preferences. One consequence is that the configurational entropy is not so large as for a random solid
solution. Likewise, the hump in the curve of E(c) in Fig. 2.9a is also suppressed. The overall result turns out
to be a significant reduction of the hump in the curve of F(c) in Fig. 2.9b. Unfortunately our point approximation, which has no independent variables to give information on pairs or larger clusters, cannot account for
these effects of short-range order. Even the pair approximation becomes inadequate as we cool closer to Tc .
Closer to Tc , the spatial range of the order increases exponentially, and diverges to infinity at Tc itself. This is
the topic of Chapter 20.

2.7 Unmixing phase diagrams
Using the common tangent construction to span between the two local minima in the F(c)
curves at low T in Fig. 2.9b, the unmixing phase diagram of Fig. 2.10 is constructed. It

16:48:42

Essentials of T–c phase diagrams

38

1.4

Free energy

d2F >0
dc 2

Temperature [zV/k]

1.2

d2F < 0
dc 2

α

1.0
0.8 α'

α"

0.6
0.4
0.2

0.0

t

Fig. 2.11

a

0.2

0.4

0.6

0.8

Concentration (c)

0.0
0.0

1.0
b

0.2 0.4 0.6 0.8
Concentration (c)

1.0

(a) Typical free energy versus composition curve for an unmixing alloy at low temperature. The curvature of the
function is negative for 0.26 < c < 0.74, but positive for all other compositions. The inflection points of
d2 F/dc2 = 0 are marked with vertical dashed lines. (b) Unmixing phase diagram with dashed “spinodal” from
inflection points of part a.
shows a solid solution at high temperatures, denoted the “α-phase.” At low temperatures
this solution becomes unstable against chemical unmixing, and forms an A-rich “α  -phase”
and a B-rich “α  -phase.” In the two-phase region beneath the arching curve in Fig. 2.10,
the chemical compositions of the two phases can be read directly from the curve itself
for any temperature.17 The horizontal double arrow points to these two compositions for
a normalized temperature of 0.5 (which is consistent with the common tangent applied
to the top free energy curve in Fig. 2.9b). In this unmixing phase diagram of Fig. 2.10,
the composition segregation is larger at lower temperatures, and is infinitesimally small at
c = 0.5 at the critical temperature Tc = z|V|/kB .
Figure 2.11a shows a typical F(c) curve for T < Tc , emphasizing its curvature. The
curvature is both positive and negative, with the null condition d2 F/dc2 = 0 at the two compositions indicated by the vertical dashed lines, approximately at c = 0.26 and c = 0.74.
A negative curvature of the free energy makes the alloy highly unstable against unmixing.
Within the composition range of negative curvature (where F(c) is shown as gray), any line
that intersects the two sides of the free energy curve will lie below F(c) for any composition between the points of intersection. A typical line is shown below the composition
c = 0.44, with an arrow showing how an alloy of this composition can reduce its total free
energy by separating into the two compositions at the ends of the line. In this region of
negative curvature, d2 F/dc2 < 0, the free energy is unstable against even infinitesimally
small changes in composition, since we could have drawn the straight line only slightly
below the gray curve, and still have unmixing.
When infinitesimally small composition changes are favorable, nucleation of a distinct
zone is not needed to start the unmixing process, and unmixing can occur with a continuous
change in composition as in Fig. 1.5d. Sometimes an additional boundary line is drawn on
the unmixing phase diagram to indicate this region of negative curvature, like the dashed
line in Fig. 2.11b. For compositions and temperatures within this dashed curve, called the
17 The fractional amounts of the phases are set by the lever rule or Eqs. 2.4 and 2.5.

16:48:42

39

2.8 Eutectic and peritectic phase diagrams

“spinodal,” the alloy can unmix by the continuous process of “spinodal decomposition,”
instead of by nucleation and growth. On the other hand, nucleation and growth is required
between the spinodal and the solid boundary lines of Fig. 2.11b.
Beware, though, a free energy curve with d2 F/dc2 < 0 is so unstable that it is not useful
for equilibrium thermodynamics, including the development of phase diagrams. Spinodal
decomposition does occur in nature, and it is a continuous process. Spinodal decomposition
is explained in Chapter 12 as a kinetic phenomenon based on diffusion, however, rather
than a thermodynamic concept based on an alloy free energy function. Obtaining a spinodal
curve with a F(c) as in Fig. 2.11a is not rigorous.

2.8 Eutectic and peritectic phase diagrams
It is possible to have equilibrium between three phases in a binary A–B alloy at a fixed
temperature and pressure. The canonical examples involve a liquid phase in equilibrium
with two solid phases. For example, Pb dissolves in Sb in the liquid phase, but at low
temperatures the Pb-rich phase is fcc, whereas the Sb-rich phase is hexagonal. A separate
free energy versus composition curve is needed for each crystalline phase.
There are two possible forms of phase diagrams for one high-temperature phase and two
low-temperature phases – “eutectic” and “peritectic” phase diagrams. The form depends
on the compositional sequence of the minima of the three Fξ (c) curves.

2.8.1 Eutectic phase diagram
The eutectic case occurs when the minimum of the FL (c) curve of the liquid phase lies
between the minima of the two solid phase curves as shown in Fig. 2.12a. This figure
shows four liquid free energy curves for four temperatures. For simplicity, the solid phase
curves are assumed constant with temperature, since they have the lower entropy. With
increasing temperature, the liquid free energy curve falls relative to the solid curves owing
to the large −TSL term in the liquid free energy. Three common tangent constructions are
shown in the figure. The most interesting one is at the temperature T2 , where the common
tangent touches all three phases, with the liquid in the middle. This temperature T2 is the
“eutectic temperature.” These three points of contact are shown in Fig. 2.12b, which is a
complete eutectic phase diagram. Please check the correspondence between the compositions from the common tangent constructions and the points on the phase diagram for the
same temperature by comparing Figs. 2.12a and 2.12b.

2.8.2 Peritectic phase diagram
The other possible compositional sequence for three free energy curves occurs when the
liquid curve has its minimum on one side of the minima of the two curves for the solids.
This case is depicted schematically in Fig. 2.13a, with the corresponding “peritectic” phase

16:48:42

Essentials of T–c phase diagrams

40

FL(T1)
T4
L

Free energy

FL(T2)
Temperature

T3
FL(T3)

FL(T4)

0.0

t

Fig. 2.12

T2

0.8

1.0

L+β

α

β
α+β

T1

0.4
0.6
Concentration (c)

0.2

a

α+L

0.0

0.2

b

0.4
0.6
Concentration (c)

0.8

1.0

(a) Free energy curves for two solid phases, α and β, and a liquid phase drawn for four temperatures
T1 < T2 < T3 < T4 . Three common tangent constructions are shown, from which seven compositions are found.
(b) Eutectic phase diagram derived from the curves of part (a) – the nine dots mark the seven compositions from
common tangent constructions.
T4

L

Free energy

Temperature

T3
α+L
T2

L+β
T1

α+β
β

FL(T2)
0.0

t

Fig. 2.13

a

0.2

0.4
0.6
Concentration (c)

0.8

1.0

0.0

0.2

b

0.4
0.6
Concentration (c)

0.8

1.0

(a) Free energy curves for two solid phases, α and β, and a liquid phase drawn for the peritectic temperature T2 . (b)
Peritectic phase diagram estimated from curves of part (a).
diagram in Fig. 2.13b. This phase diagram can be constructed by the same common tangent
exercise as was used for the eutectic case in Fig. 2.12 – lowering the FL (c) curve relative to
Fα (c) and Fβ (c) with increasing temperature. Peritectic phase diagrams are common when
one crystalline phase, here the β-phase, has a much lower melting temperature than the
α-phase. Note also that if the liquid free energy curve were to lie to the left of the two solid
curves, the common tangent construction would produce a peritectic phase diagram with
the higher melting temperatures to the right.

16:48:42

Nevertheless..3 GPa. In various combinations. Although this region occupies a two-dimensional area in the diagram. 1. (2.6 Gibbs phase rule This rule is [8]: P+F =C+2. however. 2. e. gives us one more degree of freedom. the forms of Figs.3 More complex phase diagrams Eutectic. this is a tricky feature of applying the phase rule. we expect that the eutectic point would shift in T and c but would not disappear in a set of eutectic T–c phase diagrams at different pressures. such as the α. which is not shown in a eutectic T–c phase diagram. but of course only one phase in equilibrium. The T–p phase diagram of cerium (Fig. causing a loss of a degree of freedom for the intensive variables.g.41 2. so C = 2). A line in a T–c phase diagram. (2.12b. 2. and a magnetic field is added to give the set {T.12b.38) but this is unconventional and may be declared incorrect by a casual observer. Equation 2. 2 for a binary A–B alloy). it is necessary to write P+F =C+3.6. The degrees of freedom must be handled with care for T–c phase diagrams.13b.37 becomes 3 + 1 = 2 + 2 at the eutectic point. peritectic. p. 2. plus the ordering diagrams discussed below. The same situation is true for a two-phase region of the diagram.39) The essential idea behind the Gibbs phase rule is that adding a new phase puts another constraint on the chemical potentials through expressions like Eq. and C is the number of components (chemical species. such as a liquidus line. Beware. p. and Eq. the amounts of the two phases are constrained by the lever rule. 2. If pressure is kept constant.37) where P is the number of phases.11. it is appropriate to rewrite the phase rule as P+F =C+1. For the binary alloy we add a component (species B. With one component (Ce). giving the set {T. β. however. c}. (Yes. 2.3) is simple to interpret.. (2. 2. as in a T–c phase diagram. Furthermore. “α”) we have a real degree of freedom to change c. such as the α + β region in Fig. since Eq. Here is an interpretation for the eutectic point in a eutectic T–c phase diagram. The one degree of freedom refers to pressure. 2. c}.g. the chemical composition can be used as an intensive variable.8. F is the number of degrees of freedom. so one degree of freedom is lost.8 Eutectic and peritectic phase diagrams Box 2.37 allows freedom to vary either temperature or pressure. if the magnetic properties of the phases are significant and different. B. and only two phases are in equilibrium at the liquidus. and γ at T = 230 K and p = 0. it is possible to have three phases in equilibrium at one point (zero degrees of freedom).) Within a single-phase region (e. 2. account for almost all features found in real binary 16:48:42 . for example). and continuous solid solubility phase diagrams can be assembled into more complex diagrams with multiple phases and two-phase regions (see [9].37 becomes 3 + 0 = 1 + 2.

The triangular grid at the bottom depicts the alloy composition.6b. 2. Temperature is vertical. 5. 2. The Gibbs phase rule states that adding a third component (the element C) gives an additional degree of freedom in the intensive variables of temperature. not labeled).14 shows an image of a compound eutectic diagram for a ternary alloy (a special case where each pair of elements. with region of γ phase. and C (to back. showing more clearly the solidus surfaces and the low-temperature ternary eutectic point.5 as in Fig. increases with distance from the front to the back of the model. C. and A–C has a eutectic phase diagram as labeled in Fig. B. (b) Right: sections removed to show that the ternary eutectic composition has a lower melting temperature than any of the three binaries. The two-phase areas of the binary T–c eutectic diagram become volumes in the ternary diagram. T T α+L α t Fig. A–B. and there are no other phases). the diagram is called a “eutectoid” or “peritectoid” diagram.14a.0 could be attached together at the composition c = 0. probably an ordered compound. Alternatively. respectively. a peritectic phase diagram could be attached to one side of a eutectic diagram. 5.9 Ternary phase diagrams A binary alloy has one independent composition variable.14 A β+L α+β Composition β α B A β Composition B (a) Left: a ternary phase diagram of elements A.) 2. perhaps the disordered solid solution in the phase diagram of Fig.14b. pressure or mole fractions.11b where a third solid phase (γ in Fig. two eutectic phase diagrams over composition ranges from 0 < c < 0.12b or 2. Two of these volumes are removed in Fig. the liquid phase in a eutectic diagram could be replaced by a random solid solution. and the concentration of the third element. The eutectic point in this ternary diagram has three solid phases and a liquid phase in equilibrium (compared with two solid phases and one liquid phase for a binary alloy). (When a solid solution replaces the liquid region in Figs.5 < c < 1. A two-dimensional T–c diagram is no longer sufficient.5 and 0. In another variation.Essentials of T–c phase diagrams 42 alloy phase diagrams. Figure 2.13b. 2. B–C. For example. 16:48:42 . 2. is stable.11b). but a ternary alloy has two. 2.

” Isopleths have the same usage as binary phase diagrams for locating singlephase regions. For a two-phase region in an isopleth. the phases have compositions given by the corners of the triangle. If Fig.14). Horizontal cuts as in Fig.15 0.14 and 2.15 are “isothermal sections.2 t 0. cC } are at the ends of the three intersecting grid lines. and generally lie out of the plane of the isopleth (unless the tie line is along constant cC . 2.2 0. 2.8 0. it is necessary to know the “tie-lines” that identify the compositions of the two phases in equilibrium.14 (which show all temperatures. The central triangle is a region of three-phase equilibrium.6 α +β +γ β α +β 0.15. but cannot include all compositions). and γ are rich in the elements A.4 0. and a substantial effort continues to obtain this important information by experiment and theory [10]. however. a tangent plane will touch near the minima of three bowl-shaped Fξ (c) curves. Handbooks of ternary phase diagrams usually do not show three-dimensional data structures as in Fig.15 corresponds to a temperature just below the existence of the liquid.18 and their fractions are set by a pair of lever rules for the distances from the alloy composition to the compositions of these three corners.14 at a low temperature. but instead present various cuts through them. Compared with a binary alloy. B. The isothermal section of Fig. Tie lines affect the utility of vertical sections of Fig. The three phases α.15 are actually a rather simple example.9 Ternary phase diagrams 43 C 0. 16:48:42 .19 18 At the low temperature T .6 0. The two-phase regions are shown with thin gray tie-lines indicating the specific compositions in equilibrium with each other.15. cB . 2.15 is for the alloy of Fig. Three-phase triangular regions are often handy structural units for understanding isothermal sections as in Fig.2. the compositions {cA .6 0. 2.8 B Schematic isothermal section of a ternary phase diagram (for the low-temperature part of Fig. or tie-lines.15. Ternary alloys allow for phases and microstructures beyond those possible for binary alloys.8 γ α+ 0.” For any point inside Fig. Vertical sections made with constant cC are called “isopleths. temperatures. the contact points are at the corners of the triangular three-phase region in Fig. C. 2. but today our knowledge of ternary phase diagrams is much less complete. as for a binary diagram with cC = 0). Within the three-phase region.4 0. the two phases generally do not have the same composition cC . For our ternary eutectic system at low temperature. 2. β. Figures 2. Around each triangular three-phase region there must be two-phase regions and single-phase regions at their ends. 2.2 Fig.4 γ α A β+ γ 0. adding a third component gives another degree of freedom by Eq. A vertical stack of isothermal sections for different temperatures can depict the full ternary diagram.14.39.12a shows a common tangent between two minima of the free energy 1 curves of phases α and β. 2. imposing a new feature in the two-phase regions of ternary phase diagrams. 2. 2. 2. respectively. 2. Fig. In a two-phase region of a ternary alloy. 2. 19 Today many ternary diagrams are missing phases.

1. 20 Near L = 1 the derivative of the entropy with respect to L has a logarithmic singularity in sublattice concentrations analogous to the concentration dependence of Eq. An alloy with an ordering tendency. alloys with either unmixing or ordering tendencies will be random solid solutions if configurational entropy is dominant. each of these sublattices contains N/2 sites.16b).16a identifies sublattices for the two species of atoms. specify an equiatomic A–B alloy on a square lattice. The sublattice concentrations are used to define a “long-range order” (LRO) parameter. At high temperatures. 2.16a at T = 0. 2.41) A on α B on β t Fig. N/2 (2.1 Long-range order parameterization We return to the issue of chemical preferences of atoms in an A–B binary alloy as expressed in Eq. 2. L.4). but their phases at low temperatures are fundamentally different (see Fig. for the alloy [11] L≡ R−W . For illustration of an ordering transformation. maximizes the number of its A–B pairs at low temperatures. 4 (2.23 near c = 1. (b) Misplaced atoms A and B (antisites) on generally ordered sublattices.Essentials of T–c phase diagrams 44 2. as shown in Fig.10. we expect the structure of Fig. The double arrow indicates how a duo of A and B atoms can be exchanged to eliminate antisites. For V > 0.16 a A on α B on β A on β B on α b (a) Ordered square lattice with two species {A.20 driving some mixing of A-atoms onto the β-sublattice (Fig. V > 0. and the number of “right” atoms is R. so A–A and B–B pairs are preferable to A–B pairs (eAA + eBB is more negative than 2eAB ). Figure 2. By conservation of atoms and conservation of sublattice sites.40) Recall that an alloy with an unmixing tendency has V < 0. an equal number of B-atoms must move onto the α-sublattice.16. 2.10 Long-range order in the point approximation 2. The number of these “wrong” atoms (or “antisites”) on each sublattice is W. β}. B} and two interpenetrating square sublattices {α. 2. Finite temperatures favor some disorder in the structure – putting a few atoms on the wrong sublattice gives a big increase in the configurational entropy.24: V≡ eAA + eBB − 2eAB . For a crystal with N sites. 16:48:42 . 2.

Owing to the conservation of sublattice sites and the conservation of atoms. L(T). This is ensured by considering two atoms at a time – we transform between a “right duo” that comprises an A-atom and a B-atom on their proper sublattices..42) . i.10 Long-range order in the point approximation N/2 α β Rα Wβ N/ 2 α NA A B Wα α Rβ 0 0 Fig. and B–B pairs of atoms as in Sect. The energy of the two atoms in the wrong duo (a Wrong A plus a Wrong B) is zR zR zW eAA + eBB + 2 eAB . W = 3. A–B. Figure 2. 2. 2.45 2.79.41 gives the same value of L if R and W are averaged from both sublattices (Rα = Rβ in general. R = W. R = 25. 2.6. and L = 0.16a. N/2 N/2 N/2 starting wrong atom : wrong A wrong B wrong A (B) eW = neighbor : right A right B 16:48:42 wrong B (A) (2.16b.17a depicts the general case of an alloy with two identical sublattices. The case of L = 1 corresponds to the case of Fig. 2. Because the distances between the heavy and dashed horizontal lines are equal for both sublattices α and β. To calculate the equilibrium value of LRO parameter.41 will be the same for both sublattices. we must be sure to conserve the number of each species. and L evaluated from Eq.e. The energy is obtained by counting the number of A–A. we seek the ratio of right and wrong duos as a function of temperature.10. the transfer of B-atoms onto the α-sublattice must be exactly compensated by the transfer of A-atoms onto the β-sublattice. This corresponds to swapping the two atoms at the ends of the double arrow in Fig.17b shows how the sublattice populations change as the state of order is changed (reduced in this case). unless c = 0. but an overall chemical composition c.16. This L ranges from −1 < L < 1. For the alloy of Fig. The disordered equiatomic alloy has as many right as wrong atoms on each sublattice. so L = 0 for a disordered alloy. (b) Switching A-atoms to B on the α-sublattice requires a compensating change of B-atoms to A on the β-sublattice. where R = N/2 and W = 0.17 β β NA B A cN/2 t α a NB β NB b (a) Sublattice concentrations for an alloy of overall composition c. Figure 2. and a “wrong duo” that comprises an A-atom and a B-atom on the wrong sublattices (antisites). 2. 2.5). 2. As atoms are moved between sublattices. the difference R − W is equal for both. Equation 2.2 Temperature dependence of LRO We now calculate the temperature dependence of the long-range order parameter.

2.42 or Eq.45 becomes   1+L L2Vz = exp . the energy of a single atom will be half that of Eq. N/2 N/2 N/2 (2. 2. (2. 2. remember that the neighbors of an atom are on the other sublattice.43.16) puts the two atoms on their proper sublattices. 2. Equation 2. and the third term (two terms really) is from the B-neighbors of the wrong A that are themselves wrong (plus the wrong A-neighbors of the wrong B. 2.46 becomes: L2Vz . W 2 N/2 (2.18. it is found that R = (1 + L)N/4 and W = (1 − L)N/4. W e W  z 1 R = exp −β (R − W)(2eAB − eAA − eBB ) . 2.48) (2. 13]. The analogous expression for the pair energy of the “right” duo is: eR = zW zW zR eAA + eBB + 2 eAB .49) Please compare this result to the critical temperature for the unmixing alloy with c = 0. (Notice how the swap of the variables R and W in Eqs.45) With Eq. Tc = kB ln(1 + 2L) = (2.36.42.46 cannot be reduced further into a general analytical expression for L(T).44) (2.47) (2. Unfortunately. 2. so the first term is from the wrong A in the duo that has right A-atoms as neighbors.7b. 2. Exchanging the positions of the wrong A.42 makes a simplifying assumption about counting pairs – if we pick an A-atom. the second term is from the wrong B in the duo that has right B-atoms as neighbors.41 and noting that R + W = N/2. 2. We can solve Eq. its neighbors on the adjacent sublattice sites will be determined only by the overall concentration on the sublattice.42 and 2. so Eq.) On the average.and B-atoms in the duo (as indicated by the double arrow in Fig. kB Tc L2Vz . 2. Eq. for the reason illustrated in Fig. 2. 2. 2.46) 1−L kB T where again V ≡ (eAA + eBB − 2eAB )/4.Essentials of T–c phase diagrams 46 To understand Eq. 2L = kB Tc Vz . Eq. giving the factor of 2). but the B (A)-neighbors of the right A (B) are right. the B-neighbors of the right B are wrong. The theory presented in this section is called the “Bragg–Williams” or “mean field” theory of ordering [12. The critical temperature is found analytically by taking L as infinitesimally small. for which Eq.5. It can be solved numerically.43) where the A-neighbors of the right A are wrong.43 follows the swap of the atoms in the duo.46 in an important limiting case. Because the Bragg–Williams 16:48:42 . and the result for L(T) is shown in Fig. We calculate the ratio of right to wrong duos (equal to R/W) as the ratio of their Boltzmann factors: R e−βeR /2 = −βe /2 .

spinodal ordering is possible.2 Long-range order versus temperature in Bragg–Williams approximation. • Eutectic phase diagram.0 0. there is no spatial scale. The solid phase nucleates and grows from the liquid. so V > 0. One of the phases has a lower melting temperature than the other at the peritectic temperature.18 0. and composition fluctuations evolve continuously into chemical segregations. • Peritectic phase diagram. the liquid phase is stable at the lowest temperature. • Ordering phase diagram.0 LRO parameter (L) 0. • Unmixing phase diagram. In cases allowed by symmetry.8 0.6 0.4).2. The composition of the liquid phase lies outside the range of the two solid phases. so V < 0 and unmixing occurs at low temperatures and intermediate compositions.6 0. The sublattices are infinite in size.7 Summary of T–c phase diagrams The five major T–c phase diagrams are • Continuous solid solubility. The species A and B prefer like neighbors more than unlike. There is one crystal structure that accommodates all chemical compositions from pure B to pure A.10 Long-range order in the point approximation 47 1.0 t Fig.4 0. an ordered structure maximizes the number of unlike pairs.2 0.4 0. Box 2. 14. 2. At low temperatures. theory keeps track only of the average sublattice concentrations. 16:48:42 .8 Temperature (T/Tc) 1. but its chemical composition varies with temperature. where the solid solution is unstable and the order parameter evolves continuously (Sect.2 0. Spinodal decomposition is possible when the solid solution is unstable. At an intermediate composition.0 1. The species A and B prefer unlike A–B pairs more than like. It is therefore appropriate to call L a long-range order parameter. The liquid free energy curve has a minimum at a composition between those of the two solid phases.

3–7. Some advanced concepts needed for obtaining accurate free energy functions for these minimizations are: • Enthalpy. but any quantitative understanding requires additional physical concepts and complexity. At low temperatures the enthalpy originates with static structures of the nuclei and electrons (the essential components of matter). however. 7. Both the bonding energy and the effects of pressure can be obtained from the electronic structure of the alloy.) In nature. 2. and alloys tend to exhibit “shortrange order.23 is that at intermediate temperatures it is not correct to assume that the occupancy of a crystal site is independent of the occupancy of neighboring sites. For crystals of pure elements there is no configurational entropy. This is developed further in Sect. is independent of the surrounding atoms.” even at temperatures above the critical temperature.2. The models presented in Chapter 2 offer valuable conceptual understanding. 2. (Likewise. these correlations alter the configurational enthalpy. there is no provision for any preference for the first-nearest neighbors of a wrong A-atom to be B-atoms.11 Alloy phase diagrams Compendia of binary and ternary alloy phase diagrams from experimental results and theoretical assessments have been assembled as handbooks and texts [9.23. Configurational entropy is not the only source of entropy. and these reduce the configurational entropy. • Entropy. (For example. and for shear transformations there is no change in the configurational entropy. but they are also fascinating for what they show about tendencies for interatomic interactions. 10. being the sublattice concentration. More quantitative approaches of obtaining equilibrium phase diagrams are still based on minimizing the free energy of multiphase mixtures. The occupation probability for a site. 2. Most quantitative electronic structure calculations of today are based on the approximation that the electron exchange and correlation energy depend on the density of electrons. are also available. 7. The intricacies of these real phase diagrams are useful for practical heat treatments of alloys. The problem with Eq. A better thermodynamic theory is possible by defining “pair variables” to account for the short-range preference of neighboring atoms [14]. and is often satisfactory for many predictions. The configurational entropy can be calculated with more accuracy than in Eq. The reader is encouraged to peruse them to understand the depth of information that is available today. but higher-order cluster approximations are practical. 15].4. atoms tend to satisfy their chemical preferences over short distances. and in some systems it makes a minor contribution.) The pair approximation provides a significant improvement in calculating the configurational thermodynamics of an alloy. A better expression is obtained with the pair approximation Eqs. Methods for calculating multicomponent phase diagrams from model free energy functions. There are short-range correlations between atom positions. 16:48:42 .48 Essentials of T–c phase diagrams The Bragg–Williams approach is a theory in the “point approximation” because the theory uses only a simple sublattice concentration at each site (or “point”) in the structure. The reliability of these methods is now known. such as the CALPHAD approach [16–18].

33 B0. and equal compressibilities for all compositions. In the dilute limit of the impurity concentration.1 Use the Gibbs phase rule. (c) An atom in a bcc crystal has eight nearest neighbors (coordination number z = 8). n.3 Random solid solutions (a) Use the algorithm of the binomial expansion of (A + B)N to justify why the number of arrangements of n equivalent atoms on N sites is N! /[(N − n)! n! ]. At elevated temperature the vibrations may no longer be harmonic or independent. but explain why you can ignore the term (1 − c) ln(1 − c) in Sconfig (c).5. and the energy cost for n of them (on a lattice of N sites) is nε.34 for the lowest temperature shown in the model? (d) If pressure were applied to this system.49 Problems Since atom vibrations are the major source of heat in materials. where c = n/N is   ε .33 C0. If the chemical composition of the alloy is c = 0.3 and Chapter 24. c → 0. what is the numerical probability that a given atom in a bcc A–B alloy has n B-atoms as nearest neighbors. Assume also that the three solid phases have identical specific volumes. too. 0. (a) How many phases are in equilibrium at the central eutectic point? (b) Why is the β–liquid boundary a surface in this model? (c) How many phases are in equilibrium at the composition A0.50) c(T) = exp − kB T (Hint: Use the entropy result for a solid solution. Eq.2 Suppose the energy cost of an impurity atom (or a point defect like a vacancy) is ε. to explain your answers to these questions about the ternary eutectic diagram shown in Fig. Set dF/dc = 0 to find the minimum of F(c). what would happen to the liquidus surfaces and the eutectic point? Assume that the liquid has a larger specific volume than any of the solid phases. 16:48:42 . 2. 2.37. 10.5) vs. Motions of magnetic spins and electronic excitations can make important contributions to the entropy.4).4 A “high-entropy alloy” is a multicomponent alloy that could have a higher configurational entropy than a binary alloy. where n ranges from 0 to 8? Make a graph of P(8. n. 2. where temperature causes some disorder in the occupancy of electronic states near the Fermi level (Sect. 6.2. (b) Derive the configurational entropy of mixing from part a using the Stirling approximation. show that the impurity concentration c(T) at temperature T. it is not surprising that differences in crystal structure may cause differences in heat capacity and in vibrational entropy.) 2. (2. and this adds considerable complexity to the problem as discussed in Sect.14. Problems 2. 2. Metals always have some contribution from electronic entropy.

2. Draw sketches of free energy vs.5 2. and argue why it is a maximum when cj = 1/n for all j.7 Temperature (T ) [ºC] 2.19 consists of phase boundaries made from straight lines.6.Essentials of T–c phase diagrams 50 2.23).5. 500 L L+α 400 Fig. with common tangents. Figure 2. graph them at different temperatures and construct a peritectic phase diagram like that of Fig. and by adding an eighth component to a seven-component alloy? (d) What is the effect of chemical short-range order on the result of part c? (e) For metallic alloys with larger n.  (Hint: Use the entropy expression S = −kB nj cj ln cj . Show how the positions of these F(c) curves shift with temperature.22 and 2. 2. The shape of the phase diagram will be different from that of Fig. however.8 (a) Show that the configurational entropy of an n-component alloy can be as large as Smax = kB ln n. Figure 2. For example.20 shows free energy versus composition curves for a solid phase and a liquid phase having a phase diagram with continuous solid solubility. (c) How much does the configurational entropy change by adding a third component to a binary alloy. what other sources of entropy might be comparable with the effect of part c? Calculate the critical temperature for unmixing for a binary alloy with composition c = 0. Using free energy versus composition curves like those of Fig. 2.13a. notice that two common tangent curves can be drawn for the positions of the two curves in Fig. composition curves. 400 ◦ C. 16:48:42 . for the phases α. Explain why this sets the critical condition. (a) Assuming as usual that the entropy of the liquid is much higher than that of the solid.20.) (b) Make a plot of Smax versus n for 2 ≤ n ≤ 10. 2.6 2. and liquid at temperatures of 300 ◦ C. and 500 ◦ C that are consistent with this phase diagram.13b.19 β α+β 0 t L+β α 10 50 B [%] 90 100 Simple eutectic phase diagram constructed from straight lines. construct two or three more pairs of F(c) curves for this system that are useful for constructing a phase diagram. 2. β. 2. and explain why this approach is irrelevant when the interchange energy V > 0. The curvature of F(c) is zero at the critical temperature. Use the energy and configurational entropy of the mean field theory (for example Eqs. You may need to widen the F(c) curve of the β-phase.

16:48:42 . which is a special point in the diagram? (d) Suppose the F(c) curves for the solid and liquid were interchanged in Fig.Problems 51 S F L t Fig. (b) Draw the phase diagram consistent with your curves in part a. (c) What is the curvature of the solidus curve around c = 0. 2.20 A Concentration B F(c) curves for solid (S) and liquid (L) at an intermediate temperature. Draw the phase diagram for this case.20.5. 2.

The heat equation has been known for centuries. 1. 3. finite-element methods are practical for many problems.1 The diffusion equation Writing the kinetic master equation in the form of Eq. for example..g. Mention was made of interstitial diffusion and interstitialcy diffusion. The Laplacian. and often prove more efficient than analytical methods.3 Diffusion In solids. such as the basic solutions of Gaussian functions and error functions for one-dimensional problems.3 and illustrated with Fig. such as the rising of a hot liquid in a gravitational field. Some of these methods are standard for diffusion in materials. 1. 1. the reader may consult classic texts in mathematical physics (e. For problems in ellipsoidal coordinates.1 The diffusion equation has the same mathematical form as the equation for heat conduction. and methods for its solution have a long history in classical mathematical physics. This is the subject of Chapter 22. [19]). but these are beyond the scope of this book. if solute concentration is replaced by heat or by temperature. each with their own special functions and orthogonality relationships.1) 1 Convective currents can be driven by differences in density.5. changes in the local atomic structure during diffusion can depreciate the diffusion equation because the “diffusion constant. By deriving the diffusion equation from the kinetic master equation. is not constant.” D. is separable in nine other coordinate systems. Because diffusion depends on atomic-scale processes. we can later replace the assumption of random atomic jumps with an assumption of chemically biased jumps to predict the kinetics of chemical ordering or mixing. Mass transport in glasses and liquids can also occur by atomic-level diffusion. but for gases or fluids of low viscosity there are larger-scale convective currents with dynamics quite different from diffusion. The vacancy mechanism for diffusion in crystals was presented in Sect.7. 16:48:47 . Today. Chapter 3 also presents the method of separation of variables for three-dimensional problems with Cartesian and cylindrical coordinates. This can be a serious problem when using the diffusion equation to describe the kinetics of a phase transformation. This chapter concludes by showing how the kinetic master equation can lead to thermodynamic equilibrium. atoms move by a process of diffusion. however.25 motivates a matrix description of the kinetic processes W (t) N (t) = N (t + t) ≈ ∼ ∼ (3. ∇ 2 .

9a. If t is small.3. Two such matrix elements are shown in Fig. 1. This approach has an advantage for numerical computations. after m intervals of t the new contents of the bins will be m .9b.1 The diffusion equation 53 where N (t) is a column vector that we lay out along the bins of Fig. 1. and W (t) is ∼ ≈ a two-dimensional matrix that gives the new contents after time t.

but the elements one off the diagonal are now δ/2. 0 ⎥ ⎢ ⎥⎢ .. first assume zero atom jumps in the time t. so in this case I N (t) = N (t + t).. . the probability of an atom entering a bin from an adjacent bin is also δ/2. 0 ..25 is ⎡ ⎤ ⎤⎡ 1−δ δ/2 . assume each atom has only a small probability δ of leaving its bin in the time interval t. 0 ⎥ ⎢Nn−1 (t)⎥ ⎢ 0 ⎢ ⎥ ⎥⎢ ⎢ ⎥ ⎥⎢ ⎢ 0 . and to our construction of a kinetic master equation for diffusion. δ/2 1−δ 16:48:47 .. .. : ⎥ ⎢ ⎥⎢ . right. ... . up. 1. down. N2 (t) .2) W (t) N (t) = N (t + mt) ... • All atoms have the same jump probability (unaffected by the presence of other atoms). ⎥ ⎢ ⎥ ⎢ ⎥ ⎥⎢ ⎢ ⎥ ⎥⎢ . 0 ... 0 δ/2 1−δ δ/2 0 . Nn+1 (t) . I . ... 0 ⎥ ⎢Nn+1 (t)⎥ ⎢ ⎥ ⎢ ⎥ ⎥⎢ ⎢ : ⎥ ⎢ .. ⎥ ⎢ ⎥⎢ ⎢ ⎥ ⎥⎢ ⎢ δ/2 .. • If an atom has probability δ of jumping out of a bin in Fig. The W-matrix is close to diagonal. . In this case W (t) ≈ ≈ must be the identity matrix. Equation 1.. . and its probability of entering an adjacent bin is δ/2. Nn−1 (t). ⎥ ⎢ . .. out). ⎥ ⎢ ⎥ ⎥⎢ ⎢ ⎥ ⎥⎢ . δ/2 1−δ δ/2 0 0 . • An atom can jump only into an adjacent bin (but this is not an essential assumption for obtaining the diffusion equation.. For our matrix equation we first arrange the two column vectors in correspondence with the bins in Fig... ∼ For the structure of W (t). it has an equal probability δ/2 of going left or right (in three dimensions the probability is shared as δ/6 between left. ..9a.. in. ⎥ ⎢ ⎥ ⎥⎢ ⎢ : ⎥ ⎢ . ⎥ ⎢ 0 ⎣ ⎦ ⎦⎣ .9.. . .3) N (t) = N1 (t).... Likewise. ≈∼ ∼ ∼ Next. They are important to remember whenever using the diffusion equation for a problem in materials science. . 0 0 δ/2 1−δ δ/2 . as shown by Problem 2 in Chapter 9). The probability of it remaining in the bin is therefore 1 − δ. with all 1s on its diagonal. . {n}.. (3... : ⎥⎢ . . . and 0s elsewhere. . . 0 ⎥ ⎢ Nn (t) ⎥ ⎢ ⎥ ⎥⎢ ⎢ ⎥ ⎥⎢ ⎢ 0 ⎥ ⎢ . The operation ≈ of this identity matrix on the vector N (t) preserves the contents of all bins at time t + t. 1. Nn (t).. and their contents {N}   (3.. . ≈ ∼ ∼ The following assumptions are fundamental to the diffusion equation... δ/2 ⎥ ⎢ ..

5 for δ Nn (t + t) = t/2Nn−1 (t) + (1− t)Nn (t) + t/2Nn+1 (t) . which is nearly 1...5) Consider the middle line on the right-hand side of Eq. ∂x2 a2 (3.10) . = ∂t 2 ∂x2 16:48:47 (3... which is the value of Nn at time t + t. the nth bin would have no change in the time ≈ t. 3. as t becomes small ∂Nn Nn (t + t) − Nn (t) = . such as n and n + 1.   Nn (t + t) − Nn (t) = Nn−1 (t) + Nn+1 (t) − 2 Nn (t) . and use Eq. meaning no change. times t δ = t.. The column vector on the right-hand side of Eq.. (3. ∂t t (3. With atom movements. giving a2 ∂ 2 Nn ∂Nn . .6) (3. multiplied by 1 − δ.. ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ .4 is actually quite simple.4) ⎢ ⎥ ⎢ ⎥ ⎢δ/2Nn (t) + (1−δ)Nn+1 (t) + δ/2Nn+2 (t)⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ . t 2 (3. ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢δ/2Nn−2 (t) + (1−δ)Nn−1 (t) + δ/2Nn (t)⎥ ⎢ ⎥ ⎢ ⎥ = ⎢δ/2Nn−1 (t) + (1−δ)Nn (t) + δ/2Nn+1 (t)⎥ (3. however. the nth bin receives contributions from the two surrounding bins..8) The smallest x is a. . proportional to δ/2. and Nn (t + t) = Nn (t).9) which we identify in Eq. which is the distance between adjacent atomic planes. the probability δ can be rewritten as an atom jump rate (per second).4. Its largest terms are the middle ones. 3.. ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ .Diffusion 54 ⎡ ⎤ . ⎢ ⎥ ⎣ ⎦ .7. 3. If δ were zero.7) The difference equation 3.7 becomes a partial derivative of Nn . ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ . An expression for the second spatial derivative is Nn−1 (t) + Nn+1 (t) − 2 Nn (t) ∂ 2 Nn = . W would be the identity matrix. Because δ is the dimensionless probability that an atom will jump from its bin in time t.. 3.

17) (3.14) ∂t 6 ∂x2 ∂y2 ∂z2 ∂c(x.20) known as “Fick’s first law. ∂t ∂x (3.15) = D ∇ 2 c(x. Equation 3. t) . out. (3.19) (3.11) where N0 is the number of crystal sites on the 1 cm2 plane. N0 aA (3. t) (3. 2 ∂x ∂c j = −D . The net flux is to the right because the bin (atomic plane) at the left contains more atoms than the bin at the right. in.9b can be used to obtain an equation for the net flux of atoms between planes.16) Equation 3. Figure 1.4.” sometimes called “Fick’s second law. t) . down.10 becomes the one-dimensional diffusion equation a2 ∂ 2 c(x. The three-dimensional diffusion equation is obtained by recognizing that only 1/6 of the atom jumps go to the right.n+1 and Wn+1.3. ∂t where D= a2 . Because 16:48:47 . which are often less useful than concentrations because fluxes are harder to measure. the net number of atoms moving across the boundary is  t  Nn − Nn+1 .1 The diffusion equation 55 For atomic diffusion. 6 (3. to the right across the boundary is the difference: (right flux) − (left flux). ∂x j t = (3. and define a concentration [atoms cm−3 ] at the position x for the nth bin as c(x. t) − c(x. consider atom movements between planes of area A separated by the distance a. t) + + = c(x. labeled as Wn. this equation works with fluxes.9b illustrates an important feature of fluxes.n in the figure as δ/2 or t/2. In three dimensions   ∂2 ∂2 a2 ∂ 2 ∂c(x.18) (3.12) (3. with the other 5/6 going left. t) ≡ Nn (t) 1 . 3. 2 a a (−1) [c(x + a. j.15 is “the diffusion equation.13) where the diffusion coefficient was defined as D = a2 /2 (which has units [cm2 /s]). = ∂t 2 ∂x2 ∂ 2c ∂c =D 2 . In the time t. t) .” The process of Fig. The total flux. up. t)] . t) ∂c(x. 1. We recognize the matrix elements in Eq.” Physically. j= 2 a a2 ∂c j=− .

Applying Eq. In three dimensions. Eq.21) ∇ ∂t states that a flux leaving a small volume causes a decrease in concentration in that volume. = D∇ ∂t ∂c = D ∇ 2c .24) each atom jumps with the same probability per unit time. ∂t 2t 4Dt2 4π Dt   x2 ∂ c 2x e− 4Dt − c(x.25) −→ where ∇c is the gradient of c.22) (3.1 Conservation of atoms The conservation of atoms gives an important relationship between Fick’s first law (Eq.27) (3. as it will be if all atoms jump with equal probability − −→ ∂c  · ∇c .13) is a Gaussian function that depends on space x and time t c(x. 3. The divergence theorem  · j = − ∂c . 3. ∂t which is again the diffusion equation.2 Gaussian and error function solutions to the 1D diffusion equation 3.1 Gaussian Solution One type of solution to the one-dimensional diffusion equation (Eq. 3. (3. t) = √ . 3. t) = √ . 3. 3.15.2. ∂x 4Dt 4π Dt 16:48:47 (3.28) . Fick’s first law is −→ j = −D ∇c .26) The partial derivatives of c(x. t) are   x2 c x2 ∂ 1 e− 4Dt − + c(x.25) −→ ∂c  · [D ∇c] . = −∇ ∂t Assuming again that D is a constant.15).23) (3.25) and the diffusion equation (Fick’s second law. (3. the flux to the right is greater.21 to Fick’s first law (Eq. t) = √ c 4π Dt x2 e− 4Dt . (3.56 Diffusion Box 3. (3. 3. Eq.

All have physical importance. Adding these halves of the solution is possible because the atoms are assumed to move independently. 3. (3. without interaction – there is no effect on the time evolution of the concentration profile at x > 0 by doubling it.1 Position A Gaussian function at top. t) is proportional to accumulation of solute ∂2 c(x. t) and its spatial partial derivatives are shown in Fig. and the second derivative is the rate of accumulation.3.13. t) → c(x. nonvolatile.2 The Gaussian solution is a useful solution when a very thin layer of diffusing species is placed between two thick blocks. and the concentration profile increases in breadth as √ t. t) = − . With the surface at x = 0. the second spatial derivative of c(x.     x2 c 2x 2 2 ∂2 − 4Dt e c(x. shows the validity of the Gaussian solution. t)) and added to the profile at x > 0. the first derivative is the negative of the flux.31) The Gaussian function solution becomes a Dirac δ-function when t → 0. the concentration profile for x < 0 is inverted (c(−x.1. and its derivatives below.30) and from Fick’s second law. The Gaussian concentration profile c(x. 3. the first spatial derivative of c(x. t) is proportional to the negative of the flux j ∂ c(x. preserving the integral of the concentration (and conserving atoms). From Fick’s first law. and allowed to diffuse into the solid that extends to positive x.29) Substitution of these partial derivatives into the one-dimensional diffusion equation.2 Gaussian and error functions for the 1D diffusion equation 57 c(x) ∂c ∂x ∂2c ∂x2 t Fig. 16:48:47 . ∂x D (3. t) = √ + − . It is also useful when a thin. Inflection points and zeros are indicated with vertical dashed lines. 3. layer is deposited on the surface of a solid. t) = ∂x2 ∂ ∂t c(x. The Gaussian represents concentration. 2 The peak concentration of the species decreases with time as 1/√t. Eq. t) D . 4Dt 4Dt ∂x2 4π Dt (3.

and the concentration increases with time. There is concentration loss where the curvature is negative. Overlaps of the Gaussian profiles are no problem for the diffusion behavior. The second spatial derivative gives the rate of accumulation at each point x. This is the situation at the time t = 0. n + 1.30). and allowed to evolve to give the solution at times t ≥ 0.. to its spatial derivatives.31). there is no net flux between any pair of adjacent bins. has the same concentration in bins n − 1. and how it changes at a point in space corresponding to the bin n shown in Fig. and is high at the sloping parts. so for a constant slope cx0 the net gain of atoms in bin n from n + 1 equals its net loss from n to n − 1. the flux is proportional to the difference in concentrations of two adjacent bins. taken to the limit as δ-functions. n. These points are illustrated with the Gaussian function in Fig. where the constant of proportionality is D (Eq. Its first spatial derivative gives the (negative) flux at each point x. capable of handling any diffusion problem inside a homogeneous material. 3.000 Gaussian functions. We can center a narrow Gaussian function at each plane (a narrow width will confine it to one plane).2 Green’s function solution Consider a one-dimensional concentration profile as bins corresponding to atomic planes. under the assumption of independent atom movements. The flux is zero where c(x) is flat. In practice. 3. where an arbitrary function can be approximated as a sum of narrow Gaussians.000 planes we can center 1.2 Concentration profiles during diffusion Random jumps of independent atoms was our main assumption about how atoms move by diffusion – each atom moves independently of the other moving atoms. it is 16:48:47 . has a difference in concentration between bins n and n + 1 that is larger than the difference between n and n − 1 (assuming cxx0 > 0). • A concentration having a linear variation with position. to describe a 1D composition profile. for example. The net flux into bin n is therefore greater than its loss. In principle. However. ∂c/∂x = cx0 . 3.2. It is zero at inflection points (with zero curvature). Consider a concentration profile in one spatial dimension c(x. ∂ 2 c/∂x2 = cxx0 .2a. The concentration in bin n remains constant in a linear concentration gradient. The change is related to the concentration profile itself. as expected from Fick’s first law ∂c/∂x = −j/D. this is quite a general approach. The Gaussian is the Green’s function of this 1D diffusion problem – it is convoluted with the initial profile. and a net flux of atoms to lower n (assuming cx0 > 0. c(x) = c0 .9a. t). and no change of concentration in bin n. or more specifically. 3. see Eq. An important effect of diffusion is the time evolution of a concentration profile. i. 3.e. With random jumps between bins.58 Diffusion Box 3. 1. The idea is illustrated in Fig. Our basic assumption of independent atom motions then allows us to add the concentration profiles that evolve around each atom plane at later times. Along 1. and concentration gain where the curvature is positive.1. ∂c/∂t = D cxx0 . The effects of its spatial derivatives are: • A flat concentration. • A concentration function having curvature with position. has more atoms jumping from bins of higher n than lower n.

0) δ(x − x )dx . 3. but have zero amplitude for x > 0. Near x = 0 the sum of Gaussians shows a broader interface for longer times. The individual Gaussians evolve in time. (b) A set of evenly spaced Gaussian functions with amplitudes set so that their sum matches a Heaviside function. but with unchanged area. Recall that the convolution of a Dirac δ-function with any function returns the function.33) Equation 3. c(x . and a sharp drop in c(x) at x = 0.2 a x=0 b (a) A set of evenly spaced Gaussian functions with amplitudes set so their sum approximates the function c(x). t) = ∞ (x−x )2 1 e− 4Dt dx . In what follows we use a convolution. Figure 3.34) . 3. and the same Eq. which we convolute with the initial profile.2b shows the “error function solution” obtained below. 3. 16:48:47 (3.32. this Gaussian is a Green’s function.33 gives the concentration profile at later times  c(x. 0) = ∞ −∞ c(x .3  c(x. 0) √ 4π Dτ −∞ (3. For these δ-functions we can use normalized Gaussian functions of Eq. The advantage of Eq. and for x  0 the sum is 0. c(x . 3.33 describes the initial profile in an equivalent way to Eq.34 is that the integral is usually difficult to perform. Away from x = 0 their sum remains constant because their areas are conserved – for x  0 the sum is c0 . where τ is very small.2 Gaussian and error functions for the 1D diffusion equation 59 c0 t~0 c=0 x=0 c(x) c0 t >0 c=0 x t Fig.32) This integral amounts to summing a large number of δ-functions. Bottom shows broadening of all Gaussian functions at a later time. 0) √ 4π Dt −∞ The disadvantage of Eq.33 is that all the Gaussian functions broaden independently with time. and how their sum develops a rounded corner near x = 0. each with a different center x . where all narrow Gaussians have the same amplitude for x ≤ 0. 3.3. and our goal is to calculate how this interface broadens with time. becoming wider and shorter. which is often left as an integral. challenging to obtain simple mathematical expressions for the solution.26 at t = τ . 0) = lim τ →0 ∞ (x−x )2 1 e− 4Dτ dx . (3. For the concentration profile at t = 0  c(x. Top shows narrow Gaussians near t = 0. 3. 3 Formally.

c(x.40) erf(z) ≡ √ π 0 and it is known that erf(∞) = 1. t) = c0 /2 and c(∞.41) 1 − erf √ 2 4Dt It is useful to know that erf(0) = 0 and erf(−∞) = −1. 3. and reappears many times when working with diffusion equations. θ(x ) √ (3. but with a cutoff in the upper limit of integration (at the end of the line of centers for the Gaussian functions)  0 (x−x )2 1 c0 √ (3. 16:48:47 .60 Diffusion We now calculate the time-dependent broadening of a concentration profile that initially has a sharp interface. 3. giving    c0 x c(x. which would be tedious.39 is as far as we can go in the analytical reduction. this would have to be done multiple times. where θ (x) is the downstep function illustrated at the top of Fig. From Eq. (3. t) = c0 √ e−η dη . x/ 4Dt}. 3. (3.37) 4π Dt −∞ The ratio (x − x )2 /(4Dt) is dimensionless. c(−∞.39) π √x 4Dt Equation 3. but far from the interface the composition remains unchanged for all time. initially with an abrupt change in composition in the middle. ∞} to {∞. t) = c0 √ e− 4Dt dx . The next method to solve the diffusion equation handles short bars or long bars quite nicely. and the tail of the composition profile can be inverted and superimposed on the main profile. t) = 0. The dx changes to √ − 4Dt dη (allowing us to switch the limits of integration)  ∞ 1 2 c(x. The breadth of the composition spread near x = 0 √ increases as t. t) = c0 and c(0. t) = 4π Dt −∞  0 (x−x )2 1 c(x.2b. (3. however. (3. but it is traditional to express this result somewhat differently. We make the standard substitution x − x η≡ √ 4Dt (3. The definition of the “error function” is  z 2 2 e−χ dχ . This error function solution is useful for an infinite bar of material.34  ∞ (x−x )2 1 e− 4Dt dx .39 is rewritten in terms of the error function. t) = 4π Dt −∞ The same initial concentration profile can be constructed by setting the concentration profile to a constant c0 . it reflects from the ends. Following convention. Eq.35) c(x.38) √ and note that the limits change from {−∞.36) e− 4Dt dx . sometimes it is possible to use the trick described with the Gaussian solution – when the solute is not volatile. so for all times. t) = . If there are ends to the bar. For a short bar.

e. center. Each of the special functions is treated independently for its time evolution.42) although there are numerical factors that differ from problem to problem (cf. t). the dimensionless form is D/η2 .3 Fourier series solutions to the diffusion equation 3. width. These special functions form a complete set that can describe any function of composition in the coordinate system. y. it applies to all solutions to the diffusion equation. The concentration profile at any time is the sum of these independent Gaussians. The result is a transformation to simple ordinary differential equations for the x. Using these Gaussians as “kernels” is a systematic mathematical approach to solving partial differential equations. To be used as an argument in a function. This leaves one parameter. This method of solution. and they are all added to give the total profile at later times. The one other systematic approach involves selecting an appropriate coordinate system. Each Gaussian is specified completely by three parameters – area.. For the same initial and boundary conditions. Eq. 3. t) (with two independent variables) is determined by a single parameter through a relationship between x and √ t. (3. This pertains to the error function solution. to rescale the Gaussian for a different time. and the center does not move (unbiased diffusion jumps). t). In fact. which depends on both time and position. since its component Gaussian functions evolve independently in time.77 below. as can be seen in the time-dependence of Eqs.1 Separation of variables in Cartesian coordinates The diffusion equation 3. The previous analysis assigned a sharp Gaussian function to each point in space. Evidently the value of the function c(x. and using a sum of special functions for the coordinate system. 9. z. the concentration is the same. y. and how a sharp Gaussian evolves with time into broader Gaussians. is called “separation of variables. 3.” meaning that we obtain equations where all variables of one type can be separated to one side of an equation. This relationship is η = x/ t. For the same η. 16:48:47 .61 3. too.3 Fourier series solutions to the diffusion equation Box 3.67 and 3. and select weights for the independent sine waves. z.8). Boundary conditions and initial conditions are then used to specify the problem precisely. different materials have the same concentration at the same value of D/η2 = D t/x2 . in this case the width. and then let all the Gaussians broaden in time. and t dependences of c(x.3 Characteristic distance A characteristic distance of diffusion can be obtained by a scaling argument. c(r. Consider again the Gaussian solution. explained in this section for sine waves in rectangular coordinates.3. This scaling behavior of the concentration profile during diffusion gives a handy estimate of a characteristic distance of diffusion in the time t √ x ∝ Dt . i. By normalization (conservation of atoms) all these Gaussians have the same area.15 is a partial differential equation for composition..

49) (3. 3.46 is therefore possible only if both sides equal a constant. so the term in the Laplacian ∂ 2 c/∂z2 = 0. − k12 = −k22 . Equality in Eq.43) ∂t ∂x2 ∂y2 Fundamental to the method of separation of variables is the assumption. The problem is formulated in two dimensions. which is our first constant of separation.44 into 3. Likewise. y. call it −k12 .43 and differentiating the relevant variables   2 ∂ 2 Y(y) ∂ X(x) ∂T(t) Y(y)T(t) + X(x) T(t) . 3. The diffusion equation 3. which is not true for all partial differential equations but works here. (3.45) and dividing all terms by the product XYT gives ∂T(t) ∂t T(t)  ∂ 2 X(x) =D ∂x2 X(x) +  ∂ 2 Y(y) ∂y2 = −k12 . the spatial variables x and y appear only on the right. each involving only one variable.48) 2 The spatial variables can be separated further  ∂ 2 X(x) − k12 D ∂ 2 X(x) ∂x2 X(x) =D = −D ∂x2 X(x) ∂ 2 Y(y) ∂y2 Y(y) + ∂ 2 Y(y) ∂y2  Y(y) .Diffusion 62 The method of separation of variables differs in important details for different coordinate systems.15 is4  2  ∂c ∂ c ∂ 2c + =D .50. Here we use the Cartesian coordinate system. so the two sides equal a second constant of separation.44) Substituting Eq.46) We now make a key observation about Eq. −k22 . 3. The variable t appears only on the lefthand side and not the right. and the time variable cannot account for any spatial dependence. For later convenience. (3. =D X(x)Y(y) ∂t ∂x2 ∂y2 (3. (3. 3.47) T(t) = T0 e−k1 t . The spatial variables cannot account for any time dependence. (3. This leads to a simple solution for T(t) ∂T(t) = −k12 T(t) . 4 For our two-dimensional problem there is no variation of the concentration along zˆ . t) = X(x) Y(y) T(t) .46. Y(y) (3. for which the method is simplest. that the concentration can be expressed as a product of three functions. 16:48:47 . ∂t (3. and not on the left.50) where we have again made the observation of independence of the variables x and y across two sides of Eq. Giving these three functions obvious symbols c(x. with no variation of the concentration along the direction zˆ (where zˆ is a unit vector in the z-direction).

For initial conditions. 3. t) = Ax Ay e−(k2 +k3 )t sin √ x sin √ y . t) = 0 . t) = 0 . and the opposite corner is at x = Lx and y = Ly . c(0. Since the solute is lost at the surface. that is.51) D D     k3 k3 (3. t) = e−(k2 +k3 )t X(x) Y(y) . is c(x. since they cannot be zero at the origin. At the surface.55.56) c(x. 3. t) = 0 . 3. This is relevant for atoms of hydrogen diffusing out of a solid and leaving as a gas of low density. with time dependence. D D where the two constants of separation are combined into one for clarity k32 ≡ k12 − k22 .52 give     k3 k2 2 2 (3. The boundary conditions involve only the spatial coordinates. The problem we are considering has no variation of composition along zˆ. Intersecting the x–y-plane. y.3 Fourier series solutions to the diffusion equation The ordinary differential equations for X(x) and Y(y) in Eq. With Bx = 0 and By = 0 Eqs. This is achieved when the arguments of the sine functions are multiples of π at Lx and Ly  − c(x. 2 2 (3.51 and 3. c(x. the loss of solute from the surface causes a decrease of composition near the surfaces. y. we choose a uniform composition throughout the bar. 3. one corner is at the origin. we assume that solute is lost. (3.2 Boundary conditions and initial conditions We must now specify details of the problem. t) = Amn e m2 n2 + 2 Lx2 Ly  π 2 Dt  sin πm x Lx  16:48:47  sin  πn y Ly .50 have standard harmonic solutions.52 are inappropriate.3. but only if solute cannot pass through its surfaces normal to zˆ.54) 3. with coefficients to be determined later     k2 k2 X(x) = Ax sin √ x + Bx cos √ x .57) .63 3. the boundary conditions and the initial conditions.52) Y(y) = Ay sin √ y + By cos √ y . (3. and are valid for all time. c(Lx . y. t) = 0 . y.51 and 3. c(x. the surface concentration is 0. and about what happens to the atoms at the surface. y. With time. Our solution could pertain to an infinite bar oriented so its long direction is along zˆ.53) The general solution. and eventually a loss of solute from the center of the bar. (3. A shorter bar or even a plate could also be appropriate. D D The constants k2 and k3 are obtained by imposing the other two boundary conditions at x = Lx and y = Ly . and cannot satisfy the first two boundary conditions in Eq. This amounts to information about the initial concentration profile. Ly . (3. 0.55) Immediately we see that the cosine solutions of Eqs.

62) are elementary and integrate to cosines evaluated between limits      x=Lx . Subtracting the results from the two limits gives zero when m and n are even. 3.64) 2 2 The integrals on the LHS (Eq.62) 0  m. c . and fit the uniform concentration.3. the x-part of the integrand becomes sin2 (π mx/Lx ). but is illustrated in part by Fig. which shows the orthogonality of two sine functions with m = 4 and m = 8. owing to the orthogonality of sine functions over their interval of periodicity.60) c = Lx Ly m n To solve for the Amn . k2 = Lx √ π D n .63 – when m = m . at t = 0. (3. which integrates to Lx /2 over the x-interval.65) LHS = 2   cos Lx Ly π mn x=0.Diffusion 64 so the boundary conditions give √ π D m . 3. Lx Ly LHS = 0 RHS = (3. This is a standard result. where n is an integer .n         LxLy  πm πm πn πn Amn sin x sin x sin y sin y dx dy . y=0 RHS = Am n At the lower limits. Noting that the exponential factor with the time dependence in Eq. Lx Lx Ly Ly (3. k3 = Ly (3. (3. the cosines evaluate to +1.3 Orthogonality and initial conditions Finally we impose the initial conditions. (3.60 by specific sine functions (denoted by m and n ).3. 3. 3.58) (3. 3. This requires multiplying both sides of Eq. Lx Ly  (3. At the upper limits. and likewise for the y-integral Lx Ly .63) 0 0 The integrals of Eq. y=Ly c Lx Ly π m πn x cos y . the cosines evaluate to +1 for even integers. 3. and −1 for odd.61)      π m πn c sin x sin y dx dy .59) 3. where m is an integer .63 are zero unless m = m and n = n . 16:48:47 . and integrating over the intervals from 0 to Lx and 0 to Ly LHS = RHS. Orthogonality simplifies Eq. we make use of orthogonality relations of the sine functions. We need a sum of sine functions along x and y that matches the concentration profile (and each of these sine functions will evolve independently in time). so fitting to the initial conditions forces m and n to be odd.57 equals 1 at t = 0      πm πn Amn sin x sin y .

64 and 3.68) n = 1. .69) The solution for c(x. y. 3.66)     πm πn sin x sin y .67 is a Fourier series in x and y with coefficients that depend on time. and decreases uniformly in height with time. so Lx = Ly . 3.3. the exponential is +1. Equating the LHS and RHS of Eqs. 16:48:47 . . y.2b. t) in Eq.n  − 16 e 2 π mn 16 c . and eliminating the primes for clarity.3 Fourier series solutions to the diffusion equation 65 Amplitude 1 0 –1 t Fig. Lx Ly (3. The rate of decay of a Fourier coefficient goes as m2 + n2 . 3. At right is shown one-quarter of the interval where the product cancels the product in the last quarter-interval (the dashed curve is shown to illustrate the symmetry).67 is then exp(−π 2 D(m2 + n2 )/Lx2 t).67) m = 1. the Fourier coefficients Am. π2 m n m2 n2 + 2 Lx2 Ly  π 2 Dt (3. 3.5 The time dependences of the Fourier coefficients do not depend on composition. t) = c e Lx Ly for a square bar at late times. When t = 0.70) c(x. y. meaning that the higher-order coefficients decay first. 3. (3. . only on the order of the sine waves. Suppose for a moment that the bar has a square cross-section. by adding a background concentration instead of having zero concentration outside the bar. 7. (3.n are Amn = and the full solution is  c(x. 7. This late-stage composition profile is a smooth mound of considerably less height than c . and the composition profile can be approximated with m = n = 1. 5. incidentally. 5. . so after only short times the edges broaden. The higher-order Fourier coefficients are the ones needed to give the composition profile its sharp edge. Even for 5 We can modify this solution. 3.3 0 L Interval Two sine waves on the interval 0 − L. The exponential in Eq. with sharp edges that drop to zero at the edge of the bar. much as shown in Fig. . . (3. giving     π π  −2π 2 Dt/Lx2 sin x sin y .65. and the Fourier series describes a rectangular composition profile of composition c . t) = c  m. 3. the higher-order coefficients can be neglected. to be multiplied and integrated to show orthogonality. At late times.

46.1L2x /(π 2 D).4. oblate spheroidal. It turns out that there are a total of 13 coordinate systems for which the Laplacian is separable.Diffusion 66 Concentration c' 0 – c' Solution valid 0 a Lx x 2Lx Concentration c' 0 – c' t Fig. 3.4 Bessel functions and other special function solutions to the diffusion equation Solving the diffusion equation by the method of separation of variables proves successful for coordinate systems other than the Cartesian system. The solution to the diffusion equation involves these mathematical steps: • separation of the diffusion equation to obtain a set of constants of separation. as we did in Eq. • use of boundary conditions to pick acceptable families of the special functions characteristic of the coordinate system. cylindrical. bispherical. shorter times the composition profile is soon dominated by the lowest terms in the Fourier series. 3. (b) profile after t = 0. prolate spheroidal. The important consideration for other coordinate systems is whether the Laplacian ∇ 2 is separable. 3. ellipsoidal. (a) Initial condition at t = 0. spherical. 16:48:47 . conical. 6 Rectangular.4 b Solution valid 0 Lx x 2Lx Concentration profile along x-axis (black) and individual terms with 1 ≤ m =≤ 11. paraboloidal.6 so the concentration can be expressed as a product of spatial functions (times exponential functions for time). The time dependence is always straightforward to separate from the spatial coordinates. toroidal (although the latter two require a somewhat different approach). parabolic. 3. as shown in Fig. elliptic cylindrical. parabolic cylindrical.

67 3.4 Bessel functions and other special function solutions • expanding the initial concentration profile as a sum of special functions for that coordinate system. by grouping terms in powers of r it can be confirmed that all vanish as required. however. Inside the rod. • using the orthogonality relationships of the special functions to determine the weights of the different functions in the expansion of the initial profile. We select a boundary condition for our cylindrical rod of radius a so that the concentration at a is 0.74) (3. The parameter λ rescales the Bessel function.77) . ∂r 2 2 4 2 4 6 λ2 3λ4 r2 5λ6 r4 ∂2 J (λr) = − + − + ··· . so m = 0. J0 (λr). 0= 2 + r ∂r ∂r r (3.76) Substituting the Bessel function and its derivatives into Eq. Here we sketch the approach for cylindrical coordinates. 3. 2 (3.71 and some algebra   m2 ∂ 2 R 1 ∂R 2 + λ − 2 R. Anticipating separation in the coordinates r. As we did for our problem in the Cartesian coordinate system. The diffusion equation in cylindrical coordinates is  2  ∂ 2c 1 ∂ 2c ∂ c 1 ∂c ∂c + + =D + . t).71) ∂t r ∂r ∂r2 r2 ∂φ 2 ∂z2 where the concentration is c(r. (3. scaled so their oscillations go to zero precisely at r = 0.73) Equation 3.73. z. A Bessel function is a smooth function with oscillations that swing through zero. Just like the case for a Fourier expansion. • assigning a different time dependence to each term in the expansion. we make a sum of Bessel functions.72) After substitution into Eq. t) =  αi − Ai e αi2 Dt a2 J0 α r i a 16:48:47 . 2 2 2 4 2 4 6 λ2 r λ4 r 3 λ6 r 5 ∂ J0 (λr) = − 1 + 2 1 − 2 2 1 + · · · . we ignore the z-dependence by assuming an infinitely long bar along zˆ.73 is Bessel’s equation. This is known as a “Fourier–Bessel expansion. the initial concentration is c . (3. One Bessel function cannot reproduce our initial condition with a sharp cutoff of concentration at r = a. φ. φ. 3. we write the product of three functions c(r. φ. The solutions are Bessel functions of zero order. and its derivatives are λ2 r2 λ4 r 4 λ6 r 6 + 2 2 − 2 2 2 + ··· . which shows similarities and differences to the Fourier series approach for the Cartesian system. t) = e−Dλ t R(r) eimφ . 0 ∂r2 21 22 41 22 42 61 J0 (λr) = 1 − (3.” c(r. We pick a simple case where the concentration profile is uniform inside the bar. based on the constants of separation.75) (3. t.

65. A2 = α2 J1 (α2 ) . . A Bessel function with the argument αi r/a will therefore be zero at r = a. 3. 3. Although there are sign oscillations in Ai from the values of J1 (αi ). we set the left-hand side of Eq..79 . . all the Bessel functions in the series make positive contributions to the concentration just below r = a. we need the orthogonality relationship 1 r J0 α r i a J0 α r j a dr = 0 unless i = j .84) The first terms in the Fourier–Bessel expansion for c(r. αi (3. but the solution has no meaning beyond r = a. t) are shown in Fig. 5. corresponding to where the Bessel function crosses zero.82) 0 1 0 where 1 xJ1 (x) dx = 0 (which also equals −dJ0 (x)/dx).. These expressions are also needed: 1 xJ0 (αi x) dx = 1 J1 (αi ) . The rates of decay increase as the square of αi . 16:48:47 ..78) 0 where {αi }. multiply both sides by J0 ( αa1 r ).77 equal to c . 2 2 4 2 4 6 (3.83) (3. 8.}. are {2. 11.5 for t = 0. and 74. Here the integrand is rJ02 ( αa1 r ). The coefficients of the Fourier–Bessel expansion are 2c . 2 1 (3.8 for i = 1.81) x5 x x3 − 2 + 2 2 − . 2. The values of αi2 are 5. a a a α i (3. except when αi = α1 . 3. set the exponential equal to 1. the roots of J0 . 30. The αi are special numbers – the roots of the Bessel function.405. but are rescaled in height and width. and integrate  a c 0 a α r α r α r   1 1 i r J0 Ai r J0 dr = J0 dr .79) 0 The right-hand side of Eq.. so the higher-order curves decay much faster than low-order ones.80) xJ02 (αi x) dx = 1 2 J (αi ) .52. To match the initial profile at t = 0.5. All have the same characteristic shape of J0 . 3. This sharp edge continues through to r > a.Diffusion 68 where we set different values of λ = αi /a. α1 J1 (α1 ) 2c . (3. To fit the initial concentration profile at r < a.79 is zero for all terms in the sum.78. . A1 = (3.

3. What are the conditions for which this steady state of equilibrium is the state of thermodynamic equilibrium? The master equation is readily formulated for one site i surrounded by z neighboring sites with labels j. In thermodynamic equilibrium: ≈ ci = e−εi /kT . We first consider what the master equation predicts about equilibrium in a system. and the second term is the loss of concentration on site i from the reverse processes. 3.5 1.0 t Fig. labeled with characteristic decay behaviors. 2. The concentration on site i changes with time   dci Wij cj − Wji ci .5 2 –1.85) where the matrix W is a set of transition rates.0 Radius (a) 0.5 3 0.5 Kinetic master equation and equilibrium The kinetic master equation works with state variables of a system.5 Kinetic master equation and equilibrium 69 1 1.0 –0. To answer our question about equilibrium.e.86) 16:48:47 . 3.3.0 α32 Dt e– a2 0. i.5 2.5 The first three functions in the Fourier–Bessel expansion (i = 1. where the ensembleaveraged concentrations of atoms on different sites are monitored in time as the atoms move from one site to another. The first term in Eq.5 α22 Dt e a2 – Solution valid 1.0 0. and how they change with time.5 e– α12 Dt a2 Concentration (c') 1.85 is the gain in ≈ concentration on site i as atoms move to it from the neighboring j. when none of the state variables are changing with time. =+ dt z z j j (3. It is readily understood for atoms on an Ising lattice. we first assume thermodynamic equilibrium and then find what conditions are imposed on W ..0 2. 3) for a uniform concentration in a cylindrical bar of radius a. Z (3.

93) +εi /kT =e . dt Z Z z z j j (3.89) The condition for kinetic equilibrium is that there are no changes in atom concentrations on the lattice sites.85  e−εj /kT  e−εi /kT dci Wij Wji =+ − . 3.94) j z=z.90) j Wji e−εi /kT . The second solution is more interesting. 3.. all proportional to Boltzmann factors. (3. ci e i 16:48:47 (3. (3.97 can be understood with reference to Fig. (3. (3.93 are consistent with reaching thermodynamic equilibrium z  e+εj /kT e−εj /kT = j z  e+εi /kT e−εi /kT . (3. 3. Z (3. dci /dt = 0 0=+ z  −εj /kT Wij e − j z  −εj /kT Wij e j = z  z  Wji e−εi /kT . so the rates of Eqs. Substituting into the kinetic master equation Eq. For the individual Wij we pick exponentials with positive arguments that cancel the Boltzmann factors Wij = e+εj /kT .93 become Wij = e−(E ∗ −ε )/kT j −(E∗ −εi )/kT Wji = e .e.93 with a constant.6. This solution eliminates all kinetics. Substituting into Eq. W = 0 will ≈ ≈ work.88) j where εj is the energy associated with the atom on site j. Consider equilibrium concentrations on the three states shown in the figure. we can find two conditions on W that satisfy Eq.96) .92 and 3.95) We can also multiply both rates of Eqs.87) Z ≡ e−εi /kT + z  e−εj /kT . The ratio of the equilibrium concentrations between the activated state and the initial state is ∗ e−E /kT c∗ ∗ = −ε /kT = e−(E −εi )/kT .96 and 3.91.98) . By choosing the constant factor e−E /kT . 3. 3. (3. so atoms stay on their sites with their thermodynamic probabilities. First. (3. and achieve the ∗ same condition for steady state (dci /dt = 0).92) Wji (3.91 gives equality. (3.91) j By inspection. i.92 and 3. 3. 3. Eqs.97) Equations 3.92 and 3.Diffusion 70 cj = e−εj /kT .

they are in thermodynamic equilibrium with each other. which equals Wji of Eq.6 Linear kinetic response 71 E* Wji Wij εi t Fig. it will continue to the other side. To be specific. there is no net effect on the equilibrium concentrations.101) The second derivative evaluated at L0 is a positive constant. but it requires that the system is not far from equilibrium.6 Linear kinetic response Here we develop a simple and intuitive model of how a system moves towards equilibrium.3.7 This is the basis for activated state rate theory.103) 7 If it returns to state i with some probability. F.102) (3. The force. F(L0 + L) − F(L0 ) ≡ F 2 dL2 L0 1 F A(L)2 . which was relabeled as A.99) dL L0 2 dL2 L0 Because F is a minimum at equilibrium. but has the same return probability for jumps out of the state j. 3. 3. 3.6 εj Energies and transitions between states i and j through an activated state of energy E ∗ . or at least this is the reason why it leads to thermodynamic equilibrium. Finally.100) (3. is obtained in the usual way dF d 1 =− A(L)2 dL dL 2 F = −A L . This treatment is a handy one that can be applied to a number of nonequilibrium behaviors. The free energy is approximated with a few terms in a Taylor series near the equilibrium value of a state variable. 2 (3. F =− (3. choose an order parameter with an equilibrium value of L0     dF 1 d2 F F(L0 + L) = F(L0 ) + L + (L)2 + · · · (3. We expect F to act as a potential energy. so it can be used to obtain a “force” that pushes the system towards equilibrium. and   1 d2 F (L)2 . (dF/dL)L0 = 0. 16:48:47 . When the two states i and j are in thermodynamic equilibrium with an activated state. we assume that once an atom gets into the activated state.97.

as expected for the unmixing problem at low temperatures and intermediate compositions (top curve in Fig. 3.3 Solve the diffusion equation for a long rectangular bar. c(x. c(x. In reality.7 microns at 1. When there are multiple order parameters that alter the free energy. We assume that the rate of change of an order parameter is proportional to this force. 3. however. but with its cross-section centered about the origin. but the boundary conditions are now c(−Lx /2. much as was done in Sect. Equation 3. their mobility coefficients may differ considerably. t) = 0 . (b) Explain why it is physically reasonable that the first and second spatial derivatives of c(x. and 2. (3.72 Diffusion The next assumption is perhaps the key one.104) (3. Ly /2. −Ly /2. but probably not much for small changes in T and L.000 ◦ C. and when the kinetic mechanism allows for a constant mobility. when d2 F/dL2 < 0. and 3. the mobility may change with temperature and state of order. 3.107) . We expect Eq. y. 16:48:47 (3. M. t) are odd and even functions of x. Problems 3.105 was obtained with the expectation of a decaying exponential. Here the system is stable against composition fluctuations. Eq. t) = 0 . the system is unstable.9b).105 to be useful if the system is not far from equilibrium.1 (a) For diffusion in one dimension. t) = 0 . t) = 0 . and this is true when d2 F/dL2 > 0. The characteristic diffusion length at 900 ◦ C was found to be 1. 2. y.27. 3.106) and the diffusion distances are precisely x = 4Dt.13. 3. so our assumption of being close to equilibrium is inconsistent with the behavior of the system. F dL = MF = −MA L . (b) Suppose the diffusion coefficient is of the form D(T) = D0 e−E ∗ /k BT √ . (3.104 has a simple solution L(t) = L(t = 0) e−MA t .” Equation 3. c(Lx /2. Equation 3.2 Suppose two diffusion profiles have been measured in two identical samples heated for the same time of 1000 s. 3. The initial condition is still a uniform concentration of c .29 satisfy the diffusion equation. On the other hand. (a) Calculate the activation energy E∗ for this diffusion process. dt where M is the “mobility. show by substitution that the partial derivatives in Eqs.105 predicts that both positive and negative deviations from the equilibrium L0 will be restored with the same characteristic time. What is D0 ? 3. and the rate will also scale with the size of the deviation from equilibrium.3. but at two different temperatures.28.0 micron.105) where A = d2 F/dL2 > 0 for stable systems.

Graph the approximate radial concentration profile in the cylinder for the times t = 0 (easy). and t = 0. d3 F/dL3 = 0. what is the concentration profile of hydrogen at different r through the thick wall of the tube? Graph it. 3. Is this system in equilibrium? Explain how L will change with time. with the initial and boundary conditions of Sect. The concentration of hydrogen at r = 0 is always c. all near the point where dF/dL = 0. Analyze them with an appropriate form of Eq.73 Problems 3. 3. and d4 F/dL4 > 0. so the free energy was near a minimum. and state B is in equilibrium with state C. and compare this to the case of part a. d3 F/dL3 = 0. Is this system in equilibrium? How will L change with time? (b) Suppose d2 F/dL2 = 0 and d3 F/dL3 > 0. 3. 3. t = 0.4 Consider the diffusion problem in cylindrical coordinates.5 Consider the diffusion of hydrogen through a thick-walled tube having an inner diameter that is negligibly small compared with the outer radius at r = a. both positive and negative. and d4 F/dL4 < 0. t = 0. 3. Consider some variations of this situation.5.10. where t is in units of a2 /D. (a) Suppose d2 F/dL2 < 0. Is this system in equilibrium? How will L change with time? (c) Suppose d2 F/dL2 = 0.7 In Sect. 16:48:47 .6 Use activated state rate theory for transition rates between states (through an activated state E∗ ).4.6 the system was assumed to be near equilibrium.05. and the concentration is zero at r = a as it escapes into the atmosphere. After a long time.99 and small values of L. and use the kinetic master equation to follow the probabilities of states to show: If state A is in equilibrium with state B. then state A is in equilibrium with state C. Is this system in equilibrium? How will L change with time? (d) Suppose d2 F/dL2 = 0. 3.

• Because the new phase and the parent phase have different structures.4. there must be an interface between them. but two key concepts can be appreciated immediately. for example. The atom bonding across this interface is not optimal. because a larger fraction of its atoms are at the interface. An issue for nucleation with chemical unmixing is the time required for diffusion of the different chemical species of atoms.4 Nucleation As discussed in Sect. The addition of atoms to embryos is largely a kinetic phenomenon. We will seek a steady-state rate of forming “critical” nuclei that can grow. this crystal will grow. This surface energy is most significant when the new phase is small. The discontinuous case begins with the appearance of a small but distinct volume of material having a structure and composition that differ from those of the parent phase. 1.1 A discontinuous transition can be forced by symmetry. • For nucleation of a new phase within a solid. phase transformations can occur continuously or discontinuously.” If the nucleation event is successful. The chapter ends with a discussion of the transient time after a quench from high temperature when the distribution of solute relaxes towards the equilibrium distribution for steady-state nucleation. The mismatch creates an elastic field that costs energy.5. The new crystal must appear in miniature in the liquid. The process of nucleation is an early step for most phase transformations in materials. the new phase would take this local atomic structure. 17:28:32 .2. It has many variations. a second issue arises when the new phase differs in shape or specific volume from the parent phase. a process called “nucleation. Surface energy plays a key role in nucleation. and this time can be long if atoms must move long distances between incipient nuclei (sometimes called “embryos”). since the surrounding atoms can flow out of the way. 14. although the tendency of atoms to remain on the embryos is a thermodynamic one. 1 The nucleus may or may not have the structure and composition of the final phase because the transformation may occur in stages.2 so the interfacial energy must be positive. There is no continuous way to rearrange the atoms of a liquid into a crystal. as formalized for some cases in Sect. This is not a concern for nucleation in a liquid or gas. 2 If the structure of the interfacial atoms and bonds were favorable.

the coherent nucleus in Fig. Suppose the alloy is quenched (cooled quickly) to a temperature such as 0. In solid→solid transformations. For the nucleation of solid phases. 4. The precipitating β-phase has a crystal structure different from the parent α-phase. plus some B-rich α  -phase.2a illustrates an incoherent nucleus. the B-rich α  -phase may nucleate as small zones or “precipitates” in an A-rich matrix. heterogeneous nucleation is more common than homogeneous.1 Terminology and issues 75 4. the parent phase is “supersaturated” immediately after the quench. the underlying crystal lattice remains the same while the solute atoms coalesce. and there is little registry of atoms across the interface. dislocation lines. i.3 In this case. Here the precipitation of β-phase in an alloy cooled rapidly from the α-phase requires both a redistribution of chemical elements and a different crystal structure. without preference for location. For some compositions. heterogeneous nucleation occurs on grain boundaries. A different case is shown with Fig. In contrast. the wall of the container is a common heterogeneous site.4zV/kB . Figure 4. Consider an A-rich A–B alloy having the α-phase at high temperature.1 for the unmixing of solute in a eutectic alloy. “Heterogeneous” nucleation refers to the formation of nuclei at specific sites. or other defects or heterogeneities.10. stacking faults. 2. In both these examples of nucleation. When freezing a liquid. 17:28:32 . 4. 4. “Homogeneous” nucleation occurs when nuclei form randomly throughout the bulk material. 3 A “matrix” is an environment in which a new phase develops.e. where the equilibrium state would have mostly A-rich α  -phase.2b has crystal planes that match up with Temperature (T ) L α α+L L+β β α+β t Fig. as shown in the unmixing phase diagram of Fig.4.1 A B Composition (c) Binary phase diagram depicting a quench path from a temperature with pure α-phase to a temperature where some β-phase will nucleate. The precipitate phase can be “coherent” or “incoherent” with the surrounding matrix. and is unstable against forming the new phase.1 Terminology and issues Nucleation can occur without a change in crystal structure.

but the behavior can be modified significantly by elastic energy. planes in the α-matrix. and (c) a semicoherent precipitate interface. 4. The surface energy plays a key role for both effects. 1. there have been no reported instances of homogeneous. Homogeneous. perhaps distorted crystals. including the case of Co-rich precipitates in the Cu–Co system shown in Fig. An important feature of the nucleation of a solid phase in a solid is the elastic distortion both inside and outside a precipitate.” With a greater mismatch of lattice parameters. incoherent nucleation is often stated to be the most common form of precipitation reactions in solids. (b) a coherent precipitate. Some of these will have similarities to crystals. 4. 4. coherent nucleation is well documented in some systems. To help relieve the elastic strain.2 Critical nucleus In this chapter we calculate a nucleation rate by considering the probability of finding a nucleus of a particular size. The elastic energy usually has a major effect on the thermodynamics of nucleation. however.2c. However. 4. incoherent nucleation in any crystalline solid as depicted in Fig. In the earliest stages when the precipitate is small and has a high surface-to-volume ratio. the spacing between dislocations decreases and the interfacial energy increases. 4.2.2a. Incoherent interfaces may form later during precipitate growth.6. The elastic energy increases as the precipitate grows in volume. and these crystal-like fluctuations are expected to 17:28:32 .1 Fluctuations In a liquid there are continuous rearrangements of atoms. and by considering the rate at which it grows. and may go undetected. however. and some reduction in elastic energy is often possible when the interface loses coherency. the very early stages of nucleation are more likely to exhibit coherency. an array of edge dislocations may form on the boundary as shown in Fig. To date. however. and transient appearances and disappearances of different atomic arrangements. Heterogeneous. but its importance changes with the size of the precipitate. An interface that is coherent except for a periodic sequence of dislocations is “semicoherent.2 a Incoherent b Coherent c Semicoherent Schematic representation of (a) an incoherent precipitate.Nucleation 76 t Fig. the precipitate may be coherent with good atom matching at its interface to the surrounding matrix.

and the particle is shaped to maximize these surfaces. but will have a lower surface energy. in particular a surface energy. The fluctuation spectrum in the liquid is appropriate for a high temperature.) The change in Gibbs free energy for nucleus formation is G = V GV + A σ . and eventually a steady-state distribution of local structures is achieved. this is considered later.” a steady-state distribution is attained where some local regions are large enough to continue growing as crystals by adding more atoms to their surfaces. will suppress nucleation because the crystal-like fluctuations are small in size. Crystalline regions are unstable above Tm .77 4. After a time delay. so crystal-like fluctuations may be larger in size and may last longer in time. 4. This is a positive quantity. the partition function normalizes the probability. 17:28:32 . the area of interface between the two phases times the specific surface energy. and surface energy affects a larger fraction of atoms in small particles. the difference in free energy of the liquid and solid is infinitesimally small.2 Critical nucleus become more frequent as the liquid is cooled towards the melting temperature. The local configurations in the undercooled liquid evolve over time.2. but is considerably out of equilibrium at the new low temperature.4 The equilibrium shape that minimizes the surface area-to-volume ratio is a sphere of radius R 4 This is not expected for crystalline solids. Assume the surface energy is isotropic. 1. 6. an energy penalty.2 Surface energy Consider the formation of a nucleus as a fluctuation that rearranges atoms in a small volume of the parent phase. Infinitesimally below Tm . The probability. Now consider a more extreme case where the liquid is quenched rapidly to a temperature well below Tm . (There is also a penalty in elastic energy from any size mismatch of the precipitate particle in a solid matrix. Tm . however. As the temperature falls below Tm .2) where GV is the Gibbs free energy per unit volume of forming the new phase from the parent phase. Usually low-index planes such as {110} or {111} have the lowest surface energy. Determining this shape requires the “Wulff construction” discussed in Sect.e. The slightest energy penalty. but at temperatures slightly above Tm their penalty in free energy is expected to be small by Eq.) The G includes a term Aσ .6. i. The particle will have a larger surface area than a sphere. called an “incubation time. so σ is independent of crystallographic direction. growth of crystalline regions is expected. of course. (4. and Z is a partition function comprising the sum of Gibbs factors for all other possible configurations.. (As usual. that a group of atoms is configured as a nucleus is   exp − kG BT P(T) = . The phase transformation is then well underway.6. (4. P.1) Z where G is the change in free energy when the group of atoms is configured as the new phase (compared with the parent phase). and V is the volume of new phase.

and critical free energy of formation. The surface energy is dominant at small R.Nucleation 78 4 3 (4. which is found by differentiation of Eq.3 shows that for R > R∗ .4) ∗ dR R=R GV and by substitution into Eq. G∗ are labeled for this case of GV = 1 and σ = 1. 3 (GV )2 (4. 4.3 shows the two terms of Eq. This trend reflects the relations: R2 > R3 when R → 0. and the vertical has units of σ 3 /GV2 .3 0 1 2 R 3 4 Graph of G(R) = 4π R2 − 4/3πR3 .3 dG 2σ = 0 ⇒ R∗ = − . but the volume term is dominant for large sizes of the spherical particle. G is never negative. (4.3 G∗ ≡ G(R∗ ) ⇒ G∗ = 16π σ 3 .3. The curve G(R) has its maximum value. G = 4. nucleation of the new phase is possible for large R when the positive surface energy is less than the negative volume free energy.2. and R3 > R2 when R → ∞. The critical radius.3) π R GV + 4π R2 σ . G∗ . so nucleation cannot occur. for sizes smaller than R∗ . R∗ . where the negative volume energy (varying as R3 ) counteracts the positive surface contribution (varying as R2 ).5) Approximately. particle growth causes G(R) to decrease. (Equivalently. 4. 3 When GV > 0. the critical radius R∗ separates two types of behavior. 4. 4. the 40 4πR2 ΔG(R) 20 ΔG* 0 R* –20 – 4/3πR3 –40 t Fig.3 Critical radius Figure 4. at R∗ . On the other hand. the particle is more likely to grow than shrink – Fig. the horizontal axis has units of −σ/GV . When GV < 0.) 17:28:32 . 4. with individual terms as dashed lines. For sizes greater than R∗ . It is drawn for GV < 0.

3 Heterogeneous nucleation Homogeneous nucleation in solid materials is not rate. as shown by the dashed line in Fig. a new feature of nucleation is the elimination of some of the grain boundary area. The surface area of a double spherical cap having a contact angle θ (see Fig. which is replaced by the new particle. • The precipitate widens to eliminate the surface energy of the grain boundary. shape. The last term on the right-hand side contains the grain boundary surface energy. and free energy of forming a critical nucleus are changed in the presence of a defect. 4. 4. so a nucleus of radius R∗ is in a “metastable equilibrium” with respect to the parent phase.6) − π R2 sin2 θ σgb . 16. and has a negative sign. or two spherical caps that make contact at the grain boundary. the new shape of the nucleus accommodates these characteristics: • In contact with the parent phase. heterogeneous nucleation is preferred energetically. and this is the chemical potential for atoms both in and near the critical nucleus because they are in equilibrium. but the usual type of nucleation is heterogeneous. a circular area of preexisting α–α interface is eliminated.3  2π  G = 2 − 3 cos θ + cos3 θ R3 GV + 4π (1 − cos θ )R2 σ 3 (4.79 4. For heterogeneous nucleation on a planar interface such as a grain boundary. • The precipitate is shaped as a double convex lens. 4. At R∗ the variation of free energy with respect to R is zero. the precipitate surface is spherical because the local atom structure is unaware of the presence of the grain boundary.” and is described in more detail in Sect. 1. A chemical potential can be defined as in Eq. where R is the (constant) radius of curvature. where the second phase nucleates at dislocations or at grain boundaries. 4. Critical quantities can again be found by taking the derivative of G with respect to 17:28:32 . The critical radius R∗ and the critical free energy of formation G∗ are key quantities for nucleation. and its volume is 2π(2 − 3 cos θ + cos3 θ)R3 /3. Since this decrease is not realized for a nucleus forming in the bulk.24. This effect of surface area on chemical potential is known as the “Gibbs–Thomson effect. The cost in free energy to add an atom to this region is higher when the nucleus has a high surface-to-volume ratio. The size. A sphere is a natural shape for homogeneous nucleation because it minimizes an (isotropic) surface energy for a given volume. the free energy of formation as a function of R can be calculated in the same way as Eq.4. for example. σgb .3 Heterogeneous nucleation particle is more likely to shrink than grow because shrinkage causes G(R) to decrease.3. The energy saving of πR2 sin2 θ σgb decreases the free energy of formation. When a grain boundary nucleus is formed.4) is 4π (1 − cos θ )R2 . On a grain boundary. With these expressions.

Notice that there are more lead atoms on the top side of the precipitate that has good atomic registry with the aluminum matrix. It is interesting that the value of R∗ is identical to that derived in the homogeneous case R∗ = − 2σ . G∗ = G∗hom 2 G∗ = (4. more grain boundary area can be eliminated per unit volume of precipitate if nucleation occurs at an edge.4 σ A grain boundary β-phase nucleus as a double spherical cap.5 shows such a faceted nucleus. this network also offers edges and corners for nucleation. The bottom of the crystal is less well oriented for such matching. 4. The dashed line shows the region of pre-existing grain boundary that is eliminated by the nucleus. so the free energy of formation of a heterogeneous grain boundary nucleus is less than that of a homogeneous nucleus. and the particle minimizes its surface area against the lower aluminum crystal by making a spherical surface with a large radius of curvature. Figure 4. called a “grain boundary allotriomorph. the radius of curvature and setting the result equal to zero.8) Here G∗hom is the critical free energy of formation for homogeneous nucleation. In general.4 was drawn assuming an isotropic surface energy for the α–β interface.” The image shows a precipitate of lead at the grain boundary between two aluminum crystals (the Al–Al grain boundary is about a quarter of the way up from the bottom of the image). and even more when nucleation 17:28:32 . Section 6.Nucleation 80 σ σgb α θ β α t Fig.6 gives more discussion of surface energy and grain boundary allotriomorphs. The elimination of grain boundary energy made it attractive for the lead crystal to form at a grain boundary between two aluminum crystals in Fig. Besides simple interfaces between two crystals.7) whereas the critical free energy of formation becomes 16π σ 3 2 − 3 cos θ + cos3 θ 2 3(GV )2 2 − 3 cos θ + cos3 θ . 4. GV (4. The checked pattern at the top of the lead particle indicates regular atom matching across low-index crystal planes between the lead and aluminum. Grain boundaries make a three-dimensional network through a polycrystalline material. The contact angle θ must satisfy a balance of surface energy forces depicted by the three vectors of lengths σ and σgb . The term (2 − 3 cos θ + cos3 θ )/2 can range from zero to one. Figure 4. Lee and Aaronson [20] have investigated the grain boundary nucleation problem and concluded that the double spherical cap morphology is unlikely. A faceted nucleus is generally expected. even though homogeneous nucleation did occur in this material – notice the small lead crystal near the top of the image.5.

and a second lead crystal above it.1 discusses how there is an increased fraction of edge and corner sites in nanomaterials. 4. The energy and entropy can be defined for this unstable equilibrium. so we can depict the effects of surface area described in Sect.5 4. The elastic energy.5 A high-resolution transmission electron micrograph of a lead crystal at the grain boundary of two aluminum crystals. 17:28:32 . however. t is at a corner of three or more crystals. For solid precipitates that form in a solid parent phase. any size or shape misfit causes elastic strains in both the precipitate and the surrounding matrix.1 Free energy curves The critical nucleus is in a state of equilibrium with the condition dG/dR = 0 at R∗ .4. There are fewer edge and corner sites available for nucleation than simple sites on grain boundaries. These edge and corner sites are especially favorable for nucleation when the specific grain boundary energy is much higher than the interfacial energy between the precipitate and the matrix.81 4.4 Free energy curves and elastic energy Fig. 4.2. 4. Figure courtesy of Uli Dahmen (see also [21]). which is 5 Section 16. We take this opportunity to add elastic energy to the problem. The modification (to be described with Fig.6a) is a straightforward addition of surface energy to the free energy curve of the precipitate.2 by modifying F(c) curves.4 Free energy curves and elastic energy 4.

5. For the case shown in Fig. 4. Fig. surface energy effects are often called the Gibbs–Thomson effect of Sect.6a shift with temperature. with the generic label PdV. 4. precipitates may nucleate at locations that reduce the strain energy. giving a shift of c to higher composition. 4. but both contributions suppress the formation of the β-phase.4 0. the α-phase occupies a wider range of composition. the parent phase is assumed free of internal stresses.6 0. If there are pre-existing internal stresses in the parent phase.0 1. The surface area and the misfit strains alter the local free energies of both the β-phase and the α-phase around it. raising the nucleation barrier and suppressing nucleation. Compared with the construction for the bulk free energy curves. so precipitate formation generates strain energy.3. (Both P and dV are expected to scale with the fractional misfit of the precipitate. These effects are often modest. the analysis can be made quantitative. 4. (b) Opportunity for an intermediate phase to form in the case shown in panel a.2 0. and the β-phase must be even further enriched in B-atoms. Suppose a nucleus of β-phase precipitates from a matrix of α-phase.6a shows how the common tangent line intersects the free energy curve of the α-phase at a different composition.) It is important to remember that G∗ depends on undercooling through GV (T). so the vertical shift of the curve Fβ (c) may not be quantitative. Although the γ -phase would not be expected in equilibrium if all phases were large and 6 Without the precipitate.8 1.0 Concentration (c) b (a) Effect of surface energy and misfit elastic strains on the relative free energy of a β-phase nucleus.6b.0 0.0 t Fig.6a.8 0.6 a 0. It should also be pointed out that thermal fluctuations give nuclei of various sizes.6 adds to the free energy of nucleation. 17:28:32 . 4.4 0. 4.6 Concentration (c) 0. or a smaller shift may be appropriate if the waiting time is longer. 16. For nanoparticles of a fixed size. Also added to the free energy of the βphase is an elastic contribution. always positive. the free energy of the β-phase alone is therefore elevated by the amount G∗ = 16π σ 3 /3(GV )2 of Eq. This is physically reasonable – with more free energy needed to nucleate the β-phase. although they can be responsible for chemical composition shifts of phase boundaries for nanoparticles. 7 For nanoparticles.Nucleation 82 β β ΔG* + PdV γ Free energy Free energy α α Δc 0. and this must be taken into account in an analysis of how the curves in Fig.7 Another possible effect of these energy costs of nucleation is shown in Fig. however.2 0.

are examples of such intermediate phases (if indeed they can be called phases). Elastic energy favors precipitates as flat plates (Sections 6. where T = Tc − T. evidence of strain energy dominance as discussed below.2. This is the same approximation used for Eq. 6. for an ellipsoidal precipitate. and has no internal elastic energy. 4.5. which is δ = δv/3v G =  4 3 πR GV + 6μδ 2 E(y/R) + 4π R2 σ . 6. the shapes of the γ -phase particles are flat plates. for example.2 discuss the reasons).6b can be used to explain why undercooling is required for nucleation of the β-phase. described in Sect. Etot .3 gives the free energy change for forming the nucleus from the parent phase G = 2 (δv)2 4 3 πR GV + μ E(y/R) + 4π R2 σ . We pick the simple result of Eq. 1. Often the first stages of an unmixing transformation involve the formation of solute-rich planes of atoms.5.10) Assume the change in volume free energy GV is linear in the undercooling T below the critical temperature Tc .11) In fact. so GV 8 −L T . 4. Flat precipitates have a high surface area per unit volume. so the actual shape is set by a mix of different effects. and ranges from 0 for flat plates to 1 for a sphere. 4. 3 3 v (4. The precipitate is parameterized by its fractional volume change with respect to the original configuration of its atoms in the parent phase.1.1. shown in Fig. δv/v. this γ -phase is expected if the β-phase particles are small. 17:28:32 .5 and 21. Adding the elastic energy to Eq. we need a result from Sect. 11.4.4. 3 (4.6. Many times when this occurs. 11. if the γ -phase were the liquid. and the energy scales with the shear modulus. 1. and may be easier to nucleate.4 Free energy curves and elastic energy 83 without stresses. Fig.2 Effects of elastic energy on nucleation To adapt our analysis of nucleation to include the misfit stresses in and around the precipitate. a nucleus of γ -phase may have a smaller misfit than the β-phase. discussed in Sect. μ. An additional factor E(y/R) is a dimensionless number that accounts for the aspect ratio of the ellipsoidal precipitate. such as Guinier–Preston zones.9) For an approximately spherical precipitate. Guinier–Preston zones. the ratio δv/v can be expressed in terms of the relative change in radius δ ≡ δr/r. This change in specific volume generates shear strains in the surrounding matrix.3. based on the assumption that the precipitate itself is incompressible.8 In cases where elastic energy dominates the problem. of course.120 for the total elastic energy. This is an approximate result. Tc (4.

and now 6μδ 2 E(y/R) = 0.3. is δT − Eel 6μδ 2 E(y/R) Tc = − Tc . (Equation 4.Nucleation 84 40 4pR2 +0. the critical radius. with individual terms as dashed lines. positive strain energy. or are shaped as lenses. and critical free energy of formation.4/3π R3 . 17:28:32 . so flat precipitates hint at the importance of elastic strains in the matrix. but not when constrained by the matrix. but of course a suppressed transformation would not be observed. There are degrees of freedom in the shape and orientation of a precipitate. Because the elastic energy has the opposite sign as GV . Of course a flat precipitate shape is not favored by surface energy because its surface-to-volume ratio is high. i.. 6μδ 2 E(y/R) ≥ −GV . Figure 4. Both the nucleation barrier and the critical radius are increased. G∗ . L L (4.e. There are in fact cases where elastic strains in the matrix suppress phase transformations by hundreds of degrees K. If homogeneous precipitates are observed as flat plates. Complete suppression seems to be rare. 4. the reason is often that these shapes and orientations minimize the elastic energy. a larger undercooling is required for nucleation.7 –4/3pR3 0 1 2 3 4 R Graph of G(R) = 4π R2 − 4/3πR3 + 0. δT.) What this would mean is that the phase transformation could occur when the transforming particle is in free space.) where L is the latent heat of the phase transformation. GV = 1. R∗ .7 shows how the positive elastic energy counteracts the negative chemical free energy to make nucleation more difficult.4/3πR3 ∆G(R) 20 0 –20 –40 t Fig.12 predicts such a suppression to T = 0 if 6μδ 2 E(y/R) ≥ L.1. and precipitate shapes or “morphologies” having more equal dimensions (termed “equiaxed”) are suppressed by elastic energy. σ = 1. With this additional. which would be the case if the elastic energy per unit volume of precipitate were greater than or equal to the chemical free energy per unit volume. (As in Fig. and E(y/R) approaches zero in the case of flat plates.12) where Eel and L are per unit volume of the precipitate. 4. shift as shown by the arrow. The shift in the characteristic undercooling for nucleation. It is interesting to ask whether elastic energy can suppress a phase transformation entirely.

and critical free energy of formation. 4. and Zeldovich [26]. but it does not predict an actual nucleation rate. 4. σ s1 = 0. Figure 4. with individual terms as dashed lines. It provides key insights into the nucleation process. are labeled. the critical nucleus size n∗ . two derived parameters have been useful.75.4.8 shows Gn for parameters in correspondence to those of Fig.188 atoms). For atomistic kinetics it is more convenient to work with the number of atoms in the cluster. The critical cluster size.8 10 n* 20 n 30 40 × 103 Graph of Gn = GV n + σ s1 n2/3 . 4.5 The nucleation rate The thermodynamic concept of the critical nucleus of Sect. The curve is skewed to the right.3 is Gn = GV  n + σ s1 n2/3 . Becker and Döring [25].3. and the critical free energy of formation G∗ . The equivalent to Eq. 4.002.13) where  is the atomic volume. 4. These give the σ s1n2/3 40 ΔG(n) 20 ΔG* 0 –20 ΔGVWn –40 0 t Fig.07677.3. 752. In the 1920s and 1940s. The rest of this chapter presents some of these kinetic analyses. Conversion gives GV  = 0. 4. Farkas [24].8. instead of the radius R. since the addition of an atom to a large cluster gives a small contribution to the radius. and ends with an introduction to more modern concepts of time-dependent nucleation.5. n∗ = 16. 4. To describe the curve G(n) of Fig.5 The nucleation rate 85 4. 17:28:32 . the Gibbs droplet picture of nucleation was extended to include kinetics by Volmer and Weber [23].2 is due to Gibbs [22]. (4. This graph corresponds to that of Fig.1 Clusters of atoms All our information about the G for nucleation is parameterized by the size of the nucleus. G∗ = 16. n. with R = 1 corresponding to a 10-atom radius of a simple cubic crystal (4. and s1 is the atomic surface area.

∗ δn + ∂n n 2 ∂n2 n∗ 1 ∂ 2 G G(n∗ ) − G(n∗ + δn) − (δn)2 . m 3 3 n m= .23) 9 In a hot vapor. that the curvature of this factor [m2/3 + (n − m)2/3 ] is negative 2 2 ∂2 [m2/3 + (n − m)2/3 ] = − m−4/3 − (n − m)−4/3 .16) −[GV  (n−m)+σ s1 (n−m)2/3 ]/kB T c0 e 2 −GV  n/kB T −σ s1 [m2/3 +(n−m)2/3 ]/kB T e c0 e . a factor in the exponent of Eq. when the kinetic energy of a cluster is comparable to the binding energy of an atom. 2 9 9 ∂m ∂2 [m2/3 + (n − m)2/3 ] < 0 .2 Atom additions and subtractions Consider the different ways a cluster can change its size. In a vapor especially. it is easy to imagine collisions that make a cluster of n atoms by combining a cluster of m atoms with a cluster of n − m atoms. the conservation of momentum may require interactions with a third cluster to carry momentum away from the collision. 2 (4. examine [m2/3 + (n − m)2/3 ]. by fluctuation processes. m) = c(m)c(n − m) .15) where the second term on the right-hand side of Eq. the negative curvature of G(n) at n∗ . 4.18) To find the maximum in R(n. we add a third parameter. To more completely specify the nucleation barrier. . m) = (4. m) depends on mobilities and sticking coefficients.9 The rate R(n. ∂m2 (4. but is also proportional to the product of concentrations. however. The Taylor expansion near n∗ is ∂G 1 ∂ 2 G (δn)2 + . m) is maximized. the rate R(n. 4.8). 4.18. The rates of cluster growth depend in part on how many clusters of each type are present to come together. −[GV  m+σ s1 m2/3 ]/kB T R(n.5.14 is zero because the function G(n) is a maximum at n∗ .20) (4. 4. combine with 1 atom to produce a cluster of size 6. an n = 2 cluster can attach to an n = 7 cluster creating a size n = 9.21) Note. . m). ∂m 2 −1/3 2 − (n − m)−1/3 = 0 . c(m) × c(n − m) R(n.14) (4. 2 ∂n2 n∗ G(n∗ + δn) = G(n∗ ) + (4. m) = c0 e R(n. .Nucleation 86 peak of G(n) (the point at the intersection of the vertical and horizontal dashed lines in Fig. . This factor is positive. etc. A cluster containing 5 atoms can. (4. We seek an extremum as ∂ [m2/3 + (n − m)2/3 ] = 0 .17) (4. 17:28:32 .22) (4. so if it is a minimum.19) (4.

(4.24) where Wn is a transition rate that accounts for how fast atoms depart from a cluster of size n. Nn Nn (4. The maximum values of R(n. (4. where a large supply of single-atom clusters (N1 N) contributes a flow of clusters from bin to bin of increasingly larger n. 4. m) are when m = 1 or m = n − 1. The flux Jn is the number of clusters per unit time that change from n to n + 1. We believe that cluster growth generally occurs by adding one atom at a time. perhaps with Eq. and Nn = 0 for n > 1.” {Nn }. 1. so N1 = N.24 with the equilibrium cluster eq numbers Nn  Jn Wn  eq = eq Nn − Nn+1 . however. In a material quenched from high temperature we assume that all clusters contain one atom. By analogy to Eq. et seq. The difference in W for a cluster of size n and n + 1 is assumed small.9.25) For the equilibrium distribution. the most rapid growth of clusters. we return to the kinetic master equation as illustrated in Fig. since these are the ones most important for nucleation. there are also kinetic reasons why single atom additions are faster.25 Jn = Wn [Nn − Nn+1 ] . Instead of moving atoms between adjacent bins. Assuming the clusters grow or shrink by one atom at a time (as discussed in the previous section).) eq Wn Nneq − Wn+1 Nn+1 = 0 . Very quickly. Our concern. It proves useful to normalize Eq. is for clusters with sizes around the critical n∗ .” 4. however. however. the net flux between adjacent cluster numbers must be zero (as can be shown with an approach like that of Eq.5 The nucleation rate Therefore the factor [m2/3 + (n − m)2/3 ] is a maximum at m = n/2. 4.90. Assume there is a cluster distribution in thermodynamic equilibrium that can be known. and eventually there will be clusters with a very large n that may be better described as crystals.9. In other words. so Wn is used for both. based on the types of clusters available to combine. we expect the formation of clusters with n = 2. n/2) is a minimum. In what follows we seek a steady-state condition over the time period of nucleation. which specifies the number of clusters containing n atoms. eq Nn+1 = eq Wn Nn Wn+1 17:28:32 . the movement of clusters occurs only between adjacent bins in Fig. Here Jn is positive if more clusters move from n to n + 1 than the reverse. as specified by the kinetic master equation.87 4.3 The Zeldovich factor To calculate a nucleation rate. Since single atoms tend to be most mobile. a process termed “monomer attachment. meaning that the rate R(n.5.1. we move clusters of atoms. but it is a good way to track the time evolution of the “cluster size distribution. This is not a physical process.26) (4. 1. 1. 3. occurs by adding one atom at a time. The flux of clusters depends on the numbers in the bin and their transition rates.27) .

so Nm 0 for large m: eq (B. N1 . Eq. It is the steady-state condition. This is generally quite complicated because a high flux into one cluster size can alter the fluxes to clusters of similar size at later times. Jss eq = Wn eq − eq N N1 Nm n=1 n (4. Assume the value of N1 has eq its equilibrium value. because small cluster sizes equilibrate most quickly: eq (B.Nucleation 88 and again assuming Wn = Wn+1 . (4. the values of Nn remain constant.28 gives the same cluster flux for all n. With equal flows in and out. Using these two boundary conditions to simplify Eq.30) eq − eq Nm−1 Nm and after numerous cancellations on the right-hand side   m−1  1 Nm N1 .28 for all n. (4. 17:28:32 . The next step is a clever one that combines the requirement of steady state with the boundary conditions. 2) Nm /Nm = 0.28) To track all cluster sizes {Nn }. 1) N1 /N1 = 1.. we expect few clusters of very large size. we need equations like Eq.31) Now we impose physically reasonable boundary conditions. Summing the Jn     m−1  Jn N2 N3 N1 N2 eq = Wn eq − eq + eq − eq N N1 N2 N2 N3 n=1 n   Nm Nm−1 + . .C. eq = Wn eq − eq Nn Nn Nn+1 (4. 1. 4. 4.9 equals the flow out of the bin to the next bin n + 1 in the chain. (4. and fortunately it is an important one. 4. 4..   Jn Nn+1 Nn . where the cluster distribution {Nn } does not change with time. What is important is that in steady state Jn = Jm ≡ Jss .C. There is one case where the problem is simple. By Eq. For steady state during the times of nucleation we need to assume an abundant supply of single atom clusters.28 directly.29) because this ensures that the net flow of clusters into a bin n in Fig. 4.29.31 Jss = Wn m−1  n=1 . The differences between Nn and Nn+1 are not expected to be large.10 On the other hand.32) 1 eq Nn 10 This is not true much later in the post-nucleation stage when there is a significant fraction of new phase. making it tricky to work with Eq.

34)  . 2π kB T ∂n2 n∗ 17:28:32 . Equation 4. 4. Because G∗ is the largest value of G. Its integral receives little contribution from the smallest n.5 The nucleation rate 89 The next step is to evaluate the denominator of Eq.32. called Z  .15 provides useful information about how Gn varies with n near n∗ ⎛ ⎞ ∂ 2 G 2   m−1 (δn) ∗  G ⎜ ∂n2 n∗ ⎟ (4.39) π .1  Z = m−1  n=1 Z = m−1  n=1  Z = 1 eq . and shift the center to zero   ∞   G∗  exp −an2 dn . 4. so we extend the limits. Nn m−1  n=1 (4.33) Z  n exp − G kB T .4. kB T 2kB T n=1 which is converted to a continuous variable Z  = Z exp   G∗ kB T ∞ ⎛ ⎜ exp⎝ ⎞ ∂ 2 G (δn)2 ∂n2 n∗ ⎟ 2kB T n=1 ⎠ dn .  Gn Z exp kB T (4. these clusters make the largest contribution to Z  . (4. Z = Z exp (4. using Eq.40) The complete expression of Eq.32 for the steady-state flux of clusters across the critical cluster size is ∗ Wn Z − G Jss = (4.42) .35) Clusters near the critical size n∗ are the most important for nucleation.37) The sign of ∂ 2 G/∂n2 is negative. a (4.41) e kB T . (4. Z where Z is the “Zeldovich factor” % 1 ∂ 2 G Z= − (4.38) kB T −∞ where a=− Recognizing a standard Gaussian integral ∂ 2 G ∂n2 n∗  2kB T G∗ Z = Z exp kB T  . $ (4. 4.36) Z = Z exp exp⎝ ⎠ . so the second exponential is a Gaussian function centered around n∗ .

4.1.44) n which is half the width between the vertical lines in Fig.43) (4. 17:28:32 . (b) Random additions and subtractions of atoms from a subsize cluster. which is proportional to the time required to cross the barrier.45) The distance (in number of atoms) across the nucleation barrier is δn.15 by equating the difference in free energy to kB T 1 ∂ 2 G (δn)2 .c. with markers indicating the range of n within kB T of the top of the barrier.45.c depict the fate of a cluster at the smallest size of this range. (4. The width of this region in cluster size is obtained with Eq. (c) Random additions and subtractions of atoms from a subsize cluster. The analysis of a random walk is given in Sect. 4. these events occur at random. kB T = − 2 n ∗ 2 ∂n & ' 2k T B ' δn = (− 2 . At any time it may add an atom and grow.8.9. 4. 4. 9.41 for Jss gives the steady-state flux of clusters from one size to the next.9a–c. 4.6 gives a mean-squared displacement after nj jumps R2 = nj r2 .9 n* n b n* n c n* n (a) Energy of formation of a nucleus. does not move continuously from small n to large n. In short. and even nuclei that have achieved n > n∗ can shrink and disappear. and the individual jump sizes (the r) are 1 on the cluster axis. ∂ G ∗ ∂n2 (4. resulting in successful nucleation.5. Without an energetic bias. An individual nucleus. 9. 4. The idea is illustrated in Fig. nj is the number of vectors in Fig. 40 Time kBT Time ΔG(n) 20 ΔG* 0 −20 −40 0 a t Fig. We treat the halting progress over the activation barrier as a random walk problem.4 Random walk across the critical size Equation 4. A nucleus undergoes fluctuations in its size. For our random walk of Eq.Nucleation 90 4. and the cluster executes a one-dimensional random walk along the axis of n. or lose an atom and shrink. For simplicity we treat all clusters within kB T of the top of the barrier as equivalent – there is no energetic bias to their growth or shrinkage. however. 4. resulting in disappearance of cluster.9b. Figures 4. Eq.9b. following Fig.

L = 1. The critical free energy of formation can now be found with Eq.47) n∗ where Z is given in Eq. as for Eqs.44. This onset is called the “critical undercooling. Just four degrees lower in temperature.5.47 is a very strong function of temperature. What they are good for is predicting temperature dependences of nucleation densities and rates. 4.42. when substituted into Eq. 4.47 we can estimate the density of nuclei that will grow. 4. Of course.5.2.5 The nucleation rate 91 The net time factor for the random walk across the barrier. and σ = 1 J/m2 .0 × 109 J/m3 . and for simplicity we neglect the temperature dependence of J0 . Defining T = Tm − T. gives the nucleation rate  2 16π σ 3 Tm 1 Jss = J0 exp − 3L2 kB T (T)2  . With this analysis up to Eq.48) where L is the latent heat of fusion and the approximation is good for small T.5 Temperature dependence of nucleation rate The nucleation rate of Eq. a number so low that the crystalline phase is undetectable. Wn ⎛ (4. ⎝− 2 ⎠ = 2 ∂ G n π W nZ ∂n2 (4. or obtaining quantities such as surface energies from these rates. and to illustrate the sensitivity consider a pure element that is quenched from a liquid to a temperature below the melting temperature Tm . This result for G∗ . where GV is approximately linear with temperature in the vicinity of the transition temperature. after 1 second the liquid will contain 100 nuclei per cubic meter. A word of caution – these estimates are not useful for predicting absolute numbers.11 GV −L T . and how long it takes for this to happen.6 and 4. 1. 1. with additional effects from elastic energy even these goals may be challenging. The extreme sensitivity to temperature suggests that for practical purposes a specific temperature can be identified as the onset of nucleation.” 17:28:32 . An order of magnitude estimate for J0 is 1040 /(m3 s). The time intervals are the time for an atom to attach or detach from the cluster τ= τ= (δn)2 . 4. 4. 4.4.49) where all pre-exponential factors were lumped into the quantity J0 . Tm (4. This is the situation shown in Fig. τ . (4. At an undercooling of 126 K below the melting point. Consider the freezing of a liquid for which Tm = 1000 K. 4.47.46) ⎞ 1 ⎜ 2kB T ⎟ 1 . T = 130 K. is therefore proportional to the square of δn of Eq. the nucleation rate jumps by a factor of 100.

t + δt) = N(n. (4. because clusters of size greater than n∗ = 28 continue to grow. but the rates W ↓ decrease with cluster size. t) + δt 4. t) Wn↑ + Wn↓ . The expressions for the W are specific for nucleation in a specific material at a specific temperature. There are few B-rich clusters of sizes greater than n = 1.1 Numerical example of time-dependent nucleation The kinetic master equation for transitions between the bins of different n is ∂N(n. t) ↑ ↓ = N(n − 1. using the initial condition where only clusters of size n = 1 exist at t = 0. Nevertheless.51) ∂t Results are shown in Fig.2 Analytical results for time-dependent nucleation An analytical calculation of the time dependence of nucleation requires the Fokker–Planck equation.1. minutes in this case. (4. For a nucleation event to occur at some time after t = 0. A minimum in cluster size is not expected when nucleation and growth continues at a steady state. 4. At t = 0.50) This master equation allows changes only by the addition or loss of single atoms from a cluster. t)Wn−1 + N(n + 1. 4.10. After this time. Kelton and Greer [27] provide a numerical example for oxide precipitation in silicon at a temperature of 800 ◦ C.6. a critical nucleus must form by atom transport through the α-crystal. 4. t) were found by numerical integration of the cluster size distribution ∂N(n. the steady-state cluster distribution never rises above this minimum for large n. Immediately after the quench a metastable α-phase matrix exists with B-atoms arranged more or less randomly on the lattice. In equilibrium the cluster density has a minimum at the critical size of n∗ = 28 in this case.6 Time-dependent nucleation Consider again the quench of Fig.10. 17:28:32 . The appearance of large clusters takes some time. W28 = W28 .6. obtaining its analytic solution is a formidable task. t)Wn+1 ∂t   −N(n. N(n. So before the steady-state nucleation rate can be observed. some time will pass before a cluster of critical size can form by fluctuations. Although the partial differential equation is linear. and is beyond the scope of this text. whereas W ↑ increases ↑ ↓ with cluster size.11 The time dependence of the different cluster sizes N(n. 4. Two approximate solutions exist.” the nucleation rate is very low. which is representative of infinite temperature.92 Nucleation 4. At the critical cluster size of 28 atoms for this simulation. the initial cluster size distribution has all atoms in single-atom clusters (n = 1). The cluster size distribution curves rotate counterclockwise in Fig. For times less than this “incubation time” or “nucleation time lag. Trinkaus 11 The details require extensive discussion. t) . the nucleation rate is approximately the steady-state value.

These two predictions of the time-dependent nucleation rate are also consistent with the qualitative picture described above – the rate is quite low for a time of τ .11).6 Time-dependent nucleation 93 Cluster density [/cm3] 1015 Steady state Equilibrium 105 10 min 60 s 10–5 10 s n* = 28 10–15 t Fig. Shi.4. The logarithmic scale shows an enormous range in cluster size. Seinfeld and Okuyama [29] obtained an analytic solution by employing the technique of singular perturbation. and Yoo [28] used a Green’s function approach to arrive at the following time-dependent nucleation rate:   1 π Z2 E2 (n∗ − 1)2 J(t) = Jss √ exp − .52) 1 − E2 1 − E2 where  t .52 and 4.53) (4.55) J(t) = Jss exp − exp −2 τ where √ λ = (n∗ )−1/3 − 1 + ln{3 π Z n∗ [1 − (n∗ )−1/3 ]} . For scale. (4. 17:28:32 .10 0 10 30 40 20 Cluster size [no.56) Although Eqs. the two approximate solutions are quite similar (see Fig. of molecules] 50 The evolution of cluster sizes with time. Their solution is     t − λτ . After [27]. From the definition of τ it is clear that the incubation time is a function of both the critical size and the attachment rate term β(n∗ ). (4. 4.54) where B is the rate of monomer attachment. 4. and rises rapidly to the steady-state value at later times. 4.55 have different mathematical forms. (4. the cluster density in the steady-state solution is exactly half that for the equilibrium solution at the critical size n = 28. and the graph shows enormous changes in cluster sizes over time. E = exp − τ and τ represents a natural timescale for the kinetic problem: τ= 1 2π Z2 B(n∗ ) . (4.

from [29]. The Trinkaus and Yoo [28] solutions are the dashed lines. Reproduced.) 17:28:32 .4 is 4π R2 (1 − cos θ ).0 t Fig. If a is the cube edge length.) 4. −Pr3 in three dimensions (where P is a constant.3 Suppose that nuclei formed in a phase transformation are cubic in shape. 4. where R is the radius of curvature. labeled by the cluster size n. with permission. Consider the difference in free energy between (1) a parent phase. σ is the specific surface energy. two. Problems 4. and Ev is the change in energy per unit volume upon forming the new phase (assume Ev < 0): (a) Calculate the critical size of the nucleus. and (2) a parent phase plus a nucleus of a second phase. and the critical free energy of formation W ∗ . Seinfeld and Okuyama [29] solutions are solid curves.5 0. and three dimensions. rather than spheres. A positive surface energy is also needed. Is the nucleus unstable at the critical radius? 4. in terms of σ and Ev .0 0 1 2 3 4 5 t/τ Time-dependent nucleation rates normalized by the steady-state value. for example.1 Phase transitions and dimensionality Assume a temperature where a second phase has a lower volume free energy than its parent phase. Assume that this difference contains a favorable contribution from the volume free energy of the second phase. 4. but its surface increases with tilt as 1/ cos θ.2 Perform the differentiations and substitutions to obtain R∗ and G∗ from Eq. 4. Copyright (1990) by the American Physical Society.Nucleation 94 0. with labels for three critical sizes. t) / Jss 1.4 Prove that the surface area of the double spherical cap of Fig. (Not difficult.11 100 20 50 n= J(n. (b) Explain why these differ from R∗ and W ∗ for a spherical nucleus. (Hint: One approach is to place a sphere of radius R at the origin and integrate its surface in rings from z = 0 to z = h. Calculate the critical radius in one. a∗ . and the Shi. The radius of the ring is R cos θ . perhaps involving π ). which scales as. 4.3.

95

Problems
4.5 Richard’s rule states that the entropy of melting Sm = 1 kB /atom.
(a) Neglecting pressure, show that the ratio Hm /Tm = 1 kB /atom (which is
another statement of Richard’s rule).
(b) Using Eq. 1.6 and Richard’s rule, calculate Fl − Fs for the metal nickel as a
function of undercooling below Tm , where the subscripts denote liquid and solid.
(c) The solid/liquid surface of Ni has a surface energy of 0.255 J/m2 . Using your
result from part b, and an undercooling of 10 K, graph Eq. 4.3 for Ni.
(d) Calculate the critical size of the nucleus, R∗ , and the critical free energy of
formation G∗ for Ni using the parameters of part c.
4.6 Consider the nucleation of a solid inside a solid. At a fixed temperature, elastic
energy can alter the chemical composition of the nucleus. For convenience, use the
free energy curves of Fig. 2.5, but here the “L” curve pertains to a high-temperature
solid parent phase. Consider a small, misfitting nucleus with the curve labeled “S” in
Fig. 2.5.
(a) What will be the relative effects of elastic energy on the two curves, i.e., which
curve, precipitate or matrix, will be affected most?
(b) Assuming that elastic energy causes the F(c) curve to shift rigidly upwards, how
will it affect the chemical composition of the nucleus of precipitate phase?

17:28:32

5

Effects of diffusion and nucleation on phase
transformations

A phase diagram is a construction for thermodynamic equilibrium, a static state, and therefore contains no information about how much time is needed before the phases appear with
their equilibrium fractions and compositions. It might be assumed that the phases found
after practical times of minutes or hours will be consistent with the phase diagram, since
most phase diagrams were deduced from experimental measurements on such timescales.1
However, a number of nonequilibrium phenomena such as those described in this chapter are well known, and were likely taken into account when a T–c phase diagram was
prepared.
For rapid heating or cooling, the kinetic processes of atom rearrangements often cause
deviations from equilibrium, and some of these nonequilibrium effects are described in
this chapter. In general, first one seeks to understand the effects of the slowest processes.
For faster heating or cooling, however, sometimes the slowest processes are inactive, so
the next-slowest processes become important. Approximately, Chapter 5 follows a course
from slower to faster kinetic processes.

5.1 Nonequilibrium processing of materials
5.1.1 Diffusion lengths
The diffusional motion of atoms is thermally activated, and is more rapid at higher temperatures. An important consideration is the characteristic time τ for diffusion over a
characteristic length x (see Eq. 3.42)
τ=

x2
,
D(T)

(5.1)

where D(T) is the diffusion coefficient.
Minimizing the time when D(T) is large serves to minimize x, so one way to classify
either kinetic phenomena or methods of materials processing is by effective cooling rate.
Table 5.1 shows some kinetic phenomena that are suppressed by cooling at increasingly
rapid rates. Insights into the phenomena listed in Table 5.1 are obtained with a rule of
thumb that diffusion coefficients near the melting temperature of many (metallic) materials
1 On the other hand, this does not necessarily mean that all phases on phase diagrams are in fact equilibrium

phases. Exceptions are found, especially at temperatures below about half the liquidus temperature.

16:49:53

97

5.1 Nonequilibrium processing of materials

Box 5.1

Thermodynamics and kinetics
Two necessary conditions for a phase transformation to occur in a material are
• a driving force from a reduction of free energy, and
• a mechanism for atoms to move towards their equilibrium positions.
The phase diagram gives information about the first condition, but most phase changes in materials occur by
diffusional motions of atoms.

Table 5.1 Cooling rates, methods, typical kinetic phenomena
Infinitesimal 10−6 K/s
Slow 100 K/s
101 K/s
103 K/s
Medium 104 K/s
105 K/s
Fast 106 K/s
109 K/s
Ultrafast 1010 K/s
1011 K/s


Geologic cooling

Equilibrium (sometimes)
Suppressed diffusion in solid
Dendrites
Suppressed precipitation
Suppressed diffusion in liquid
Extended solid solubility
Metallic glasses

Casting
Iced brine quench
Melt spinning
Piston-anvil quench
Laser surface melting

Amorphous elements
Physical vapor deposition
Shock wave
High-energy ball milling
Heavy ion irradiation

Melting
Nanocrystallinity, glass formation
Chemical mixing, glass formation

are D(Tm ) ∼ 10−8 cm2 /s.2 For an interatomic distance of typically 2 × 10−8 cm, Eq. 5.1
gives a characteristic time of 4 × 10−8 s. For ultrafast cooling, it seems possible to suppress
all atom diffusion in solids below the melting temperature, suppressing crystal nucleation
and growth. This allows the formation of amorphous metallic elements, but these are highly
unstable and may persist only at cryogenic temperatures. At 2/3 of the melting temperature,
the timescale for suppressing atom motion at the atomic scale is increased by a factor of
104 or so, owing to a decrease in D(T).

5.1.2 Quenching techniques
Some practical methods to achieve high cooling rates are also listed in Table 5.1. The
cooling rates are approximate, since these depend on the sample thickness and thermal
conductivity, which vary with the particular material and the equipment. Like the diffusion of atoms, the diffusion of heat has a quadratic relationship between the characteristic
cooling time τ and the sample thickness, x, as in Eq. 5.1, where D(T) is now a thermal
2 Sometimes an estimate of D ∼ 10−5 cm2 /s for the liquid proves useful, too.

16:49:53

98

Effects of diffusion and nucleation on phase transformations

diffusivity. Approximately, the techniques listed in Table 5.1 make thicknesses of materials that are proportional to the square root of the inverse cooling rate. Samples from melt
spinning are tens of microns thick, and samples from laser surface melting are often less
than 0.1 micron, for example.
Iced brine quenching is an older technique, where a sample at elevated temperature is
quickly immersed in a solution of rocksalt in water, cooled with ice. The salt serves to
elevate the boiling temperature and improve the thermal conductivity. The rate of cooling
depends on the thickness of the sample. The quench rate also depends on the formation of
bubbles of water vapor on the sample surface, which suppress thermal contact to the water
bath. Stirring the mixture can improve the cooling rate.
In melt spinning, a steady stream of liquid metal is injected onto the outer surface of a
spinning wheel of cold copper, for example. The liquid cools quickly when in contact with
the wheel, making a solid ribbon that is thrown off the wheel and spooled. This method is
suitable for high volume production. The liquid metal should have modest wetting of the
spinning wheel, and optimizing the parameters of the system can be challenging.
Piston-anvil quenching, sometimes known as “splat quenching,” uses a pair of copper
plates that impact a liquid droplet from two sides. The alloy is typically melted by levitation melting in an induction coil. When the radiofrequency heating current is stopped, the
liquid droplet falls under gravity past an optical sensor that triggers the pistons. Like melt
spinning, the sample is thin, perhaps 20 to 30 microns, but wetting properties are less of a
concern.
Laser surface melting can be performed with either a pulsed or continuous laser. The
sample may be moved in a raster pattern under laser illumination so a significant area can
be treated. Once melted, the surface is cooled by the underlying solid material, and the
thinner the melted region, the faster the cooling.
Physical vapor deposition may use a high-temperature heater to evaporate a material
under vacuum. The evaporated atoms move ballistically towards the cold surface of a
substrate. When these atoms are deposited on the cold substrate, their thermal energy is
removed quickly, in perhaps a hundred atom vibrations (approximately 10−11 s), leading
to very high cooling rates when the deposition rate is not too rapid and the substrate is
isolated thermally from the hot evaporator. Sputtering guns, in which a plasma of inert gas
ions knock atoms off a surface, can also be used instead of a thermal evaporator.
Some of the other methods in Table 5.1 have more ambiguous temperatures, or their
cooling rates are harder to define. Shock waves can be used to consolidate powders into
bulk materials. A high-power shock, delivered by a detonation or a high-velocity impact,
can melt the surfaces of the powder particles as it passes through them. The powder surfaces are cooled by thermal conduction to the powder interiors. Nonequilibrium materials
such as metallic glasses can be prepared in bulk form by such methods, although these
methods are not routine.
High-energy ball milling is typically performed by sealing the starting powder(s) in
a sealed steel vial containing hardened steel balls. Shaking the vial violently for several
hours causes severe mechanical deformation, fracture, and rewelding of the metallic particles. The sizes of the powder particles are difficult to understand, but the process does
produce chemical mixing on the atomic scale. How this occurs is not known in detail, but

16:49:53

99

5.2 Alloy solidification with solute partitioning

Box 5.2

Historical anecdotes
The liquid quenching of steels has a storied history, perhaps because early metalworkers did not have a good
understanding of the relevant time-dependent phenomena. The cooling medium does affect the cooling rate,
and hence the properties of steel. Especially prized were good mechanical properties of steel swords, which
could make the difference between life and death. Different cultures took different approaches to optimizing
the quenching medium and quenching procedures. Quenching methods passed on from earlier generations of
metallurgists include the following.
• The steel blade, heated to the color orange to bring it into the austenite phase, was quenched into the
urine of a red-haired boy.
• Another practice was to use urine from a three-year old goat, fed nothing but ferns for three days. Salt
quenching does have merit, although a large quantity of liquid is usually required.
• A procedure chronicled in a temple in Asia Minor was “Heat until it glows like the rising sun in the desert,
then cool it to the color of royal purple, plunging it into the body of a muscular slave. The strength of the
slave, transferring to the metal, imparts to it hardness.”

sometimes the process is modeled as brief bursts of elevated temperature. The effective
temperature cannot be particularly high, however, since this method is also useful for
producing nanocrystalline materials.
Ion beam bombardment with Xe+ ions accelerated to a few MeV is a way of knocking
individual atoms off their lattice sites, sometimes moving them tens of nanometers inside
the material. The ion or the primary knock-on atom may dislodge other atoms in a “collision cascade” that involves many atoms. A temperature is not appropriate for such ballistic
processes. The effective “cooling rate” is very high.

5.2 Alloy solidification with solute partitioning
5.2.1 Summary of nonequilibrium phenomena
Here we consider the kinetics of solidification, but several issues are common to other
first-order phase transformations. These include:



nucleation of the new phase(s);
effects of temperature gradients on the movement of the reaction front;
effect of the latent heat on the temperature profile; and
restricted partitioning of chemical elements between two phases when diffusion is slow.

All these effects are considered in this chapter, but we begin with the partitioning of solutes
between the solid and liquid phases. Slow diffusion in the solid suppresses equilibration,
and together with conservation of solute, the solidified alloy has a heterogeneous chemical
composition, different from what is predicted by its equilibrium phase diagram. A good

16:49:53

Effects of diffusion and nucleation on phase transformations

100

L+β

α+L

T
α

T

L

α+β

β

t

Fig. 5.1

c

c0

0

1
fL
0
1

0
1

0
1
cL
0
1

0
1

T0

Analysis of eutectic phase diagram for phase fractions, f , and compositions, c, upon cooling under equilibrium
conditions (thick curves at right), and when there is suppressed diffusion in the solid phases (thick dashed).
point of reference, however, is how solidification occurs under the equilibrium conditions
given by a T–c phase diagram.

5.2.2 Alloy solidification: case of equilibrium
We start with equilibrium predictions for phase fractions and compositions for an alloy of
composition c0 , cooled very slowly from the liquid and allowed to freeze and equilibrate
at all stages (Fig. 5.1), even though at lower temperatures equilibration may take geologic
times. The relevant graphs to examine are on the right of Fig. 5.1 (rotated by 90◦ for
correspondence to the temperature scale of the diagram), specifically the thick solid curves.
The six graphs are the fractions and compositions for each phase, L, α, and β, over a range
of temperatures. At high temperatures the alloy is entirely liquid of composition c0 , but
below the liquidus temperature the fraction of liquid decreases as crystals of the β-phase
grow in the liquid. Using the lever rule (Eq. 2.3), for our choice of c0 the fractions of
β-phase and liquid phase are approximately equal at the eutectic temperature of Fig. 5.1.
Their chemical compositions are simply read from lines on the phase diagram. Note that
the α-phase does not exist above the eutectic temperature for this alloy of composition
c0 , but the lever rule shows that immediately below the eutectic temperature the α-phase
should jump to a phase fraction, fα , of approximately 0.3.

5.3 Alloy solidification: suppressed diffusion in the solid phase
5.3.1 Mixing in liquid, suppressed diffusion in solid (general features)
Kinetic deviations from the equilibrium analysis of Sect. 5.2.2 can be predicted approximately. The kinetic processes with the longest timescales will be the first to alter phase
formation with faster cooling or heating. Atom diffusion in the solid phases is a kinetic
process that is typically slow, and is appropriate to consider first.

16:49:53

5.3 Alloy solidification: suppressed diffusion in solid phase

101

t

Fig. 5.2

a

b

(a) Nucleation of four solute-rich precipitates, with increasing volume added in shells at lower temperatures. (b) Later
stage of the phase transformation, with less solute in the liquid phase and less concentration of solute in the outer
shells, following the composition of the β-phase solidus at T.
Solid-state diffusion modifies the equilibrium cooling curves of Fig. 5.1 as follows. At
the highest temperature where the β-phase first forms during cooling (just below the liquidus for the composition c0 ), the β-phase is rich in B-atoms. The precipitates grow in
shells of changing solute concentration, as shown in Fig. 5.2a. At lower temperatures, the
phase diagram shows that this initial β-phase is too rich in B-atoms, and some B-atoms in
the solid should exchange with A-atoms in the liquid. This usually does not occur because
it requires atoms in the interior of the crystals of β-phase to diffuse to the outside of the
crystal to reach the liquid. Diffusion in the solid is usually too slow to ensure equilibrium
during cooling, so a heterogeneous solute concentration is expected in the solid, as shown
in Fig. 5.2b. Although the average composition of the β-phase is too rich in B-atoms, at the
surface of β-phase crystals, new layers of solid are forming with their equilibrium composition (for the appropriate temperature). The adjacent liquid composition is given by the
equilibrium liquidus line. One way to summarize this behavior is that the extra B-atoms in
the β-phase are immobilized and unable to participate in the equilibrium process, and the
effective alloy composition can be considered to decrease below c0 .
The effects are shown in the thick dashed curves of Fig. 5.1. By immobilizing B-atoms in
the B-enriched β-phase, conservation of solute requires that the total amount of β-phase be
somewhat suppressed at lower temperatures (consistent with the effective overall composition decreasing below c0 ). The same argument holds for temperatures below the eutectic
temperature – the B-rich cores of the β-phase allow fewer B-atoms to be available at temperatures below the eutectic temperature, and the effective chemical composition of the
alloy is shifted somewhat to the left. Then by the lever rule, less β-phase, and more αphase, will form at low temperatures. Incidentally, the areal fractions of solid and liquid
in Fig. 5.2b correspond approximately to the phase ratios at the eutectic temperature in
Fig. 5.1. The “pro-eutectic” β-phase in Fig. 5.2b has formed above the eutectic temperature, but at lower temperatures both the α- and β-phases form from the remaining liquid in
a “eutectic transformation,” depicted in Fig. 11.15 and described in Sect. 11.7.

5.3.2 Mixing in liquid, suppressed diffusion in solid (Analysis 1)
An approximate analysis of nonequilibrium solidification with no diffusion in the solid
phase is performed with the geometrical construction of Fig. 5.3. The equilibrium solidus

16:49:53

Effects of diffusion and nucleation on phase transformations

102

T3
T2
T1

L

Δc

T3
T2

α+L Liquidus

α

T1

Δc

Fig. 5.3

c

0
c0/k
cL c
0
0
c0
cα c/k
0
0

t

kc0 c0

1
fL
0
1

Solidus

Liquid–solid phase boundaries for the A-rich region of a T–c phase diagram. Black lines are the equilibrium phase
diagram. Using the gray triangles, the gray solidus curve is constructed as described in the text, giving the altered
compositions and phase fractions in gray at right.
and liquidus curves are drawn with black lines that are straight, so at each temperature they
give the same “partitioning ratio,” k
cs
,
(5.2)
k≡
cl
where k is a constant (equal to the inverse ratio of slopes). Figure 5.3 shows that as a
liquid of composition c0 is cooled, the first solid forms at temperature T3 , and has the composition k c0 . With the equilibrium solidus curve, by the temperature T2 an equilibrium
system would be completely solid, and the solid would have a uniform composition c0 . For
nonequilibrium cooling with no diffusion in the solid, at this temperature T2 the solidification is incomplete because there is an excess of B-atoms in the remaining liquid. We make
the simple approximation that the last solid to form will have a composition c greater
than c0 , because c is the deficiency in B-concentration of the first solid to form, and the
final solid must have an average concentration of c0 . This is shown in Fig. 5.3, where the
gray triangles are congruent. The phase fractions and their compositions are then drawn
with the same rules as for the equilibrium phase diagram, except that the gray line at left is
used for the solidus, not the equilibrium black one. This analysis gets the qualitative features correct, and highlights the importance of the partitioning ratio in Eq. 5.2 (often called
the “partitioning coefficient,” or “distribution coefficient”). The triangle construction is
qualitatively reasonable, but not rigorous.

5.3.3 Mixing in liquid, suppressed diffusion in solid (Analysis 2)
If we keep the assumption that the liquidus and solidus curves are straight lines, it is possible to perform a more rigorous analysis of the composition profile in the solid phase after
solidification, again assuming no diffusion in the solid. Start with the lever rule, Eq. 2.3,
which is an equation for the conservation of solute between solid and liquid phases
fs (c0 − cs ) = fl (cl − c0 ) , 
c 

s
fs (c0 − cs ) = fl
− c0 ,
k

16:49:53

(5.3)
(5.4)

5.3 Alloy solidification: suppressed diffusion in solid phase

103

where the partitioning ratio was used to relate cl = cs /k. We consider how small reductions
in temperature cause the transformation of small amounts of liquid into solid, and how
small amounts of solute are rejected into the remaining liquid. This requires a differential
form of the lever rule, obtained by differentiating Eq. 5.4 
1 dc
dfs
dcs
dfl  cs
s
(c0 − cs ) − fs
=
− c0 + fl
.
dT
dT
dT k
k dT

(5.5)

The transformation of a small fraction of liquid, dfl , into a small amount of solid, dfs ,
requires dfl = −dfs , since only two phases are present (and fl = 1 − fs ). After a convenient cancellation of all terms containing c0 , terms with dfs /dT are separated from those
with dcs /dT  

dc 
dfs  cs
1 − fs
s
− cs =
fs +
.
(5.6)
dT k
dT
k
At this point we make a simplification and an approximation. We simplify by ignoring
the details of the temperature dependence, because we have equality for each temperature
interval in Eq. 5.6. Our result below therefore gives cs vs. fs , but no dependence on T. The
approximation is that the term (dcs /dT)fs is small, and we neglect it. This proves reasonable
until fs is large in the later stages of solidification. The approximate Eq. 5.6 is
dfs
1 dcs
=
.
1 − fs
1 − k cs

(5.7)

For the limits of integration, start with cs = c0 /k at fs = 0 (because the earliest solid to
form has concentration c0 /k), and consider the different compositions cs = cl /k of the
solid phase at the fractions fs 
fs 
cs
1
1 
1 

dfs =
dc .
(5.8) 

1 − k kc0 cs s
0 1 − fs
The integrand on the left-hand side can be simplified by the change of variables from solid
to liquid fractions fl = 1 − fs , but the limits start at 1 (fully liquid) and decrease to 1 − fs . 
cs 
1−fs
1
1 
1 

(−1)df
=
dc .
(5.9)
l 

fl
1 − k kc0 cs s
1
Both integrals are elementary, giving logarithm functions
1−fs
c s

(k − 1) ln fl
= ln cs ,
1
kc 
0
cs
(k − 1) ln(1 − fs ) = ln
.
kc0

(5.10)
(5.11)

Exponentiating
cs = kc0 (1 − fs )k−1 ,

(5.12)

cl = c0 (1 − fs )

(5.13)

k−1

.

16:49:53

Effects of diffusion and nucleation on phase transformations

104

1.0

0.8
c0 (k = 1)
k = 0.9
k = 0.5
k = 0.25
k = 0.1

cs

0.6

0.4

0.2

0.0
0.0

t

Fig. 5.4

0.2

0.4

0.6

0.8

1.0

fs

Solid composition versus fraction of solid from the Scheil equation 5.12. The overall alloy composition was c0 = 0.25.
Equation 5.12 is the “Scheil equation,” and has been in use for a century [30; 31]. It gives
the composition of the new solid that freezes when the fraction of solid is fs . Equation 5.12
is especially useful in one-dimensional freezing, where the fs is proportional to the distance
across the solid.
When k = 1, so the solid and liquid have the same compositions (the phase diagram has
a liquidus and solidus that coincide), the composition of the solid is uniform, and Eq. 5.12
gives cs = c0 for all fractions of solid. As k decreases from 1, however, the composition
profile of the solid becomes increasingly nonlinear. Figure 5.4 shows some examples. For
this alloy of composition c0 = 0.25, the first solid forms when the temperature cools to
intersect the liquidus at c = 0.25. This first solid has the composition cs = c0 , and this is
the y-intercept at left of the graph. With decreasing temperature, the new solid forms with
increasingly more solute, crossing the composition c0 when fs is at least 0.5, depending
on k. When k is small, the final solid has to incorporate a large amount of solute that was
rejected into the liquid during the first half of freezing, and there is a spike in composition for the final solid that freezes. Unfortunately, the Scheil equation is not quantitatively
correct for the very last solid that forms, owing to the neglect of a term in the differential form of the lever rule.3 Nevertheless, the accumulation of solute in the final solid is
correct semiquantitatively, and an extremely heterogeneous composition is predicted when
the partitioning ratio is small (k  1).
Temperature was not considered in this analysis, but we can understand it qualitatively
from the first analysis with Fig. 5.3 (which we now see is most reliable when k is not much
less than 1). Most of the solid forms at higher temperatures that lie above the solidus for
the composition of c0 . Nevertheless, a strong solute enrichment of the remaining liquid can
delay the solidification of a small remainder of liquid until much lower temperatures.
3 For example, it predicts compositions in excess of 1.

16:49:53

105

5.4 Suppressed diffusion in solid and liquid

5.4 Alloy solidification: suppressed diffusion in the solid and liquid
Mixing in the liquid is usually incomplete, so the liquid is not homogeneous throughout, as
was assumed when deriving the Scheil equation. Typically there is a concentration gradient
in the liquid near the solid–liquid interface, and this gradient depends on the rate of freezing. Consider the two extremes. If the interface moves very slowly, we expect diffusion to
occur over a long distance in the liquid, ensuring homogeneity of the liquid composition
(homogeneity will also be promoted by convective currents, or perhaps stirring). By “long
distance” we mean the dimension of the liquid region. The other extreme is when freezing
occurs very rapidly. In this case it is tempting to say that the freezing interface outruns
the diffusion front, which may be a reasonable statement at long distances. At very short
distances the diffusion time goes as τ = x2 /D, so τ decreases rapidly at small x. Nevertheless, with a rapid velocity of freezing, the solid must accommodate all atoms, both A and
B, because the B-atoms do not have a chance to escape the moving interface.
When the very first solid forms, the liquid has a concentration c0 at the solid/liquid interface, but this concentration increases as solute is rejected into the liquid. The liquid with
higher solute concentration has a lower freezing temperature, so the solidification transformation is pushed to lower temperatures. If liquid diffusion is extremely slow, the temperature may decrease to T2 in Fig. 5.3. At this temperature T2 , the solid of composition c0 is in
equilibrium with a liquid of composition c0 /k, and there is no need to reject more solute at
the growing interface. After a transient period to form the concentration gradient, this case
of very slow diffusion in the liquid leads to a growth of a solid with concentration c0 at its
surface, unlike the previous case of Sect. 5.3 where good mixing in the liquid was assumed.
It is straightforward to calculate the solute diffusion profile in front of the moving
solid/liquid interface, and this is an example of a “moving boundary problem.” The diffusion profile depends on the diffusion equation (Eq. 3.13), so in one dimension the
concentration changes with time as
∂c
∂ 2 c
=D 2 .
∂t
∂x

(5.14)

Figure 5.5 shows a second contribution to the change in composition with time. When the
interface and its attached diffusion profile move to the right by the distance x per time 
t (so the interface velocity is v = x/t), the concentration at a point in the liquid
increases by c per shift x, as shown in the figure. This c/x gives a change with time
as (c/x)(x/t), or (c/x)v, giving a second term as
∂c
∂ 2c
∂c
=D 2 +v
.
∂t
∂x
∂x

(5.15)

There is an initial transient and a final transient, but ignore these for the moment. In
steady state with an interface velocity v, the concentration profile in front of the moving
interface remains unchanged, so ∂c/∂t = 0, and we seek the solution to the equation
D

∂c
∂ 2c
+v
=0.
2
∂x
∂x

16:49:53

(5.16)

The solution has the form exp(−xv/D). the following solution gives the correct features of Fig.1 c0 Solid Liquid 0. Nevertheless. there must be a depletion of solute at the far left where the solidification started and the solid had the lower concentration kc0 . 5.6. Here c0 = 0. and when the interface is far from the initial and final surfaces of the solid. After scaling this function to match the boundary conditions of the problem.6.3 Δx Δc Interface Δ x in Δ t 0. 5. Nevertheless.4 0. the actual situation is often somewhere between these two cases. 5. the shape of the concentration profile differs significantly – the two results are compared in Fig. 5. These two features are qualitatively similar to the case shown in Fig. as can be readily verified by substitution into Eq. 5. which gave different concentration profiles in the solid as shown in Fig. somewhat as described by the Scheil equation 5.4 with a depletion to the left and an excess of solute to the right of the solidified material.12. In reality.Effects of diffusion and nucleation on phase transformations 106 c0/k Concentration (c) 0. c(x) = c0 + k This concentration profile moves from left to right until it comes within the distance of D/v to the end of the liquid.17) − c0 e−xv/D . with partial mixing in the entire liquid giving a concentration gradient in the growing solid.0 t Fig. Here the excess solute in the liquid must freeze into an excess of solute in the final solid. 5.25.5 0 Distance in front of interface (x) Concentration profile in front of a moving solid–liquid interface when the interface velocity is constant.1 Alloy solidification At practical solidification rates.5.2 0. Likewise.1 and k = 0.16 (the argument is dimensionless because D has units of l2 /t).5 Practical issues for alloy solidification and evaporation 5. but a moving diffusion front tending to homogenize the solid at the average 16:49:53 . it is typical to ignore diffusion in the solid.5 c  0 (5. 5. we made two different assumptions about the solute mixing in the liquid.

0 Concentration (c) 0. 5. Also in reality. 5. or a few hours. 16:49:53 . although below its boiling temperature.6.0 0.0 t Fig. A substrate is placed some distance away. some tendency to follow the Scheil equation. Eq. where an alloy is heated above its melting temperature. it is not typical to have concentration profiles of centimeter dimensions near a solid–liquid interface.000 s.2 0.12) and mixing by a diffusion front in the liquid that is smaller than the length of the liquid (limited diffusion in liquid). which can be kept cold to suppress diffusion. 5.4 4 There is usually not much chemical mixing in the thin film material. and significant compositional heterogeneities in the frozen alloy.8 1. 5.6. There is a moving interface as the alloy evaporates.5.2 0. our one-dimensional analysis of a flat interface usually proves too simple.1 mm.6 0.6 0. At a distance of 0. and no diffusion occurs in the solid after freezing. In both cases c0 = 0. A diffusion time for a length of 1 cm is therefore τ =10.6 Scheil 0. this falls to 1 s because τ x2 /D. Diffusion coefficients for solutes in liquid metals are typically in the range of 10−5 to −4 10 cm2 /s. At macroscopic length scales. To reduce the free energy more rapidly.5 Practical issues for alloy solidification and evaporation 107 1.2 Alloy evaporation Thin films of materials can be prepared by high-vacuum evaporation. In the small gaps between the spikes there is some mixing in the liquid. 5. solute segregation more consistent with the phase diagram is possible if the solid forms as small spikes that jut into the liquid rather than as a flat sheet (the dendrite instability is described in Sect.0 Concentration profile along solidified bar for the two cases of complete mixing in the liquid (Scheil.25 and k = 0. composition c0 . and the dynamics can become unstable in three dimensions as shown in Sect.5. and a thin film of solid phase is deposited on this substrate with a chemical composition determined by of the fluxes of atoms leaving the hot metal.25.2).4 Limited diffusion in liquid c0 0.4 Distance along solidif ied alloy 0.8 0. and the hot liquid shrinks as it evaporates.

We assume that the film composition is the same as the density of the elements in the vapor above the evaporating metal surface. (5. is proportional to a vapor pressure pvB (T) that depends strongly on temperature. The diffusion profile below the evaporating liquid surface has the form   c0 pvA (T) (5.5 We need only replace the peak enrichment in the liquid.23) − c0 e−xv/D . c(x) = c0 + pvB (T) Suppose the B-atoms are the more volatile species.18) pB . the surface of the liquid alloy is depleted in Batoms so csA > csB . This requires an inverse relationship between the surface concentration and the vapor pressure of the element. but then the surface concentration ratio csB /csA can also vary by a large factor if the surface becomes enriched in the element that is less volatile. After an initial transient when evaporation starts. Using a typical D = 10−4 cm2 /s and a metal pellet 5 The physical picture should show a vacuum gap between the solid film at left and the diffusion profile below the liquid surface.Effects of diffusion and nucleation on phase transformations 108 A challenge in physical vapor deposition is controlling the chemical composition of the deposited film. 16:49:53 . with higher pvB (T) > pvA (T). c0 /k. Assuming an ideal gas with pressure pB pB V = NB RT . with c0 pvA (T)/pvB (T). the composition ratio is cB ρB = . the pressures at temperature T are in the ratio cs pv (T) pB = sB B . pA cA pvA (T) (5. the concentration of atoms in the vapor must be c0 because the atoms cannot diffuse away from the surface before they evaporate. Especially if a shutter can be used to block the deposition of the intial and final atoms at the substrate. (5.21) For an alloy of two elements. 5. It turns out that the same analysis of Sect. The liquid–solid phase diagram is no longer relevant. fast evaporation is a method that ensures the composition of the deposited film is c0 . but is also proportional to the concentration of B at the surface of the liquid. The pressure of element B above its liquid.19) ρB = RT where ρB = NB /V is the density of B-atoms. (5. but the picture of Fig.5 is useful.4 on slow diffusion in the liquid can be readily adapted to the problem of alloy evaporation. pB . and homogeneous. With a similar equation for the A-atoms. 5. (5. If the evaporation rate is much faster than the characteristic diffusion time across the liquid metal.22) The ratio of vapor pressures pvB (T)/pvA (T) can be orders of magnitude.20) cA ρA so we seek the pressures pB and pA above the hot liquid alloy. pB = pvB (T) csB . csB .

(The opposite situation is unstable. more heat needs to be transported through the solid.20) − κs dTs dTl v L. 5.6. so v is negative. the temperature gradient in the liquid is less than that of the solid. and v is positive.5. an evaporation time considerably shorter than 103 s is needed. In our one-dimensional problem (as in the growth of a flat interface between solid and liquid).) Figure 5. where the subscripts denote heat. so the difference in Eq. and liquid. (The effect of the latent heat on the temperature profile near the interface is discussed below with Fig. The extra flux of heat through the solid depends on the velocity of growth v as follows v L. Because the heat flux has the form of Fick’s first law (Eq.3 cm. 3. = −κl − dx dx Vmol (5. solid. There is melting of solid near the interface.1 Temperature profiles and interface motion After solidification is underway. but can occur as described below. however. Jhs . The latent heat is always a consideration. In this case.6 Heat flow and kinetics 5. and the interface moves slower. the diffusional flux of heat in the solid. With a larger latent heat. 5. 5.7e. Figures 5.25) The velocity of the solid–liquid interface is determined by the difference in temperature gradients in the solid and liquid   Vmol dTs dTl v= κs − κl .7b shows a smaller temperature gradient in the solid.) If there were no latent heat released at the solid–liquid interface. 16:49:53 .7a shows a typical situation where heat is extracted to the left. and the interface moves to the left.26 is positive. A total evaporation time of a few seconds may suffice. (5.d have a change of sign of the temperature gradients.7c. and the solidification front moves to the right.6 Heat flow and kinetics 109 for evaporation of 0.24) Jhs = Jhl − Vmol where Vmol is the volume per mole. Figure 5. What this means is that the heat will flow from the hot liquid to the interface faster than the solid can conduct the heat away. The solid cannot be overheated above Tm . would equal that of the liquid Jhl .26) L dx dx although the thermal diffusivities κl and κs also play a role. so Lv/Vmol has units of energy/(s cm2 ). As far as we know. as do the heat flux terms. (5. the solid must take away the latent heat. its rate depends on the removal of heat from the material. the solid melts instantaneously (within a few periods of atom vibrations) if heated above Tm . where the solid phase nucleates at a cold surface and grows into the liquid. The usual situation is to have the liquid hotter than the solid. and this makes an important difference.

(e) Modification of profile in part a with latent heat release at interface.2. 5.2 Dendrite instability Pure elements A common phenomenon in freezing is the growth of “dendritic” or tree-like microstructures. 33]. its detailed behavior cannot. Dendrite instability in alloys: constitutional supercooling A second type of supercooling can occur for liquids of alloys. 13. 5. 5.6.7c. 5. making a liquid that freezes at a lower temperature. 16:49:53 .” as described in Sect.Effects of diffusion and nucleation on phase transformations 110 Tm Solid Liquid Solid Liquid x Tm x a c Tm Solid Liquid Tm Solid Liquid x x b d Tm Solid Liquid unstable stable x t Fig. as shown in Fig. 5.7c. The interface. One important cause of this instability is the latent heat released during freezing. Tm . The solid front will grow rapidly into the liquid as a protuberance. is the warmest part of the growing solid – it is at the melting temperature. The protuberance can grow until it reaches liquid at a temperature of Tm or higher. If the latent heat causes the interface to be warmer than the nearby liquid. Suppose a B-rich solid phase rejects A-atoms into the liquid. Dendrites occur when the interface between the liquid and solid becomes unstable against asperities in the solid.7e.d require that the temperature of the solid is below Tm .d.6. where the situation is locally similar to those of Fig. and can grow to the right in the region labeled “unstable” in Fig. which are unstable.7 e (a)–(d) Four possible temperature gradients near solid–liquid interface. in contrast to a liquid which can be undercooled below Tm for observable times. which makes the interface warmer than the solid and liquid immediately around it. which grow forward rapidly into the liquid [32. so although the instability of a planar solidification front can be understood from the analysis of this section. meaning that the liquid is severely undercooled below Tm . This is an unstable situation.4. 5. These cases of Figs. A dendrite is a three-dimensional structure.7e. A protuberance that juts into this undercooled liquid can grow quickly into the undercooled liquid. 5. a “dendrite. as shown in Fig. there must be a temperature inversion in the liquid near the interface.

The general kinetics of a nucleation-and-growth phase transformation is indicated at the left of Fig. Dendritic freezing produces coarse compositional inhomogeneities in the solidified material. Note that the abscissa (x-axis) typically uses the logarithm of time.111 5. This chemical contribution to the liquid instability against protuberances is called “constitutional supercooling. although some undercooling is also typical of liquid→solid phase transformations. 5. A typical nucleation-and-growth transformation has a “nose” at a temperature where the transformation occurs in the shortest time. Undercooling is especially important for solid→solid phase transformations. Nevertheless. and held at this temperature for various times. and are slower at lower temperatures. superheating tends to be much less of an effect – an alloy will usually begin to melt close to the temperatures predicted from the phase diagram.2 explained that there is an energy penalty associated with the solid–liquid interface. 5. It pertains to a particular phase transformation at a particular chemical composition. The undercooling depends on many factors. The nose is the leftmost part of the curves in the TTT diagram of Fig. at exactly the critical temperature of a phase transformation.8. The number of nuclei of the new phase that are able to grow increases rapidly with the undercooling below the critical temperature of the phase transformation (Sect. and processings such as forging may be required to eliminate them.5). 6 On the other hand. The rate of the phase transformation is zero at the critical temperature (owing to nucleation). the free energies of the two phases are equal. 4. especially the contact between the liquid and the walls of its container. the diffusional processes needed for growth are thermally activated. Some undercooling is necessary before the new solid phase can nucleate. 5.6 Section 4. even for fast heating rates.7 Nucleation kinetics Nucleation often slows the formation of a new phase during cooling.8 is a kinetics map called a TTT diagram (for time–temperature–transformation). and is often charted as 10% and 90% complete on the TTT diagram.5. 16:49:53 . On the right side of Fig. this enriched liquid between the dendrites may freeze as a eutectic microstructure. A dendritic asperity that protrudes beyond this A-rich diffusion zone near the solid–liquid interface will grow rapidly into the liquid. so undercooling is needed to boost the thermodynamic driving force for nucleating a new volume of solid.” For alloys with eutectic phase diagrams. much as described in Sect. The assumption in using a TTT diagram is that the hightemperature phase is cooled instantaneously to the temperature on the ordinate (y-axis) of the plot. and sluggish at low temperatures (owing to diffusion). which is lower than the freezing temperature for the liquid beyond this diffusion zone. The extent of completion of the phase transformation varies with temperature and time.8. and it takes infinite time to nucleate the new phase at the critical temperature. On the other hand. 11. so it is not easy to predict.7 for the pearlite transformation in steel. Between these dendrites and their side branches is a liquid of composition that solidifies at lower temperatures. 5.7 Nucleation kinetics Freezing at a flat interface requires cooling the A-rich liquid to its freezing temperature.

5. In general. amenable to convenient analysis. at temperatures near the critical temperature.1 the β-phase forms with a lower concentration of B-atoms than predicted.Effects of diffusion and nucleation on phase transformations t Fig. A set of free energy versus composition curves is shown in Fig. Such crystallization does not occur.8 shows microstructures for two nucleation-and-growth transformations that came to completion in the same time. but liquid-phase diffusion can be suppressed at high cooling rates. and with the phase diagram of Fig. a phase transformation that occurs at a lower temperature is finer-grained than one occurring at a higher temperature. At these rates. solute partitioning between the solid and liquid is impossible. the formation of a crystal of β-phase of composition c0 would cause an increase in free energy. D.8 Glass formation Liquid diffusion is much faster than diffusion in a solid. Without solute partitioning between the α. An extreme case. 2. can occur for cooling rates in excess of 106 K/s.D Tc Temperature (T) N(T) D(T) 112 90% T 10% log(t) At left are shown the number of nuclei. Finally the top of Fig. An effective extension of A-atom solubility in the β-phase is expected owing to this kinetic constraint. 5. The common tangent construction for thermodynamic equilibrium predicts the coexistence of two solid phases (α and β) at this temperature. N.9. The partitioning of B-atoms between the solid and liquid is suppressed by high cooling rates. the number of nuclei was high.and β-phases. The TTT diagram is at right. which formed at the lower temperature. The opposite case was true for the microstructure at the right. 5. Tc . but their growth rates were low.12). 5. 5. Near the eutectic composition. deep undercoolings of the liquid are possible 16:49:53 . and all regions of the alloy maintain a composition of c0 at all temperatures. consistent with the TTT diagram. and typical microstructures are shown above it. but without solute partitioning the liquid phase is preferable over a wide range of compositions. 5. as indicated by the line with the arrowheads in Fig.8 N. The liquid phase is especially stable for compositions in the middle of Fig. For the microstructure at the left.9 (consistent with Fig.9. 5. of a nucleation-and-growth type of phase transformation. and diffusivity.

5.8 Glass formation

113

Free energy

FL

0.0

t

0.2

0.4

0.6
c0

c‘

0.8

1.0

Concentration (c)

Fig. 5.9

Free energy versus composition curves, similar to those for the phase diagram of Fig. 2.12 at a temperature below the
eutectic temperature. If solute partitioning between phases is suppressed by kinetics, at this temperature the liquid
phase may be expected for compositions 0.28 < c < 0.72.

FL(T1)
Free energy

T4
L
Temperature

FL(T2)

FL(T3)

T3
L+β

α+L
T2

α
α+β

F L(T4)

T0

β

T1

0.0

t
a

Fig. 5.10

0.2

0.4
0.6
Concentration (c)

0.8

1.0

0.0

0.2

b

0.4
0.6
Concentration (c)

0.8

1.0

(a) Free energy curves for two solid phases, α and β, and a liquid phase drawn for four temperatures as in Fig. 2.12.
The vertical gray lines mark the crossings of the curves for the liquid and the β-phase. These define the T0 -line in
panel b. (b) T0 -line constructed from points in panel a. It can be considered a liquid–solid phase boundary if chemical
segregation is suppressed.
at high quench rates. At low temperatures, however, the liquid becomes increasingly more
viscous. As the viscosity increases past 1012 Pa s, the alloy will not crystallize in measurable times. This frozen liquid is a “metallic glass” [34]. Many alloys having phase diagrams
with low eutectic temperatures can be quenched into a metallic glass by rapid cooling.
Figure 5.10 shows a modification of the previous eutectic free energy curves and phase
diagram of Fig. 2.12. The new features in Fig. 5.10a, shown in vertical gray lines, are the

16:49:53

114

Effects of diffusion and nucleation on phase transformations
crossings of the free energy curves of the liquid and β-phase. If no diffusion were possible,
the liquid would be most favorable at compositions to the left of these points, and the βphase would be favored at compositions to the right. The compositions and temperatures
of these points define the gray line labeled “T0 ” in Fig. 5.10b. (Another such point for one
temperature is indicated in Fig. 5.9 as the composition labeled c .)
The T0 line denotes the limit of solubility of A-atoms in the β-phase competing with a
liquid of the same composition, or equivalently the limiting composition where a metallic
glass could form upon rapid cooling. It is an important construction from the free energy
versus composition curves, but it is not directly measurable and does not appear on equilibrium T–c phase diagrams. It is often approximated as being located halfway between
the liquidus and β-solidus lines on an equilibrium phase diagram, but it probably is closer
to the solidus line because the free energy of the solid phase usually increases more rapidly
with composition.

5.9 Solid-state amorphization and suppressed diffusion
in a solid phase
Starting with two solid phases α and β, another example of a kinetic constraint is demonstrated with Fig. 5.11, which shows free energy versus composition curves for three solid
phases and one liquid phase. The third solid phase, the γ -phase, is typically an ordered
compound of A- and B-atoms. Many such compounds have large unit cells, requiring substantial diffusion for their nucleation. Their formation may be suppressed by rapid cooling,
but if the γ -phase formation is ignored, the analysis of glass formation upon cooling is the
same as that used for Fig. 5.9.
More interesting, however, is the possibility that intermetallic phases can be suppressed
upon heating, because this can cause a crystalline material to go amorphous. Suppose
that solid phases of A-rich α-phase and B-rich β-phase are brought into contact at low
temperatures, and atoms of A and B are allowed to interdiffuse across the interface. An
important consideration is that there are often large differences in the mobilities of A- and
B-atoms. As an extreme case, suppose that only the A-atoms are diffusively mobile in the
solid phases. The formation of the γ -phase, with its large unit cell, usually requires the
coordinated movements of B-atoms. If these movements are suppressed kinetically, the
nucleation of the γ -phase is suppressed. Without the presence of the γ -phase, the free
energy may reduce to the next-best phase, the liquid phase, as indicated by the top arrow in
Fig. 5.11. This process of “solid-state amorphization” produces an amorphous phase at the
interface between the α- and β-phases, and the amorphous phase is metastable at low temperatures. Solid-state amorphization was first reported in 1983 [35], but many examples
have since been found [36]. In the time required for the layer of metallic glass to grow to
macroscopic dimensions, however, the γ -phase usually nucleates and then grows quickly,
consuming the amorphous layer. A tricky issue in this analysis is the definition of the initial
composition c0 , because the reaction layer can extend to various distances in the α- and

16:49:53

5.10 Reactions at surfaces

115

L

Free energy

α+L

T
α

FL

γ+L

L+γ

L+β

γ

β
γ+β

α+γ


0.0

t

Fig. 5.11

a

0.2

0.4

0.6

Concentration (c)

0.8

1.0

0.0

0.2

b

0.4

0.6

0.8

1.0

Concentration (c)

(a) Free energy versus composition curves, similar to those of Fig. 5.9, but with a curve for an additional γ -phase.
Arrows indicate reductions in free energy that are possible from a starting mixture of α- and β-phases for overall
composition c0 = 0.33. (b) Equilibrium phase diagram for the four curves of part a.
β-phases below their interface. This is a common question for phase transformations in
thin films, at least in the early stages.

5.10 Reactions at surfaces
5.10.1 Linear and parabolic oxidation
The kinetics of phase transformations involving reactions at surfaces are often analyzed in
one dimension. Oxidation of metal surfaces is a typical example, but the analysis here
applies to many other surface reactions. A cross-section is shown in Fig. 5.12, with a
metal on the right and a gas on the left. Between them is an oxide of thickness x. In
this problem we assume that the oxygen atoms must move through the oxide layer to
reach the metal surface and react. The opposite case is often seen, however, where the
metal atoms diffuse through the oxide layer. Not surprisingly, the permeability of the oxide
layer for gases, or the diffusivity of ions in the layer, plays a major role in determining
the kinetics.
The simplest model for an interface reaction ignores rate limitations caused by diffusion
of atoms through the reaction layer, and sets a constant rate of growth
dx
=K,
dt
x(t) = Kt .

(5.27)
(5.28)

Equation 5.28 is often relevant when the reaction layer is thin, so the reaction time at the
interface, rather than atom transport through the reaction layer, controls the rate of growth.

16:49:53

116

Effects of diffusion and nucleation on phase transformations

c

c2

c1

x

t

Fig. 5.12

Top: concentration profile for oxygen atoms perpendicular to a metal surface. At the surfaces of the oxide layer, the
concentrations of reactive oxygen species may differ from the oxygen concentration in either the gas on the left, or the
metal on the right. Bottom: schematic diagram of layers.
The second simple reaction model assumes that the rate of growth of the reaction layer
is determined by the rate of arrival of atoms that diffuse across the reaction layer
J
dx
= ,
dt
ρ

(5.29)

where ρ is the density of the oxide (atoms/cm3 ). Fick’s first law (Eq. 3.20) gives the flux
of atoms across the reaction layer
J = −D

dc
.
dx

(5.30)

The concentration gradient is obtained approximately from Fig. 5.12
dc
c2 − c1
=
,
dx
x

(5.31)

With this concentration gradient, and equating the two fluxes of Eqs. 5.29 and 5.30
dx
c2 − c1
= −D
dt
x 
t 
x(t)
x dx = D(c1 − c2 )
dt ,
ρ
ρ

0 

x(t) =

0

2D(c1 − c2 )
ρ 

1/2

(5.32)
(5.33)

t.

(5.34)

Oxidation reactions that follow the kinetics of Eq. 5.34 are called “parabolic oxidation,”
and those following Eq. 5.28 are called “linear oxidation.” Both can occur in the same phase
transformation – in the early stages the reaction may have linear kinetics because it is not
limited by the diffusive flux, but with a thicker reaction layer the atom diffusion through
the layer may be rate-limiting. Fitting experimental data to both types of curves is typical
practice. Sometimes the rates of the two processes, set by K and D, differ enormously.
Aluminum metal, when exposed to air, forms a reaction layer of Al2 O3 almost instantly,
but this layer allows very slow diffusion for either aluminum or oxygen ions. The reaction
effectively stops after forming about 10 nm of Al2 O3 . We sometimes say that the reactive
surface of aluminum metal has been “passivated.”

16:49:53

5.10 Reactions at surfaces

117

cA
x
T

α

γ

0

cB

β
1

α

cA

γ

β

γ

β

x
α

t

Fig. 5.13

x

Top: A generic but naïve concentration profile for element A after some interdiffusion with element B (cf. Fig. 3.2b).
Immediately below it is a phase diagram giving compositions of phases at temperature T. The middle composition
profile shows how the different phases can be accommodated in the diffusion zone. Discontinuities in the composition
are of course expected at sharp interfaces.

5.10.2 Phases in concentration gradients
Consider a flat interface between two pure elemental materials, heated at temperature T
so interdiffusion occurs. The top image of Fig. 5.13 shows a typical interdiffusion profile
for a single phase. A single solid phase occurs for some systems, but more typically the
material prefers different phases at different chemical compositions, as indicated in the
double eutectic phase diagram below the profile. In this case there will be a tendency to
form different phases with different compositions in the diffusion zone near the interface.
The phases α and β remain pure outside the diffusion zone, but in the zone they accommodate atoms of the other chemical element. The phase diagram suggests, however, that a
γ -phase of intermediate composition will form between the α- and β-phases. This requires
two distinct interfaces between the phases, α/γ and β/γ . At these interfaces, the chemical compositions in the adjacent phases are expected to be those at phase boundaries in the
phase diagram. The locations of the interfaces cannot be predicted from the phase diagram,
of course, and they move as the reaction proceeds.
The γ -phase is expected from the equilibrium phase diagram of Fig. 5.13, but it is not
necessarily present at all stages of the reaction. It is possible for the interdiffusing atoms to
spread over a wide spatial range of the α- and β-phases, even if the composition variations
in these phases are not large. This may be the case if the nucleation of the γ -phase is
suppressed.

5.10.3 Suppressed nucleation in steep concentration gradients
Nucleation can be suppressed by the large concentration gradients that occur in the early
stages of an interdiffusion reaction [37]. Consider the situation shown in Fig. 5.14, where

16:49:53

118

Effects of diffusion and nucleation on phase transformations

R*

cA

Free energy

x

FL


ΔF
0

t

Fig. 5.14

γ

0.5
Concentration (cB)

1

Top shows a generic interdiffusion concentration profile. The critical radius for nucleation has the width R∗ as marked
in the figure. Lower figure, with composition aligned approximately in correspondence with the top spatial profile,
shows free energy versus composition curves. The curve for the γ -phase rises steeply for small changes in
composition.
the γ -phase compound has a rather narrow range of composition where its free energy is
low compared with other phases. In a thin-film reaction with a steep composition gradient,
there may be only a narrow range of x where the composition is appropriate for the γ phase to have a low free energy. When this range of x is smaller than the size of the critical
nucleus of the γ -phase, nucleation of the γ -phase cannot occur. If F(c) is parabolic with
curvature ∂ 2 F/∂c2 , nucleation will be suppressed when 
F <

1 ∂ 2F
2 ∂c2 

dc ∗
R
dx 

2
,

(5.35)

where F is the free energy change favoring the formation of the new phase (see
Fig. 5.14), R∗ is the critical radius for nucleation, and dc/dx is the concentration gradient. Equation 5.35 states that nucleation is suppressed if the edges of the new phase are
at compositions with specific free energies above the common tangent of Fig. 5.14. The
larger the concentration gradient, the more difficult is nucleation. Nucleation in a diffusion
zone is also suppressed if the new phase has a large critical nucleus.

5.10.4 Phase growth in thin films
A phase diagram often contains many phases of different compositions. For thin-film reactions, important questions are “Which phase forms first?” and “Do all the phases form?”
These questions sound simple, but prove complicated, and require kinetic information well
beyond what can be gleaned from the equilibrium phase diagram. For example, the suppression of nucleation in concentration gradients (the topic of the previous subsection) is
not the same for different phases, so some nucleate more easily than others.

16:49:53

5.10 Reactions at surfaces

119

JA

cA
x

α

t

Fig. 5.15

x

γ
x0

δ
v1

β
v2

Top shows a typical average interdiffusion concentration profile, with typical concentration discontinuities at
interfaces between phases. Lower figure shows two new phases, γ and δ, that have nucleated and started to grow.
The growth velocities of the phases are another important source of kinetic control [38].
Consider the picture shown in Fig. 5.15, where for simplicity it is assumed that only Aatoms move (with net flux to the right), and the interface velocities depend on the arrival of
A-atoms and their attachment to the growing phase. Both the processes of A-atom arrival
by diffusion, and their attachment at the interface, cause the interfaces to move at different
velocities, labeled as v1 and v2 in Fig. 5.15. Using the position of the α/γ interface as a
reference at position x0 , suppose that the interfaces move to the right.7 If growth is to the
right, and if the v1 is less than v2 , the widths of both the γ -phase and the δ-phase will
grow with time. This need not be the case, and it is quite possible that v1 will exceed v2 , so
the γ –δ interface at left may outrun the δ–β interface to its right. In this case the δ-phase
will disappear. Perhaps the δ-phase will never form, since it would always be overtaken
by the γ -phase. On the other hand, perhaps the δ-phase nucleates more easily than the γ phase, so there is an early period of time where only the δ-phase is present, but once the
γ -phase forms, it quickly consumes the original δ-phase. All these cases occur in different
interfacial reactions.
Owing to their technological importance, some of the most thoroughly studied thinfilm reactions involve metals and silicon, typically with metal layer thicknesses of 10 to
100 nm [39]. In these cases when the metal layer is thin, and silicon is the majority species,
eventually the final phase is a compound with the largest fraction of silicon. This is the
state of equilibrium expected for a silicon–metal alloy with low metal composition.
The binary phase diagrams of transition metals and silicon have numerous phases at
low temperatures. What is interesting is that in chemically pure systems, usually only one
phase forms at a time. It often consumes all the metal layer, and remains metastable until
a second phase forms and consumes it. For predicting the first phase that forms, an empirical correlation is the Walser–Bené rule, which states that the first silicide to grow is the
highest-melting compound next to the lowest-melting eutectic in the phase diagram [40],
but exceptions occur. The next phase to form is often a phase with high melting temperature
that is rich in the unreacted element, but again this rule is not fundamental. Another independent, and perhaps contradictory, rule for silicide formation is that the phase formation
7 In Sect. 9.2 we will see that the entire block of material moves a bit to the right under the conditions shown, so

all interfaces can be argued to move to the left. Location markers and reference positions are important issues
for interdiffusion.

16:49:53

120

Effects of diffusion and nucleation on phase transformations

sequence depends on whether the silicon or the metal atom is the more mobile. When the
metal atom is more mobile, the equilibrium phase tends to form first, but when the silicon is more mobile a series of silicide compounds form first. In spite of the technological
importance of phase transformations in thin films of metals and silicon, they are not well
understood.

5.11 The glass transition
Figure 5.16 shows the heat capacity of a metallic glass, CuZr, as it is heated. The alloy was
prepared by rapid cooling from the liquid state (>1,000 ◦ C) to room temperature by injection into a cold copper mold. The as-cast alloy was largely amorphous (a metallic glass).
Upon heating in the calorimeter, it remained in the amorphous state at low temperatures
up to the crystallization temperature around 445 ◦ C. Before the large exothermic peak of
crystallization (usually called “devitrification”), however, the amorphous material softens,
becoming a viscous liquid that is deeply undercooled below the usual melting temperature
of 935 ◦ C. The heat capacity of this undercooled liquid is about 1.9 kB /atom greater than
that of the amorphous alloy. This increase in heat capacity originates primarily from the
additional degrees of freedom that accompany atom mobility. The onset of this change,
labeled Tg in Fig. 5.16, is called the “glass transition.” It is usually not considered a phase
transition, although it can be reversible and it involves a change in free energy.8
In 1948 Walter Kauzmann noted that with decreasing temperature, the heat capacity
of a liquid decreased more rapidly than that of a solid [41]. Extrapolating these trends,
Kauzmann found that the entropy of the liquid would be lower than that of a solid at
finite temperature, sometimes called the “Kauzmann temperature.” This would be odd if
the liquid phase were the high-temperature phase. There are other problems because the
configurational entropy of the liquid is typically larger than for crystals that are the equilibrium phases at low temperatures. The “Kauzmann paradox” is part of the history of glass
science. The resolution of the paradox for metallic glasses is that the liquid loses degrees
of freedom in atom mobilities at low temperature as it becomes a solid glass, so the entropy
of the liquid does not decrease so rapidly at low temperature. Kauzmann’s extrapolation is
unreliable.
The loss of atomic degrees of freedom upon cooling originates from a stronger chemical
bonding in some regions than others. There are numerous local configurations of atoms
or molecules in a liquid, and some are more favorable than others for molecular or atomic
motions. The increased viscosity of a liquid upon cooling originates from the locking-up of
atom positions by chemical bonding. For different materials, this lock-up with temperature
may occur more quickly or slowly. If the viscosity decreases rapidly above Tg , as is the
case for nonpolar organic molecules such as toluene or metallic glasses, the glass is termed
“fragile.” A “strong” glass, such as SiO2 , undergoes a more gradual change in viscosity
with temperature [42; 43].
8 Please do not confuse the glass transition with devitrification; these are completely different processes.

16:49:53

Problems

121

Tg Tx

10

Heat capacity (kB/atom)

0

t

Fig. 5.16

–10
–20
–30
–40

200

300

400
500
Temperature (˚C)

600

700

Calorimetry measurements on CuZr metallic glass heated at a rate of 20 ◦ C/s. Heat absorption (endothermic feature)
occurs at the glass transition temperature Tg , and a large heat emission (exothermic peak) occurs when the alloy
crystallizes at Tx . (The broad feature around 540◦ C is another crystal relaxation, but no correction was performed for
the background signal of the calorimeter.)
The prominent devitrification (crystallization) peak in Fig. 5.16, starting at Tx = 445 ◦ C
with a maximum at 447 ◦ C, is not part of the glass transition. Nevertheless, it is of practical
importance because it is an irreversible transition from the amorphous state.9 The nucleation and growth of crystals requires some diffusion of atoms. As such, the temperature Tx
increases if the heating is faster, giving a larger temperature range where the undercooled
liquid is present, and the opportunity to have a lower viscosity liquid for shaping or forming. The glass transition involves more localized motions of atoms, however, and does not
involve significant diffusion distances. Typically Tg is less sensitive to heating rate than
Tx . Shape forming of metallic glasses by heating is considerably more robust if the heating
and processing times are short.

Problems
5.1 This problem uses the eutectic phase diagram of Fig. 2.19 with phase boundaries
made from straight lines. The phases α, β, and liquid have the same density. An alloy
of 70% B is cooled from the melt.
(a) Plot the volume fraction of each phase versus temperature, assuming thermodynamic equilibrium.
(b) Show how each plot of part a changes when there is no diffusion in the solid
phase that has formed.
9 The kinetics of devitrification may follow the behavior shown in the TTT diagram of Fig. 5.8, since

crystallization occurs by nucleation and growth.

16:49:53

Effects of diffusion and nucleation on phase transformations

122

Motion of heater and melt zone
Liquid
Solid

Liquid

Solid

T

Heater
Solid
A

t

Fig. 5.17

a

b

Composition

B

Initial composition of bar

(a) Conceptual design of apparatus for zone refining. (b) A-rich region of phase diagram used for purifying A-rich solid
by zone refining.

5.2

5.3

5.4

5.5

(c) Using free energy vs. composition curves for the phases α, β, and liquid, justify
why an amorphous phase could form if there were no diffusion in the liquid
during rapid cooling.
The formation of “pipes,” or sometimes internal cavities, is a common problem
in the centers of thick castings, where a liquid metal fills a cup-shaped mold and
freezes.
(a) Considering the sequence of how the liquid freezes when poured into a cold
mold, and how the solid phase contracts as temperature is reduced, explain the
formation of cavities or pipes.
(b) If the linear thermal expansion of the solid is 2 × 10−5 K and the freezing temperature is 1,300 K, estimate a size of a pipe or cavity for a mold of thickness
10 cm.
The Scheil equation that relates cs and fs does not include temperature. Draw a diagram that estimates the freezing temperatures corresponding to fs for the case of
k = 0.5 in Fig. 5.4.
“Zone-refining” is a technique to make high-purity crystals of a pure element. Without zone-refining of silicon, it is unclear whether there would be a semiconductor
industry. In this technique, a heater melts a small zone of material, and the heater and
zone move along a bar of material (see Fig. 5.17a). The solutes segregate to the far
end of the bar. The technique works for removing B-solutes from A-metal when the
A-rich part of the A–B phase diagram is as is shown in Fig. 5.17b.
(a) Why does zone-refining work?
(b) Would the material become more pure after a second pass of the heater?
(c) For what combination of slopes of the liquidus and solidus would zone-refining
work best?
Figure 5.9 shows the sequence of free energy reductions that pertain to the formation
of a glass in a binary alloy system with a composition c0 = 0.6.
(a) How do you expect the glass-forming tendency to change for compositions c0
greater than or less than 0.6?
(b) Explain why phase diagrams with “deep eutectics” are promising for glass formation by quenching from the liquid. In your explanation, draw and discuss the
T0 lines on a eutectic diagram.

16:49:53

123

Problems

5.6 Interfacial reactions
(a) Consider nucleation in a concentration gradient of 0.01/nm. If the F(c) for
a new phase is quadratic in composition, and rises from its minimum to
1.0 eV/atom over a composition range of 0.05, what is the largest critical radius
that is possible if the change in free energy, F, must be less than 0.1 eV/atom?
(b) Interfacial reactions are often performed at low temperatures, where there are
substantial differences in rates of nucleation and growth of the different phases.
Consider the diagram of Fig. 5.15, and suppose that the growth velocity v1 is set
by the diffusion of A-atoms through the α-phase, whereas the velocity v2 is set
by the diffusion of A-atoms through the γ -phase. Suppose the diffusion coefficients for the two phases are both of the form D(T) = D0 exp(−Q/RT), where
D0 = 10 cm2 /s for both phases, but Qα = 50 kcal/mole and Qγ = 60 kcal/mole.
Assume the velocity of growth is proportional to the characteristic diffusion
length.
At a temperature of 400 ◦ C, what is the ratio of v1 and v2 ?
Approximately how long will it take for the δ-phase to disappear at 400 ◦ C if it
has an initial thickness of 10 nm?
5.7 Suppose the Kauzmann paradox really occurred, i.e, the entropy of the undercooled
liquid decreased below the entropy of the optimal crystal structures.
(a) Why would this pose difficulties for thermodynamics, or would it?
(b) Suggest a resolution to the Kauzmann paradox, different from the one in the
text.
(c) Is it possible in principle for an amorphous phase to be the thermodynamic
ground state of an alloy? Why or why not?

16:49:53

16:49:53 .

The next chapter. 16:49:54 . has a singularity. Chapter 9 explains chemical effects on diffusion in alloys. Chapter 7. The Gibbs free energy is G = E − TS + PV . This coverage of E. and thin film reactions. which can be calculated with an activated state rate theory that uses a free energy of activation. and Chapter 8 discusses effects of pressure. addresses the important sources of entropy. P. S. or a derivative of its free energy. Chapters 3 and 4 discussed the kinetics of diffusion and nucleation. Finally. Chapter 2 showed how to use the dependence of free energy on composition or order parameter to obtain thermodynamic phase diagrams. Chapter 6 discusses the sources of energy of materials that are important for phase transitions. and G∗ makes up Part II of the book. Chapter 5 showed how the free energies of equilibrium phases and the free energies of activation give rise to competition between thermodynamic and kinetic phenomena in phase transformations such as alloy solidification. which depend on the free energy of an activated state.PART II TH E ATOMIC ORIGINS OF TH ERM ODYNAMICS AND KINETICS Free energy is a central topic of this book because a phase transition occurs in a material when its free energy. glass formation.

16:49:54 .

As isolated atoms. giving an electron density that is not a linear sum of densities from two separate atoms. At a more general but more phenomenological level. The concepts presented here are important. 6. Surface energy is also described. The effect of the second electron is to push around the first electron. interatomic potentials are described and used to explain the elastic behavior of solids.1 Molecular orbital theory of diatomic molecules 6. the number of electron states depends on a wavevector k. The elastic energy of a misfitting solid particle in a matrix is discussed. their nuclei are far apart. along with the Wulff construction for predicting the shapes of crystals and precipitates. 2m − (6. and this misfit energy is generally important for precipitation reactions in solid materials.2) where ψA and ψB are single-electron wavefunctions at atoms A and B. Now bring the nuclei close enough together that their wavefunctions overlap. which can be used to obtain the spectrum of electron energies.1) (6.1. and some techniques for calculating them. but quantitative results require quantum chemical computer calculations. and understand the chemical bond in the new diatomic molecule. We seek single-electron wavefunctions for the diatomic molecule. There are states for a single electron about each atom of energy A and B . It begins with the chemical bond between two atoms – a fundamentally quantum mechanical phenomenon that depends on the coherent interference of an electron wavefunction with itself.6 Energy This chapter explains the different types of energies that are important for the thermodynamics of materials phases and materials microstructures. 2m 2 − ∇ 2 ψB (r) + VB (r)ψB (r) − B ψB (r) = 0 . A and B. set by the Schrödinger equations 2 2 ∇ ψA (r) + VA (r)ψA (r) − A ψA (r) = 0 . but 17:29:19 .1 Interacting atoms Start with two isolated atoms. Our goal is to understand what the individual electrons do in the presence of both atoms. The potential proves to be a real challenge because the potential for one electron depends on the presence of the second electron. In a periodic solid or in a large box for electrons.

2m A 2 − ψB∗ (r)∇ 2 ψ(r) + ψB∗ (r)V(r)ψ(r) − ψB∗ (r)ψ(r) = 0 .3) This approach does not always work.6) and (2) integrate A|H|A. (6. (6. lay out the molecular Schrödinger equation twice and do two standard tricks: (1) multiply by ψA∗ (r) and ψB∗ (r) 2 ∗ ψ (r)∇ 2 ψ(r) + ψA∗ (r)V(r)ψ(r) − ψA∗ (r)ψ(r) = 0 . so the coefficients cA and cB are less than 1 (the atomic wavefunctions ψA and ψB accommodate one electron each). This ψ is a “molecular orbital” for one electron.5) (6. so we need to find two molecular orbitals. though. We started with two electrons. 2m − (6. The approach works best when the overlap of the atom wavefunctions is small. especially when there are large electron transfers between atoms. but here assume that the total potential is simply the sum of potentials of the isolated atoms V(r) = VA (r) + VB (r) . which alter the atomic potentials. and the potentials tend to retain their original character. To do so.4) It is important to remember that ψ pertains to a single electron.Energy 128 this alters the potential and wavefunction of the second electron. Iterative methods are the most accurate for this problem. We make a related assumption that a single electron is in a wavefunction ψ constructed from the original atomic wavefunctions ψ(r) = cA ψA (r) + cB ψB (r) .

cA + A|H|B.

cB − (cA + A|B.

7) B|H|A.cB ) = 0 . (6.

cA + B|H|B.

cB − ( B|A.

8) where the integrals are written in Dirac notation. (6. Equations 6.7 and 6.8 can be arranged as a matrix equation ⎡ ⎢ ⎢ ⎣ ⎤⎡ ⎤ A|H|A.cA + cB ) = 0 .

−  ⎡ ⎤ A|H|B.

−  A|B.

⎥ ⎢cA ⎥ ⎢0⎥ ⎥⎢ ⎥ = ⎢ ⎥ . ⎦⎣ ⎦ ⎣ ⎦ cB 0 B|H|A.

−  B|A.

B|H|B.

9 for  and then for cA and cB . 6. The integrals A|B.9) 6.2 Definitions and conventions Before solving Eq.1. we evaluate some terms and change notation. −  (6.

and B|A.

but the tails of these wavefunctions overlap. These are “overlap integrals. are not zero – the wavefunctions are centered on different atoms.” defined as S S ≡ A|B.

= B|A.

17:29:19 (6. .10) .

1 Molecular orbital theory of diatomic molecules Compared with the case of isolated atoms.129 6. (6.11) so the integrals on the diagonal of Eq. the one-electron molecular Hamiltonian H differs only by the potential VA or VB H = HA + VB = HB + VA .9 have the form A|H|A. 6.

= A + A|VB |A.

8.12 is the presence of the A-atom electron at the B-atom potential. discussed in Sect. this relatively small term is neglected. giving A|H|A. . This is a “crystal field” term. (6. At present.12) The second term in Eq. however. 6.4.

13) B|H|B. = A . (6.

and the difference of A and B as  = B − A A|H|A. = B . (6.14) Taking the average of A and B as  = (A + B )/2.

15) B|H|B. (6. =  − /2 .

(6. but certainly not least.16) And last. we define the “bond integral. =  + /2 . as h = A|V|B.” h.

(6. the bonding level  + and the antibonding level  − are determined by the bond integral. .20) The two molecular orbitals for single electrons are shown in Fig. 6.17. (6.17) where the average of the potential is V = (VA + VB )/2. h of 17:29:19 .3 Bonding and antibonding energies Using Eqs. 6.18) Setting the determinant of Eq. so the difference  = 0. 6. For identical atoms.19)  =+ 1 − S2 and neglecting small terms of order S2  ± =  + |h|S ∓ * h2 + (/2)2 .18 equal to zero. ⎦⎣ ⎦ ⎣ ⎦ /2 − ( − ) cB 0 (6.13–6. A = B . 6.1. (6. and bonding of two different atoms. and solving the quadratic secular equation.b for two cases – bonding of two identical atoms.1a. gives two solutions for the energy ) |h|S ∓ h2 + (1 − S2 )(/2)2 ± . 6.9 becomes ⎡ ⎢−/2 − ( − ) ⎢ ⎣ h − ( − )S ⎤⎡ ⎤ ⎡ ⎤ h − ( − )S ⎥ ⎢cA ⎥ ⎢0⎥ ⎥⎢ ⎥ = ⎢ ⎥ . When the identical atoms are brought together. Eq.

The more electronegative the atom. The c± A and cB need normalization to 1 because there is one electron in either the bonding or antibonding molecular orbital + + + (6.18 becomes ⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎢−|h|S ± |h| ⎢ ⎣ h ± |h|S h ± |h|S ⎥ ⎢cA ⎥ ⎢0⎥ ⎥⎢ ⎥ = ⎢ ⎥ . ⎦⎣ ⎦ ⎣ ⎦ cB −|h|S ± |h| 0 (6.17. should be added to both of these energies. so this term is small compared with the bond integral. 6. In contrast. 6. With B > A .Energy 130 ψ– ψA ЄA ψ– Unfavorable 2h ψA – ЄB Є — ∆Є 2√h2+(∆Є/2)2 ЄA ψ+ t Fig. Eq. h. 6. and cA /cB = −1 for the antibonding molecular orbital.1. |h|S.22) 1 = +|+. and the energy difference between bonding and antibonding grows with . Perhaps the middle term of Eq. The bonding energy increases with . 6. the more it attracts electrons to itself in the bonding state.20 confirms that the homonuclear case of Fig. Equation 6.1 Favorable a ψB Є– ψB Є– Є+ E ЄB Є+ ψ+ b Energies and wavefunctions of molecular orbital formation in diatomic molecules.4 Molecular wavefunctions We now solve for the cA and cB in Eq. as shown in Eq. specifically their difference in “electronegativity. 6.20. 6. the antibonding level becomes more unfavorable. With identical atomic energies and wavefunctions. such as HF.21) where terms of order S2 were neglected.4. 6. 6. the one-electron molecular orbitals are symmetric. the bonding level ψ + is lower because it puts more of the electron in the favorable atomic orbital. Eq. Energy scale and definitions are in the middle. The antibonding orbital ψ − puts more electron in the unfavorable orbital.1a has a bonding energy of  − h and an antibonding energy of  + h (here  = A = B ). for which  = 0 and  −  = +|h|S ∓ |h| from Eq. which tell us about the molecular orbitals and electron density in the bonding and antibonding states. (a) Homonuclear molecule. 6. Recognizing that the bond integral h is negative. This  is a difference of atom properties. 6. the stronger the bond.” set by A and B . (b) Heteronuclear molecule. such as H2 . For the homonuclear molecular orbitals. + using either line of Eq. 6.20 and in Fig.1b.20. but the overlap S is relatively small.21 we can show that c+ A /cB = +1 for the bonding molecular − − ± orbital. The larger the difference in electronegativity.

= c+ A A| + cB B| cA |A.

+ cB |B.

23) 1 = −|−. . − − − − (6.

= cA A| + cB B| cA |A.

+ cB |B.

17:29:19 . .

+2 +2 +2 1 = +|+.131 6.1 Molecular orbital theory of diatomic molecules + − − and with c+ A = cB and cA = −cB .

= c+2 A A|A.

+ cA A|B.

+ cA B|A.

+ cA B|B.

24) 1 = −|−. (6. .

= (6.25) −2 −2 −2 c−2 A A|A.

− cA A|B.

− cA B|A.

+ cA B|B.

10. 1 + . It is zero for the antibonding. the bonding wavefunction is smoother in this region between the nuclei. Obtaining the electron density as ψ ∗ ψ. the wavefunction amplitude is doubled for the bonding wavefunction. The potential energy actually favors the antibonding level over the bonding in molecular orbital theory with atomic wavefunctions. Molecular wavefunctions of the H+ 2 molecular ion have been calculated accurately. Unfortunately. 6. and the antibonding some increase in amplitude. 6. and the bonding level a decrease. and the kinetic energy increases because of a larger ∇ 2 ψ. c+ A = +cB = √ √ 2 1+S 1 − . This can be understood from the electron density in the regions of overlap. Both the electron potential energy (which decreases when the electron is bunched near the 17:29:19 . so it has the lower kinetic energy. compared with the wavefunction in the tail of an atom wavefunction.1a. c− A = −cB = √ √ 2 1−S .26 and 6. it is interesting that with increasing overlap S. to extend the argument of the previous paragraph to explain the origin of the chemical bond. the bonding wavefunction needs some reduction in amplitude.26) (6. 6.27) using the relations between overlap integrals. is why the bonding level is more favorable energetically. whereas the density for the antibonding is of course zero. Its node between the nuclei requires that the electron wavefunction undergoes a more rapid oscillation. Here the disadvantage of the antibonding wavefunction is more obvious. 6. Midway between the two atoms. as shown schematically in Fig. This explanation of bonding is now more correct. this argument has a fundamental flaw. so putting more electron density between the atoms must remove electron density from the rest of the wavefunction. where the electron can see both nuclei and benefit from both attractive potentials.5 Origin of the bond energy It is tempting. The normalization therefore decreases the total amplitude of the bonding wavefunction and increases it for the antibonding. from Eq. (6. the density for the bonding case is increased by a factor of 2 over that for two independent atoms. but dangerous. since there is only one electron. Normalization removes electron density from nearest the nuclei where the potential energy is most favorable. S. Comparing Eqs. and the complicated electron–electron interactions are absent. The first step to repair this embarrassing situation is to include the electron kinetic energy.27. The argument is that the higher density of electrons in the region between the atoms. The problem is that the electron density is normalized. The consequence is that the antibonding level sees an increase in electron density at the nuclei.1. and places it where the potential is modestly favorable. In contrast.

17:29:19 . The antibonding state is still favored slightly over the bonding state for most interatomic separations. 6. there is an annoying difference between Eqs. since in Eqs.18. Approximately2 % 1 δ−S c± 1± √ . Assuming the A-atom has the higher electronegativity as in Fig.6 Ionicity and electronegativity We return to the heteronuclear molecule. The charge 1 A further improvement can be obtained by optimizing the atomic wavefunctions at each interatomic separation. (6. reducing the kinetic energy somewhat. One should also consider the crystal field term in Eq. perhaps not surprisingly. and αc weights the charge density of covalent bonding that was considered in the previous section for a homonuclear molecule such as H2 . This can be done fairly well by changing the characteristic length for the exponential decay of the atomic wavefunctions.Energy 132 nucleus) and the electron kinetic energy (which increases when the electron is confined) need to be considered in detail.29) cB = ± √ 2 1 + δ2 where the new variable δ ≡ /(2|h|) scales with the electronegativity difference of the atoms. electron density will transfer from B to A. Here we expect a difference in electronegativity. 6.29. 6. so  = B − A = 0.28 and 6.12. so |cA | = |cB |. For atomic s-electron wavefunctions it is found. for both the bonding and antibonding states. The electron kinetic energy is increased a bit because the wavefunction is more confined.29 some terms of order S2 were neglected.26 and 6. but the potential energy becomes much more favorable. 6. What needs to be done is alter the shapes of ψA (r) and ψB (r) to minimize the energies of the bonding and antibonding states. which destabilizes the antibonding state and favors the bonding state. (6. 6.28 and 6. 6.1 The real situation points out the risk of relying too heavily on analytical models. and we expect unequal electron densities on each atom.27 are obtained directly from Eqs. The antibonding wavefunctions tend to lower their energy if the atomic s-states are expanded a bit. that the atomic wavefunctions are inadequate even for a qualitative explanation of the chemical bond.26 and 6. We seek to write the molecular charge density ρ = ψ ∗ ψ in the form ρ(r) = (1 + αi )ρA (r) + (1 − αi )ρB (r) + αc ρbond (r) .28) A = √ 2 1 + δ2 % δ+S 1 ± 1∓ √ . (6. it is possible to place more electron density at the nuclei.1b. and normalization was not done directly. 6. By making a more rapid decay with distance of ψA (r) and ψB (r) for the bonding state.1.27 and Eqs. Expressions analogous to Eqs. 2 When δ = 0. even for calculating bonding and antibonding states of a diatomic molecule. and the bonding state does become bonding. but here the result is not changed qualitatively. 6.30) where αi is a fractional charge transfer caused by the ionic character of the bond (the difference in electronegativity).

32) (6.00.31) where ρA = ψA∗ (r)ψA (r) and ρB = ψB∗ (r)ψB (r). for example. (6.56. 3 In practice. ZnSe.133 6.1 Molecular orbital theory of diatomic molecules redistribution upon bonding. 0.1 helps explain one approach. 0. 0. measurements on suitable heteroatomic molecules can be used to find . and CuBr. ZnSe. but interpretation requires some understanding of how to estimate the center of the band from the gap and bandwidth. αi = √ 1 + δ2 1 αc = √ . it is appropriate to assign an electronegativity to each chemical element. 0. using an element. say Ge. and the bond is mostly ionic. There are other types of electronegativity scales. GaAs. One way is to make spectroscopic measurements of the gap between the bonding and antibonding levels.06 Here the second row of numbers is the difference in Pauling electronegativity. 0. This is a purely covalent bond. ρbond (r). and the values of αi are GeGe. these measurements have been done on solids of similar crystal structures.71 1. the Phillips–Van Vechten [44. Assuming that these bonding tendencies are characteristics of atoms. Furthermore. but it is a challenge to account for bonding between different types of outer electrons about different atoms. 1 + δ2 (6. which was largely deduced from heats of chemical reactions.3 As discussed by Pettifor [5]. CuBr 0. offering a good basis for comparing their molecular orbitals. These molecules all have the same number of electrons. is defined with respect to the charge already present around two overlapping but independent atoms   ρbond (r) = 2ψA (r)ψB (r) − S ρA (r) + ρB (r) .32. Both ψA and ψB were assumed real in Eq. so the atoms have the same electronegativity. Figure 6.90. The general trends of the electronegativity scales tend to be in agreement. This is not so.37. GaAs. and then compounds of elements equidistant to the left and right of Ge on the periodic table. there is a fundamental problem with all electronegativity scales – they predict that all pairs of chemical elements will form compounds. making the series GeGe. 45] electronegativity scale was developed this way. αi > αc . 0. It turns out that δ . or differences between atoms.00. When δ = 0.31. 17:29:19 . and proper explanations of compound formation require concepts and physical properties that cannot be assigned to individual atoms alone. The Mulliken scale is based on ionization energies and electron transfers between atoms. αi = 0 and αc = 1. 6.33) and αi2 + αc2 = 1. When δ > 1 and the effects of electronegativity are dominant. If the bonding and antibonding levels can be measured for homonuclear molecules to obtain h.

This cannot be true.2. . eik·1a . . These phase factors are complex numbers. we need many more states. the sequence of phase factors . In this case when k = π/a aˆ . . we obtain a fully bonding wavefunction. One approach is to use the dimension of a large box as L. When k = 0. An important consequence of translational symmetry is that the distinction between bonding and antibonding states needs to be modified. ei4π . a translation vector of the lattice. Figure 6. which force the phase to be the same when the same atom is reached after the translation L. +1. because eiπ even integer = +1 and eiπ odd integer = −1. −1. . . has a similar form in a periodic solid when it switches phase between atoms in a chain as . and translational symmetry has a profound effect on the spectrum of electronic energies in crystals. which switched from +1 to –1 for an antibonding state in a diatomic molecule. and all atomic orbitals have the same phase. ei3π . The repeat length of the wavefunction at the bottom is simply a. Another imposes periodic boundary conditions over the length L. With many electrons in a solid. the electron density ought to be the same if position is shifted by a repeat distance of the crystal. +1.. .1 Translational symmetry and phase factors Translational symmetry is the fundamental difference between a molecule and a crystal. i. where a is a primitive translation vector of the lattice. . . the  phase factors are eik·nr = ei0 = +1 at every atom. −1. which involve taking products   like e−ik·nr e+ik·nr = +1. 4  This k sets the energy levels of the electrons. . but this is not a problem when calculating electron density or energy. It has a repeat distance of 2a. we have this vector k. varying  between these extremes of fully bonding and fully antibonding. At top is the wavefunction of highest energy. eik·4a . where L is a large length. eik·3a . by changing the value of k..2 Electronic energy bands in solids 6. an antibonding state. −1. . which is . using three different values of k. How can we construct enough states if we have the same type of atomic wavefunction ψ(na) at the different atom positions na along a linear chain? The trick is to multiply the ψ(na) by phase factors like     . of course. but it gives only one state for one electron. if k = π/a aˆ . . . +1. The atomic wavefunction. + 1. As one specific example. . for a short diatomic molecule. Neglecting end effects.e.2 shows the construction of three periodic wavefunctions on a linear solid of atom spacing a. eiπ . What is new is that we can now obtain a wide range of different wavefunctions. − 1. For k = 0. +1 . −1. ei2π . . 4 The number refers to an interval in k-space. and is a new quantum number to describe the electron states in a periodic potential. and it has the characteristics of a low-energy bonding orbital. The state and its energy depend on  which can take many values..Energy 134 6. An even easier case is k = 0. . we recover our antibonding state. as for the diatomic molecule. eik·2a . . 17:29:19 . . The interval has the approximate width 1/L. . This alternate switching of phase gives a valid antibonding state.

but this is easy) by na in an infinite crystal gives the same Hamiltonian. (6.2 Electronic energy bands in solids Box 6.135 6. Assume the Hamiltonian of the Schrödinger equation is translationally symmetric.1 Bloch’s theorem Bloch’s theorem justifies our use of phase factors. so H0 = Hna .34) and a shift of the potential energy operator (and kinetic energy operator. The wavefunctions. |0.

and |na.

. of the unshifted and shifted Hamiltonians satisfy H0 |0.

= 0 |0.

.35) Hna |na. (6.

= na |na.

(6. .36) H0 |na.

= na |na.

It turns out we can have a multiplicative factor.37) but by Eq. (6. 6. . but this is incomplete.34 Do the wavefunctions change with this translation na that leaves the Hamiltonian invariant? A naïve guess is that they are identical. λn for the translation na in the form |na.

= λn |0.

(6.38) The wavefunction must remain normalized. so 1 = na|na. . however.

= λ∗n λn 0|0.

For a long chain there is a nearly continuous range of k between 0 and π/a. A wavevector k = 1/8 π/a would give a wavefunction with twice the repeat distance. there is no problem with an imaginary phase factor because the electron density is ψ ∗ (x) ψ(x). The middle construction has a repeat distance of 8a. . and an energy that is closer to that of the fully bonding state. eik·r .40)  r ik·n e = λn . For the neighboring atoms in the middle construction. (6. 17:29:19 .39) 1 = λ∗n λn . which is always real. the wavefunctions generally overlap with the same sign. (Again.41) Equation 6. and Bloch’s theorem is A wavefunction of a periodic Hamiltonian is a product of a function with the periodicity of the lattice times a phase  factor of modulus unity.) The middle construction of Fig. (6.41 shows that λn must be a phase factor of modulus 1. 6. for both the real and imaginary parts. giving a spectrum of electron energies from fully bonding to fully bonding.2 with k = 1/4 π/a is only one possibility. (6. and is closer in energy to the bonding wavefunction than the antibonding.

. The top is for the wavevector k = π/a... at intermediate energy.7i. showing phase factors of .2 Tight binding So if k = 0 gives the most bonding state. +1.Energy 136 ε +i –1 +1 –i t Fig. –1.7i. 6. –0. Also for simplicity.. For simplicity assume that only bonds between an atom and its first-nearest neighbors have significant energy. showing phase factors of . –i. sometimes called “tight-binding” theory.+1.7–0.7+0..+1.2.In the middle.. +0.. continue with the example of a one-dimensional chain of atoms. 6. +1. +i . giving the energy per atom  = −a| + +a| H|0. and k = ±π/a + a gives the most antibonding.+1. The constructions of three electron wavefunctions are shown at right. showing phase factors of . +1 .7i.  This section explains a how do we calculate energies for the intermediate values of k? straightforward approach. Each atom in the chain makes a bond to a neighbor to its left and a neighbor to its right. –1 . and has two bond integrals equivalent to Eq. is for the wavevector k = π/4a. –0. 6.. +1.17. intersecting the real axis at ±1 and the imaginary axis at ±i..2 a At left is the unit circle on the complex plane... –i .The imaginary parts of the wavefunction are shown in gray.7+0.At bottom is for the wavevector k = 0..

is not of concern in the formal manipulations that follow. or even its origin. Simplifying  = −a|H|0. . (6.42) The form of the bonding interaction H.

+ +a|H|0.

a −ik· a +ik· a|H|0. .

+ e a|H|0.

.  = 2 cos(k · a) a|H|0.

43) (6. =e (6. a|H|0. The result for  depends on the bond energy between a neighboring pair of atoms.44) (6. 6.45) where Bloch’s theorem was used to obtain Eq.44. .

3 will show how equal intervals in k have equal numbers of electron states. Section 6. and there are more electron states near the top and bottom energies 17:29:19 . Figure 6. 6.2. but the phase of the wavefunctions  gives rise to a k-dependence of 2 cos(k · a).3a shows (k) for our one-dimensional chain. a function of energy. the intervals in k near k = 0 and k = π/a are wider. putting the lowest-energy (most bonding) state at  = 0.3b. ρ(). Although the energy intervals are the same. This allows us to construct the “density of electron states”.. using Fig.

More rigorously. compared to the number of k-points in an interval at larger k.3 a 0 ε 2〈a|H|0〉 k +/a –/a 0 ρ(ε) k +/a b 0 2〈a|H|0〉 c (a) Band structure of linear chain. Nevertheless. The tight binding model extends the concept of molecular orbitals to include the translational periodicity of a crystal.2.” and begins with free electrons in a box. the constructive interference between the atomic wavefunctions. with peaks where the dispersion is flat. raising their bonding energy. 6. (c) Plot of widths in k-space for the energy intervals in part b. 6.3c. The electronic DOS in three dimensions requires an additional concept concerning how  In 3D there are very few k-points in an interval the density of k-points increases with |k|. there is a suppression of the high density of states shown for ρ() at small  in Fig. 6.3 does illustrate key concepts of a band of energies (energy band) for electrons in solids. 6. • Bloch’s theorem gives a phase relationship between wavefunctions at atoms in a periodic solid. the inverse of the slope is 1/ sin ka. k · a. 6.’ It is called the “nearly free electron model. Fig.3c. the density of states is proportional to the inverse of the slope of the dispersion (k). and no more than one electron can occupy each quantum state. but within this box they behave as free electrons in a vacuum.3 A gas of free electrons in a big box This section presents a completely different view of bonding in metallic solids.3a. showing construction of the density of states ρ(). • There is a bonding integral for atomic wavefunctions at neighboring sites. there are more values of k near k = 0 and k = π/a (giving peaks in the electronic density of states. but we will not take this step here. 6. A histogram density of states (DOS) is shown in Fig.45. The electrons are confined to a box by a background of positive charges.  the increasing phase.) It is remarkable that a number of properties of metals are explained fairly well by assuming the electrons behave as a gas with no interactions between electrons and no interactions with the atomic structure. of our one-dimensional chain. They do obey the Pauli principle. 17:29:19 . (It later adds in perturbations from the periodic crystal potential. 6. with no connection to the previous model of ‘tight binding. ρ()). give the essence of the “tightbinding model” for electron energy bands in solids. These assumptions. and the approach taken in this section. so ρ() = csc(ka). near k = 0. One consequence is that in 3D. however.2 Electronic energy bands in solids 137 ε 2〈a|H|0〉 –/a t Fig.6. as in Eq. For the cosine function of Fig. (b) In each interval of equal width in . between the first neighbors decreases • With increasing k.

y. (6. is assumed a constant within the box (assume it is from 0 to L along each Cartesian axis). which are basic properties of metals.46) 2m ∂x2 ∂y2 ∂z2 This partial differential equation is solved by separation of variables. y. the free electron model is unreliable for transition metals and main group elements for which the tight-binding model is better. The result is kx = nx π/L. The quantum levels are separated by an amount kx = π/L. z)(x. ψ must be sine functions that go to zero at 0 and L. a typical equation for ψx (x) is − 2 ∂ 2 ψ(x) = ψ(x) . For a cubic box of edge length L. Particle in a box The “particle in a box” is a classic problem in quantum mechanics because it is one of three for which the Schrödinger equation has an analytical solution. y. A constant offset in potential energy is arbitrary. It can also explain electrical conductivity and electron screening of charges at impurities. We divide the 5 The other two being the harmonic oscillator and the hydrogen atom. and x y z isolating functions of an independent variable on one side of the equation. 2m ∂x2 (6. z. z) = sin(kx x) sin(ky y) sin(kz z) . Density of states in k-space The boundary conditions set the allowed values of kx because a half-integral number of wavelengths must fit in the box. y.50) where nx . z) = ψx (x)ψy (y)ψz (z).48) 2m With our boundary conditions. z) + V(x. so  3 2 2 (x. kz = nz π/L. 3. y. explicitly (x. z) .Energy 138 This free electron model sounds too simple to be true. nz are positive integers.3 by substituting the product ψ (x)ψ (y)ψ (z) into Eq.46.47) and all the energy is kinetic (k)2 . z) =  (x. ny . Beware. 6 The separations are done much as in Sect.5 Assume the potential box has infinite walls so the wavefunction does not tunnel outside the box and is zero at the edges of the box. the Schrödinger equation is   ∂2 ∂2 2 ∂ 2 + + − (x. V. so setting it equal to zero. (6.49) L = where the prefactor is a product of three normalization factors for the three dimensions. y. and likewise for intervals in the y. (6. but it can account for the elastic stiffness of metals like sodium. 17:29:19 .and z-directions. ky = ny π/L. where the solution is the product of independent functions in the coordinates x. and the separations6 lead to ordinary differential equations in one variable. though. The potential. and often it is. 6. (6. y.

48) √   dN  2m 1 2m L 3 . = d 4π 2 3 17:29:19 (6. volume in a spherical shell of radius k. three-dimensional k-space into small cubes of edge length π/L. 6.55) 3 dN (2m) 2 L3 1 2 .54) (6.4.56) . and assign one quantum state to each cube.   1 2 L 3 dN . = π 2 d m 2  π (6. so in three dimensions there are few states at small k. 6. π √ 2m/ (also from Eq.52) = πk dk 2 π This DOS increases as k2 . 6. There are more of them in a shell at larger k. 6. N(k) = π 8 3 L  or the “density of states” (DOS). as illustrated in Fig.48 for k2   dN 1 2m L 3 . Using Eq. In three dimensions the number of states in the positive octant of a sphere (here nx . (6. first find the To find the number of states at a given |k|. 6.2 Electronic energy bands in solids 139 ky p/L p/L k t Fig. Density of states in energy It is often more useful to know the density of states as a function of energy. but many at large k. nz are positive numbers) is the volume divided by the volume of each element 1 4π 3 1 (6.51) k 3 . (6.6.4 0 kx 0 Quantum volumes in k-space for a particle in a box of edge length L.53) = π 2 dk 2  π and also from Eq. rather than k. ny .48 d/dk = 2 k/m dN 2 k 1 2m = π 2 d m 2  and with k =  3 L .

The highest energy electrons will be at the highest value of k. 6. so the electronic DOS.4 Electronic heat capacity The breadths of energy bands are often of order 1 to 10 eV – far wider than the spread of thermal energies.59) kF = 3π 2 n This gives kF from the density n. and hence to its entropy at finite temperature.62) . At low temperatures. and those above F are unoccupied. (This is strictly true at T = 0. leaving holes within kB T below F . the electrons fill the lowest states first. To a first approximation. (6. so the free electrons contribute to the heat capacity of a metal. F 17:29:19 (6. which are at low values of k.60) (6. (6. temperature does not alter the occupancy of electronic states.58) k 8 3 F π 3 π  1 3 . more electrons must occupy states of higher energy. The total number of electrons in the box is (cf.5). Heat is absorbed by these electronic excitations.57) 2π 2 3 √ This gradual dependence of ρ() on energy as  is characteristic of a free electron gas (see Fig. With increasing electron density. and the states below the Fermi energy are fully occupied. such as 0.61) 6.025 eV at room temperature. 2m 2  2  23 F = . there are no electrons in states with k exceeding kF . At low temperatures. Eq. 6. is 3 (2m) 2 L3 1 ρ() = 2 . which then gives the corresponding Fermi energy F (kF )2 . (6. n ≡ N/V.Energy 140 Two electrons can be placed into each state if they have opposite spins.2. however. Fermi energy What is the highest energy of an electron in the box? The answer depends on the density of electrons. ρ() ≡ dN/d. 3π n 2m F = (6. which we call kF in honor of Enrico Fermi. thermal energy can excite electrons to energies around kB T above F . The total energy of thermal electronic excitations is proportional to the fraction of electrons that are excited (approximately kB T/F ) and the average energy of the excitations (proportional to kB T) Uel (T) kB T kB T .) To a second approximation. but all states with k < kF are filled.51 with two spin states per cube)     2 4π 3 L 3 π 3 L 3 3 nL = = kF .

63) A more accurate calculation [46] gives the electronic heat capacity Cel (T) = π2 ρ(F ) kB2 T . constructively interfering for the lower-energy bonding state but destructively interfering for the higher-energy bonding state. and a free electron gas. The electron DOS curves of real materials show all features of Fig. Although this is much less than the vibrational entropy. and this seems to be approximately right for most metals. This ρ(F ) is assumed constant within a few kB T around F . Molecular crystals tend to have electronic states 17:29:19 . Vibrational entropy and configurational entropy dominate the thermodynamics of metals over a wide range of temperatures.000 K. At temperatures of 1. and gives a distribution of energies between the fully bonding and the fully antibonding levels. it is a well-known result that the phonon heat capacity increases at T 3 . 6.2 Electronic energy bands in solids The heat capacity is obtained as Cel = ∂Uel (T)/∂T Cel (T) 2 kB2 T.65) Sel (T) = T 0 π2 Sel (T) = (6.2. (6.64. 6.5 presents electronic DOS curves for the systems considered so far – a diatomic molecule. The electronic heat capacity is predicted to increase linearly with temperature by Eq. The phase of the atomic wavefunctions. 3 and Sel (T) also increases linearly with T. it is the difference in entropy between phases that is important for phase stability. but the electronic entropy is important at low and high temperatures.5 From molecules to metals Figure 6. The tight-binding model of a periodic solid gives a distribution of phase interferences between the atomic wavefunctions in a crystal.5 in varying degrees. At low temperatures. the electronic entropy of metals is typically tenths to one kB /atom (or more if ρ(F ) is large).64) where ρ(F ) is the electronic DOS at the Fermi level. give the two states in the energy spectrum of the diatomic molecule. 6.66) ρ(F ) kB2 T . a tight-binding crystal. The free electron gas is a somewhat different concept because the electrons are assumed independent and noninteracting. so sometimes electronic entropy plays a role in the phase stability of metals. 3 (6. F (6.141 6. so at very low temperatures (under a few K). Sharp spikes in the density of states are characteristic of crystals that do not have facile electron transport between atoms. the electronic entropy is more important than the vibrational entropy. and we also assumed that the electronic DOS is unchanged with temperature. The electronic entropy can therefore be calculated easily as  T Cel (T  )  dT .

2. that the electrons are all in states like the low-energy bonding state of a molecule in Fig. a crystal structure has an electronic DOS with more states at low energy. for example. Pressure can drive crystals into a metallic state.6 Fermi surfaces The total electronic DOS determines the bonding energy of a solid. Such details of the electronic DOS are often what gives specific materials their special character. and many oxides have local bonding giving discrete local states for electrons. although entropic differences cause phase changes with increasing temperature. If.5 ε Electronic DOS for a diatomic molecule. perhaps arranged as a molecular crystal. and often its thermodynamic stability.Energy 142 ρ(ε) molecule ε ρ(ε) tight binding ε ρ(ε) free electron t Fig. This. less important than the total electronic DOS is often the DOS near the Fermi surface. we want the electrons to do something interesting in response to a stimulus. like the molecules of which they are composed. compared with an alternative structure. and this often means changing their states. 6. The phase diagrams of materials depend on such energy differences. In essence.5. transition metals have d-electrons that contribute a spiky electronic DOS function on top of the broad s-electron background. but transition metals also have s-electrons that make a broad DOS. for example. 6. Nevertheless. 17:29:19 . For electronic properties of materials. and a free electron gas in three dimensions. it will be more stable energetically. there are electronic states just above the Fermi level. with a parabolic shape. is the origin of electrical conductivity. however. for example when atoms are forced together by pressure. The electrons remain in their states because it costs too much energy for them to change. Suppose. These sharp peaks become broad with increased overlap of the wavefunctions at neighboring atoms. approximately. with bonding and antibonding levels. 6. however. Now apply a small electric field to a collection of such molecules. For a metal. a periodic solid in the tight-binding approximation (the shape depends on the nature of the wavefunctions and the dimensionality). It was mentioned earlier that alkali metals have a free-electron-like electronic DOS. and it takes only a little energy for the electrons to change into these states.

electrons in states near the Fermi level can change their state in response to internal disturbances such as an impurity or a phonon. The F lies in the band gaps of insulators and intrinsic semiconductors. For a metal. even at very low temperature. semiconductor. The gray shading indicates states that are filled.6 gives a general picture of insulators. F .phys. 6. and this is reflected indirectly in the DOS. 6. all aligned with the same Fermi energy.6. and metal. but please take at least a quick look at the University of Florida website to see the structures in three dimensions.ufl. 6. white indicates empty states. but the gap of an intrinsic semiconductor is narrow enough so that temperature can excite some electrons from the lower valence band to the upper conduction band. Usually Fermi surfaces are anisotropic because chemical bonding in many crystals is directional.6 ε εF Electronic DOS for an insulator. The point to notice is the energy required to excite an electron from an occupied state to an unoccupied state. but many are well known.2 Electronic energy bands in solids ρ(ε) insulator ε ρ(ε) semiconductor ε ρ(ε) metal t Fig.7 Such excitations are required for moving electrons and electrical conductivity. 6. The Fermi surfaces of the elements are shown in the website: http://www. semiconductors. These Fermi surfaces are worthy of study if the electronic properties of materials are of interest to you.7 Band structures of alloys Extending the concept of electronic energy bands from pure elements to alloys is sometimes possible in an easy way.2.143 6.edu/fermisurface/ Examples of some Fermi surfaces are shown in Fig. ρ(). The three-dimensional shapes of real Fermi surfaces are quite a bit more complicated than the schematics in Fig. The solute atoms in an 7 Of course the Pauli principle forbids electrons from entering a state that is already occupied. and metals. The free electron gas model already ignores the chemistry of the elements. Figure 6. there are always states that are accessible by thermal excitation. For a metal. and all properties depend on the electron density.7. or impurity (dopant) atoms can be added to place states in the gap and reduce the energy required to change the states of electrons. 17:29:19 . Temperature can assist these excitations if the band gap energy is less than several kB T.

such as p. 48]. The hexagonal crystal structures of Zn and Cd at right are evident in the anisotropy of their Fermi surfaces. or outside in the case of “hole pockets” [47. The occupied states may be inside the surfaces. The alkali elements at left are reasonably well approximated as a free electron gas. 17:29:19 . Note that isoelectronic elements in the same column tend to have similar Fermi surfaces. Copyright (2000) by the American Physical Society.or d-electrons. The different shaded “sheets” of the Fermi surfaces may correspond to bands made from different types of electrons.Li Be Na Mg K Ca Sc Ti V Cr Mn Rb Sr Y Zr Nb Mo Tc Ru Rh Cs Ba Lu Hf Ta W Re Os Ir t Fig. but the transition metals are considerably more complicated.7 Fe Co Ni Cu Zn Pd Ag Cd Pt Au Hg Part of the periodic table of Fermi surfaces of metallic elements. 6.

It is straightforward to count the total number of valence electrons added by the solute atoms. 6.1 Metallic bonding If we calculate the average energy of the free electron gas.8 Even if the band structure is distorted by alloying. there are separate bands for ↑-spin and ↓-spin electrons. 17:29:19 . (Some thought may be needed about which valence electrons are available for donation – s-electrons may be more available for donation than d-electrons. Surprisingly. These plots of magnetization versus electron-to-atom ratio for Fe–Co and other 3d transition metal alloys are the “Slater–Pauling curve. a central question is how much the electronic bands change with alloying.” which has the electronic DOS at the bottom of Fig. does alloying cause the band structure to simply fill with electrons. Large differences in electronegativity or atomic size usually cause large distortions in the alloy band structure. In this case. At an intermediate composition. In such cases the e/a ratio is again a useful parameter. Specifically. when the chemical elements are similar. and there is an increase in the magnetic moment of the alloy. the ↑-spin states fill more rapidly because of their higher density. Although it is often not quantitative. This is a mechanical response. parameterization with the e/a ratio is easy and it can sometimes provide qualitative understanding. When bcc Fe is alloyed with Co. or is the band structure also distorted? Distortions of the band structure of an A–B alloy are expected when there is a big difference in the bonding between A–B pairs and the A–A or B–B pairs of neighbors. around c = 0. On the other hand. with more electrons in the ↑-spin band by definition.) In such cases. For more detailed band structures such as obtained with the tight-binding model or with modern electronic structure calculations. the states for the ↑-spin 3d electrons become full. 6.” It is explained by the filling of “rigid” electron bands that do not change their shape. we can calculate pressures needed to reduce the volume of the electron gas. and assign them to the electron gas. the distortions can be small.3. and we may assume a rigid band structure is filled with electrons during alloying.3 Elastic constants and the interatomic potential alloy are assumed to donate (or remove) electrons to the gas. sometimes the distortions are not essential to understanding general trends. The e/a ratio serves a role like that of chemical composition.145 6.6 (examples are 8 Because Fe is ferromagnetic. the ρ() depends on the overall electron-to-atom ratio of the alloy. for example. such as how the magnetic moments of 3d transition metals change with alloying. so the extra electrons go into the ↓-spin states. but it can accommodate results from more than two elements in the periodic table. the rigid band model can sometimes explain rather delicate behavior. and this “e/a” ratio is commonly used as an axis when plotting properties of alloys. for example.3.3 Elastic constants and the interatomic potential 6. using the electrons pertinent to the band structure that is being filled. Changing the chemical composition alters ρ() as electrons are added to the gas. and the magnetic moments decrease. and how this energy depends on volume (it goes as V −2/3 ). This section calculates the bulk modulus of a “free electron metal. as for understanding the cohesive energy of d-electron metals (see Problem 10. the e/a ratio is again a valuable parameter for organizing properties of the elements and their alloys.3).

B=− N 5 V 3 2N B= F .61. The energy of the electrons increases as they are confined to a box of smaller volume. (6. 6. giving rise to a mechanical pressure.72 d  F  2 . and this success gives hope that some electrons in metals do behave as a gas in a box. 6. P. 3V 17:29:19 (6. Electrons with states near k = 0 have the lowest energy. 6.77) . the essence of F fraction of Eq.71) so with Eqs. 6.69 and 6. 6. 6.71   2 2 F −1 − .68 is identified in the first 3 (6. 5 Pressure. 6.73) B ≡ −V dV and with Eq. The kinetic energy of an electron in the gas varies with wavevector. it is straightforward to show P=− dF 2 F =− .72) F . 5 F 2π 2 3 (6.70) (6.61. dV From the dependence of F on n = N/V in Eq. is a response from the free electron gas when its volume is changed = d .Energy 146 the alkali metals in Fig. It is remarkable that a macroscopic property of a real metal can be calculated with reasonable accuracy with the simple assumption of an electron gas.70 2N (6.57 as the probability distribution for computing an average  F =  ρ() d . The average energy  is calculated using the DOS of Eq.7).68) −3/2 and using the expression for F from Eq. 6. (6.76) (6.69) N F .74) B = − NV 5 dV V   1 dF F 2 . dV 3 V (6. 5V The bulk modulus is defined as the pressure from a material with a fractional change in volume dP . (6.75) B = − NV − 2 + 5 V dV V P= Using the volume dependence of F from Eq. and those near the Fermi level the highest.67) 0 (2m)3/2 3 2 5/2 = L  . (6.

or more generally the covalent bonding in numerous other materials. 6. Typically the potential decreases as the atoms come together.147 6. increasing B. where they no longer see each other. 6. Equation 6. The force is F=− d 2/3 −1/3 (r) . such as transition metals. and these peak-like features rise like mountains over the free electron DOS. and the nuclei are not fully screened by the electrons. is popular   M = C e−2αr − e−αr . loosely called a “hard core repulsion. comprising two exponential functions. such as sodium or potassium. owing to the attraction of a chemical bond.3 Elastic constants and the interatomic potential From the electron density. This force. at the low 17:29:19 . Although there is no rigorous justification for either M or L−J . the d-electrons (up to 10) have a split DOS from bonding and antibonding levels. however.6.80) where N is the number of atoms in the crystal. N 2/3 is the number of atoms in a plane.” and an attractive contribution that has a more gradual variation with distance.2 General interatomic potential A simple but important relationship between an interatomic potential and elastic modulus is given here. Interatomic potentials go to zero as the atoms are separated by long distances. For alkali metals.2).8. For transition metals. Here an elastic modulus is calculated for the case of a cube of a simple cubic crystal with the force applied along a Cartesian axis. the energy rises steeply because the electronic kinetic energy rises as the electrons are squeezed into smaller volumes.78) as is a Lennard–Jones “6–12” potential L−J = A B − 6 . Interatomic potentials are typically represented as the sum of two functions. In fact. and the second is the attractive. which has been measured for elemental metals and many alloys. the s-electrons have a free-electron-like band structure. we can calculate F .3. in many applications it is possible to use either of them by suitable choice of parameters. A Morse potential. the success of this free electron gas model in predicting B is impressive (see Problem 6. but the DOS is low because it accommodates not more than two s-electrons per atom. n ≡ N/V. r12 r (6. N N dr (6. (6. and N −1/3 is the reduced force associated with lining up the atomic springs in series. 6. F. To accommodate these d-electrons. Usually this more direct covalent bonding is stiffer. we need a different approach to the interatomic potential. bonding contribution. For other metals. They are shown in Fig. At very small separations.79) In both cases the first term is the hard core repulsion. many transition metals have electronic band structures that are a superposition of the bottom two diagrams of Fig. is zero when the interatomic separation is of minimum energy. On the other hand.77 can then be used to obtain the bulk modulus. the electrons form chemical bonds that are not well described by this free-electron-gas model.

Energy 148 0. and L−J = r121 − r16 . consistent with Hooke’s law.1 Morse −0.3 −0. points of the (r) curves of Fig.84) Y= . Here the Morse potential was shifted right.0 −0. 6. 17:29:19 .9 Now consider the increase in energy as the interatomic separation is changed from this equilibrium r0 . For a small change r (r0 + r) (r0 ) + d(r) 1 d2  (r)2 .8.80 F(r) = − A d2  r .83) using the area A and length L for the cubic crystal. so there is a lateral contraction when a block of material is elongated. (a) Here M = e−2r − e−r . r + dr r0 2 dr2 r0 (6.2 −0.1 Lennard−Jones Morse F(r) F(r) 0. The ratio of lateral 9 At finite temperature the equilibrium separation may increase a little.8.2 0. Y.2 0.0 1 2 3 r 4 5 6 0 1 2 b 3 r 4 5 6 Functional forms of Morse and Lennard–Jones potentials. so the term linear in r vanishes (r0 + r) = (r0 ) + 1 d2  (r)2 .1 0.8 Lennard−Jones −0. and pulls back against an extension to positive r. The restoring force is linear in r.81) The zero-force condition at r0 is d(r)/dr|r0 = 0. The constant is a Young’s modulus. defined as d2  (6.2 t Fig. and the Lennard–Jones potential was shifted in r and rescaled in r. L dr2 r0 (6. Usually there is a tendency of a material to maintain a similar density during deformation.82) Calculating the force as in Eq. owing to the skewness of the potentials of Fig.3 0 a 0. dr2 r0 A (6. 6. 6. L The restoring force is linear in the displacement.85) F = −Y r . (b) By choosing the appropriate parameters and rescaling the two axes. 2 dr2 r0 (6.1 −0. 6. This zero-force condition gives the equilibrium separation r0 of the adjacent atoms for the temperature T = 0. these two potentials can be made to look similar.

y). (b) Nonuniform displacement of a square with varying u(x. all parts of the square must undergo the same translation u = ux xˆ + uy yˆ = constant. This does not necessarily create strains. This is an example of a shear strain. 17:29:19 . ∧ ∧ u(x. 6. Bonds also resist shear forces. even if the basic idea of Hooke’s law remains valid. Notice the distortion of the square.4 Linear elasticity 6. there are no lateral forces. 6. so some of its internal volume elements are translated with respect to the laboratory frame by the vector u. (Incidentally. which shows a displaced but undeformed square. the problem is much more complicated than for a row of springs in a simple cubic crystal because the applied forces are not always parallel to the bonds.6. ν.y) y u = const a ∧ y b ∧ x Δy ∂ux ∂y Δy φ φ Fig.y) x + uy(x.4 Linear elasticity 149 contraction to elongation is the Poisson ratio. The effects of the variations of the components of u are labeled. 6.9 ∂y Δx ∂uy ∂x c t ∂uy d Δ x ∂ux ∂x (a) Uniform displacement of a square with constant u. 6. We begin by defining the strains with the help of Fig. The square is undistorted.9. however. and our set of elastic constants needs to be expanded. typically ν ∼ 1/3. (c) After translating the distorted square of part b back to overlap the initial square. the angle φ is equal to ∂ux /∂y. The elastic energy distribution in a material can be calculated once the elastic field is known. but the converse is not necessarily true.1 Strains An elastic field has energy that is important for thermodynamics. as illustrated in Fig. For the square to move without deformation. and stresses (forces per unit area [pressure]). In general.9a.) (d) Example of a uniform dilatation. Suppose a large elastic body is deformed significantly. where ux and uy are constants. showing the effects of the variations of the components of u as labeled. For our simple bonding model. and ν = 0.y) = ux(x. The mechanics of elasticity is based on strains (fractional changes in length [dimensionless]).4.

10 depicts six shear stresses. The six shear stresses therefore obey relationships such as σxy = σyx .10.2 Stresses Stresses are applied to a material by making contact with a surface and applying a force to it. ∂x ∂y ∂z (6.direction . and the second for the direction of the force. y.9d. as in Fig. 6. the strain is dilatational.Energy 150 Strains occur when ux and uy vary with position. where the numbers {1. 6. and how the indices are mixed for the three shear strains. When ux varies with x and uy with y. The stress σxy is a shear stress with the force along yˆ that pulls on a surface with normal along xˆ . yy = . as shown in Fig.88) 6. Opposing vectors are assumed on the three opposing faces. A normal stress such as σxx describes a force along xˆ applied to the face with normal along xˆ . but when ux varies with y and uy with x as in Fig.4.9c. 3} are used interchangeably with {x. defined as       ∂uy 1 ∂ux 1 ∂ux 1 ∂uy ∂uz ∂uz + .9b. In three dimensions the displacement vector is u(x. (6. y. and three normal stresses. 17:29:19 . The notation is shown in Fig. Figure 6. It is also necessary to apply a counteracting force (along −ˆx) to the opposite face of the sample to prevent movement. γij . 6. the strains are shear strains. A shear stress has its force perpendicular to the normal of the face. the first for the face to which the force is applied.direction . but counteracting stresses are needed to avoid rotational acceleration. γxy = 2 ∂y ∂x 2 ∂z ∂x 2 ∂z ∂y (6.86) The three tensile (if positive) or compressive (if negative) strains are defined as xx = ∂uy ∂ux ∂uz . σface. The notation for a stress uses two indices. 6. γxz = + . σ33 σ32 σ31 z^ σ23 σ13 σ12 σ 21 x^ t Fig. γyz = + .87) Notice the consistency of all indices for tensile or compressive strains. 6. ii . z) = ux xˆ + uy yˆ + uz zˆ . 2. There are only three independent shear stresses. with nine stress vectors labeled in the notation σface. z}.10 y^ σ22 σ11 Cartesian coordinate systems. zz = .

the number of independent elastic constants is only three. an important relationship for the bulk modulus is 1 (6.96) For low-symmetry crystals.89 becomes σ11 = C11 11 + C12 22 + C12 33 (6. Iron. C11 and C12 . so Ciiij = 0. with the following convention { 1.10 For a cubic system. yz.89) j Equation 6. 3 A cubic crystal generally does not have isotropic elasticity. The anisotropy is parameterized by the ratio.95) σ12 = C44 γ12 (6. For a cubic crystal. the elastic behavior is more complicated. which give a linear relationship between the stresses and the strains through Hooke’s law σi = 6  Cij j . A. 2. Cxxyy . zz. 6 } {xx. and this is roughly true for aluminum. C11 − C12 (6. and crystal symmetry reduces them further. C12 .94) σ31 = C44 γ31 (6. 6.3 Elastic response In linear elasticity theory.4 Linear elasticity 6. xy} (6. for example. 5 .89 is more complicated than it looks because the subscripts are not single Cartesian indices.92) σ33 = C12 11 + C12 22 + C11 33 (6. zx. but rather pairs of them. if A = 0 there are only two nonzero elastic constants. 6. 6.151 6. the relevant material properties are the elastic constants Cij . but also (2) the fact that the sign of ij cannot alter σii (positive or negative shear forces perpendicular to ˆi should cause the same effect). In particular. Cyzyz }. 3. sometimes requiring more parameters than can be measured.97) (C11 + 2C12 ) . (6. 17:29:19 . is distinctly anisotropic.98) Incidentally.91) σ22 = C12 11 + C11 22 + C12 33 (6. Eq.90) There are 36 constants Cij in Eq.89. however. The contracted indices i and j in Eq. yy. however. For cubic crystals.89 are ordered pairs of Cartesian indices.4. but symmetry across the diagonal reduces this number to 21.11 10 Symmetry requirements include (1) the fact that the three Cartesian axes are equivalent. 4. involving the three elastic constants B= A≡ C44 . 11 For bcc tungsten it is approximately true that A = 0.93) σ23 = C44 γ23 (6. a bcc crystal is often stiffer along its body diagonal than along a Cartesian axis. the set {C11 . The elastic constants Cij actually have four Cartesian indices. C44 } ≡ {Cxxxx .

This often requires special methods of partial differential equations that can account for the shape of the body and the boundary conditions of free 17:29:19 . the springs in regions of largest variation will have the largest stretching or compression. The energy of this elastic chain is 2 N  1  Eel = cx (6. i. 2 i 1 Eel = cx 2 L  du dx 2 dx . meaning a uniform displacement. For example.4.100 as the strain. In general. If u were constant. On the other hand. we could put all the elongation in one spring. however. giving the energy cx /2 (δx)2 . and considering all six strains j . if a material with internal elastic energy from a misfitting precipitate is simultaneously subjected to an external applied load.100) 0 where the transformation to a continuum was done by assuming the solid of N atoms (masses) has the length L = Na (a is the interatomic separation). 6. Within its elastic range. a bar of uniform thickness will elongate uniformly when pulled. Suppose we have a chain of springs and masses. For example. If there is a variation of u(x) with x. the general expression for the elastic energy per unit volume is 1  Cij i j . A most important aspect of elastic energy is that it depends on the square of the distortion in an element of volume. For real 3D solids. [ σij dij ] dv.. There is no energy cross-term in linear elasticity theory. using Eq. and a subtle but important consequence is that elastic energies from internal and external loadings are additive.Energy 152 Linear elasticity theory considers only small strains. the elastic energy of a 1D solid is simple to calculate. stretching our chain of elastic springs will produce a uniform distortion of all springs. spreading this distortion over n springs gives more springs that are distorted. 6. The total elastic energy is reduced by a factor of n by spreading out the distortion. 2 6 Eel = 6 (6. there is no change in lengths of the springs. For this reason. but each has the energy cx /2 (δx/n)2 . if we stretch the ends of the chain by the length δx. 6.99) u(xi ) − u(xi−1 ) . Another important consequence is that strain fields tend to spread over long distances in materials.101) i=1 j=1 using the contracted notation of Eq. 6. to calculate the elastic energy we need to calculate the strain fields inside a material. We recognize du/dx in Eq. and by assuming that the spatial scale for variations in u(x) is long compared to a.e. 6.101 is obtained by integrating force times distance (strain) in a small volume. and suppose we move this entire chain by the distance u along ±ˆx.90. Equation . the elastic energies of both the precipitate and the external load can be calculated independently.4 Elastic energy Without the complexity of stress and strain tensors (of rank 2) and elastic constants (of rank 4).89 for σij . (6.

The elastic energy for this “stress-free strain” is zero. the surrounding material can flow out of the way. 6.6. A diagram with definitions of variables is shown in Fig.1 Eshelby cycle and elastic energy First consider a hypothetical series of steps to account for the elastic energy. full calculations may have to be done numerically. and is not hard to calculate. This is the largest energy of the entire sequence. This step creates some surface energy of the small sphere and the hole. but remains within a surrounding matrix of untransformed material.5. 6. This problem is unique to phase transformations in solids – in a liquid or gas. • Expand the small sphere to its new volume. The problem of expanding a small sphere inside a large spherical body is one that can be solved analytically. 17:29:19 . Neither the body nor the extracted sphere has strains. 6.5 Misfitting particle For phase transformations in solids. t Fig. The energy cost is positive. in addition to the strains within itself. it is important to know the elastic energy when a volume of material changes its shape. so the energy cost in this step is zero. We want to know the elastic energy from these strains. 50]. In this third step. forces are applied over the surface of the new sphere to compress it back to its original size. and remove a small sphere from inside it. The new misfitting particle generates long-range strain fields in the surrounding matrix. but surface energies are ignored.11. but it is too big.5 Misfitting particle 153 or constrained surfaces. which is the natural volume of the new phase. 6. and gives insight into the strain energy of solid–solid phase transformations.11 Expanded sphere inside a large spherical body. sometimes called an “Eshelby cycle” [49. • Start with a uniform body of material. • The expanded sphere needs to be fitted back in the old hole. There is no internal stress if the phase transformation occurs in free space. so there are no strains and no elastic energy. Especially when the problem includes two phases with different elastic constants.

but is compressed along rˆ and expanded perpendicular xx + yy + zz = 3 12 There is a boundary condition of a free surface. but less in magnitude than for the previous step of compressing the sphere. (6. φ). convert to Cartesian coordinates. The surface forces that compressed the sphere are now removed. 6. (6. (Typical values are about a third of the compression step. 6. and it turns out that the net volume of a shell remains the same.104) and in general δij 3xi xj − 5 . however.e. and is the subject of Chapter 21.2 Strain fields for misfitting sphere in infinite matrix For a misfitting sphere. especially near the sphere. however. they expand their surface area. reducing its internal elastic energy. but we can understand it by considering shells of material around the sphere. giving a positive elastic energy. θ . 6. It fits. though. The energy cost is negative. θ . the outer radius of the body. and for isotropic elastic constants all displacements are radial. An important result from Eq. gives xx = ∂ux 3x2 1 = 3− 5 . and differentiating the component ux with respect to x. This result (Eq. 17:29:19 .103) r r x + y2 + z2 r Finding the common denominator of (x2 + y2 + z2 )3/2 . recognizing that r2 = x2 +y2 +z2 x 1 y z  u(x.105) r3 r where i and j denote Cartesian coordinates x.102) To calculate the strains.106) is not intuitive. but away from the outer surface of the body12 ur = δv .102 to ensure no radial stress at R. ∂x r r (6.105 is that the total dilatational strain is zero ij = 1 3x2 + 3y2 + 3z2 − =0. the displacement field is naturally described in spherical coordinates u(r. this step must give a reduction of energy. u(r. x3 ≡ z. The sphere expands a bit. and x1 ≡ x. Here δij = 1 when i = j.Energy 154 • The compressed sphere is replaced into its original hole. z) = 2 + x + + y + + z . but the surrounding material is strained. x2 ≡ y. i.) This energy change is the most difficult to calculate. and 0 otherwise. Near the sphere. These shells are compressed radially when the oversize sphere is inserted. By shifting outwards. Each volume element in the shell keeps the same volume. y.. y. 4π r2 (6. 6. Overall. even though they change shape after the transformation. depending on the elastic moduli.106) r3 r5 meaning that the volume elements in the matrix do not change their volumes. or it would not occur. (6. z. The strain in the matrix near the misfitting sphere is pure shear. φ) = ur rˆ . An additional term proportional to rR−3 is added to Eq.5.

13 The strain has a component  . x1 k2 (6.107 shows that the stress on the surface of a spherical shell of radius r pushes it back towards the origin.5 Misfitting particle 155 to rˆ . 17:29:19 . This contribution to the strain energy scales with the bulk modulus. Actually. 6. so k1 x2 = .) In spherical coordinates.12 1 2 2 k2 x2 1 2 2 k1 x1 k2 = k1  x2 x1 2 . but also  and  rr θθ φφ because a differential volume is sheared when a spherical shell expands. Eq. so there are compressive strains inside the sphere. In Fig. Around the interface of the misfitting sphere. without shear contributions. B. 15. (6.12 illustrates how. π r3 (6. 6.3 Elastic springs in series Figure 6.17c. When the misfitting sphere has a larger volume than the original matrix. the stiffer spring (with large spring constant k2 ) has the smaller displacement (x2 < x1 ). when a force is shared by two springs in series. inside the sphere the strains are pure compressive.108) The energies in the two springs are in the ratio E2 = E1 t Fig.13 The transformed sphere is constrained to have a smaller volume than is natural for this new phase. and their displacement ratio is obtained from F = k1 x1 = k2 x2 . The displacement of the soft spring is much larger than that of the stiff spring. for example. (This can be accomplished by the shears shown in Fig. (b) Force applied to the series springs. much like the membrane of an inflated balloon.5. but perhaps less intuitive is the fact that it also has the lower elastic energy. obtaining the stress from this displacement involves another differentiation with respect to r.107) Note that this result depends only on the shear modulus μ. and the stress near the misfitting sphere turns out to be σrr = − μ δv . and not on the bulk modulus B. This is intuitive.109) (a) Two springs in series – a soft spring at left and a stiff spring at right. 6. the outwards elastic forces from the sphere are counteracted by the forces from the surrounding matrix.6. 6. the springs share the force F.12b. We can gain insight into this problem by considering two springs in series.

6. E1 k2 (6. (6. The fractions of energy in the two springs. using Eq.111) The total energy in the system is the sum of the energies in the two springs Etot = 1 1 k1 x12 + k2 x22 . 6.Energy 156 and using Eq. 2 2 (6.112) Rewriting Etot in terms of the displacement of only the stiff spring x2 .113) 2 k1 2 and multiplying through by the fraction 1 = (C2 /C2 )2   1 k1 + k2 2 .108. are C1 = k2 k1 + k2 and C2 = k1 .110) which is the same ratio as for the displacements in Eq.108 for the ratio x2 /x1 E2 k1 = .108  2 1 1 k2 Etot = k1 x22 + k2 x22 . 6. k1 + k2 (6. C1 and C2 .

2 k1 (6. and it is different for particle shapes other than spheres. Emx 3B (6.110). The constant 4/3 is not simple to calculate. (Here B is the bulk modulus and μ is the shear modulus. 3B + 4μ Cpt = (6.117) 17:29:19 . where elastic energies are calculated to be 1/2B 2 for compressive (or tensile) strains in the sphere.5.) Comparing to the preceding description of the energy partitioning between springs in series (Eq. it is not surprising that the ratio of the elastic energy in the spherical particle to that in the matrix proves to be inversely proportional to their relevant elastic constants Ept 4μ = .115 4μ . Etot = k1 (C1 x2 )2 + k2 (C2 x2 )2 . and 1/2μγ 2 for shear strains in the surrounding matrix. and μ is a property of the infinite matrix with its shear strain. 6. The partitioning coefficients for the infinite matrix and spherical particle are the fractions consistent with Eq.4 Energy of misfitting sphere in infinite matrix We return to the more complicated problem of the 3D misfitting sphere in an infinite matrix of Sect.115) where B is a property of the sphere with its compressive strain.5.2.116) Cmx (6.114) 6. Equation 6. 6. 3B + 4μ 3B = .115 shows that the partitioning of elastic energy between the particle and matrix depends on the relative values of B and μ. which have comparable importance. 6.

6.118 is Etot = Etot = 2 μ B (δv)2 .2). 6. and E(∞) = 3/4. the particle shape approaches a long cylinder. Nabarro described the ellipsoid of revolution by the ratio y/R.6. 21. it becomes a sheet that separates two regions of matrix. and Eq. so E(1) = 1.116 to 6.114.118 simplifies to 2 μ (δv)2 (6. these three layers impose no stresses on each other and can relax their strains entirely. these energy partitioning fractions can be used to calculate the total elastic energy B 2μ (6. Cmx = 1 and Cpt = 0. Nabarro considered ellipsoids of various shapes. Now consider the effects of surface energy. it often finds an orientation where the elastic distortions are confined primarily to regions near its edge (see also Sect. When y = R. the strain energy of Eq. For y < R the particle approaches a flat disk.120) 3 v for a spherical particle. but it is well beyond the scope of this text. if it is multiplied by a factor E(y/R). 6. which goes to zero for a very flat disk. A thin plate has the maximum surface area per atom in the particle. With this simplifying assumption. with the assumption that the matrix and particle had the same elastic constants. Since the two interfaces are free to slip.3. Summarizing for this case of a stiff particle in an isotropic matrix Etot = Edisk  Eneedle < Esphere . and y is half the length along the axis of revolution. 6.. 2v 3v where the first term gives the elastic energy in the spherical particle and the second the elastic energy in the infinite matrix. When y  R. (Recall that the strain is δv/(3v). In the limit where the particle is infinitely flat and wide.5 Misfitting particle 157 so that Cpt + Cmx = 1.e. thin particle is embedded entirely within a matrix. so the particle is incompressible and contains no elastic energy). B  μ. As for Eq.5. leading to complex shapes 17:29:19 . and E(y/R) = 3π y/(4R). with the further assumption that all elastic energy is in the strain field of the matrix (i. where R is the maximum radius of revolution.119) where again the value of B is for the precipitate material and μ is for the matrix.5 Energies for other shapes of misfitting particles in an infinite matrix Eshelby solved the more difficult problem of a misfitting precipitate of ellipsoidal shape.114. It is understandable that the strain energy is minimized when the particle is a flat plate. In the more typical case where a small. 6. 3B + 4μ v (6. the particle is a sphere. the basic form is that of Eq. For different shapes.) Although the prefactors differ. so thin plates are disfavored if interfacial energy is significant.120 could still be used.118) (Cpt δv)2 + (Cmx δv)2 . both elastic energy and surface energy can be important. The total elastic energy in the misfitting sphere of volume v and its surrounding infinite matrix from Eqs. In reality. 6. This is a classic result.

0] surface is 2 times longer than the [1.13 a [1. which will have a different surface energy than the [1. as illustrated in Fig.1] surface may be favored over√the zig-zagged [1. (b) Construction of a series of “vicinal” surfaces with [1.1 Surface structures The surface energies of crystals are generally not isotropic.1] surface have two bonds to neighboring atoms.0] [1.0] surface is then 2/3 × 1. the relative importances of surface energy and strain energy can change as a particle grows in volume and its surface-tovolume ratio is reduced.1] surface. Along the circled part of the surface.1] surface.0] surface. 6.1] b (a) Approximation of a circle by square elements. The idea is that although atoms in close-packed planes have lost neighbors on one side. 17:29:19 .414 times longer. they are surrounded snugly by other atoms.6. and a discussion of this choice illustrates key issues for the surface energies of crystals. so this simple bond-counting model predicts that a zig-zagged [1. However. for this 45◦ angle a bond-counting model predicts that a [1. 6. The total energy for the [1.414 = 0. Assuming large areas of [1. An alternative at lower right is to have a diagonal [1. giving favorable chemical bonding. the atoms at a [1.0] surface because it has less length – for a 45◦ slope a [1.1].1] surface.13a.6 Surface energy 6. t Fig. a diagonal [1. it is possible to construct a macroscopic surface shape with zigs and zags of atom planes.0] surface would have 2/3 the surface energy.13a for surfaces of [1. 6. A choice is presented at the lower part of Fig. but must be 1.0] surface is favored over straight lengths of [1.0] surface compared with the atom dimensions (so corner atoms can be neglected).0] surface.94 times that for the [1. rather than the three bonds for atoms at a [1. and depend on orientation.0] planes and one-atom ledges of different separations.0] planes of a square lattice. To make matters more complicated. Geometrically. Close-packed crystal planes tend to have lower specific surface energies than higher-index planes.Energy 158 of precipitates. 6.

also minimizes the total surface energy.14d shows the case where the specific surface energy of a [1.14c we assume that the higher index (approximately [3. 17:29:19 .14 In Fig. presented in Fig.6. This “Wulff construction” arranges the planes of low specific surface energy to optimize their coverage of the crystal surface. • The length of the vector should be proportional to the surface energy of its corresponding plane. The lower pictures in Fig. The [1. given the equilibrium shape of a crystal.14 a b c d e (a) Steps of Wulff construction. Figure 6. All planes have the same specific surface energy. 6. with equal specific surface energies for all planes. each one pointing along a direction normal to a crystal plane.6 Surface energy 159 Fig.0] t Fig. (c)–(e) Equilibrium crystal shapes with increasing specific surface energy of [1. 6. • The crystal shape is bounded by the planes closest to the central point.6.14. intersecting their tips.7] σ[1.0] surfaces.14b–e. 6.0] plane.1] surfaces are still assumed to have equal energy. 14 For isotropic surface energy. 6. and need not be considered. It can sometimes be used backwards to find the relative surface energies for different crystallographic planes. that predicts the optimal shape of a crystal when the shape is determined entirely by surface energy.0] surfaces. which minimizes surface-to-volume ratio. Figure 6.2 Wulff construction There is a clever geometrical construction. • Draw planes perpendicular to the arrows.0] and [1. To make a Wulff construction: • Starting at a point. which has the lowest surface area.1] planes. or in this case a circle. This construction naturally gives a sphere. The Wulff construction is presented here without proof. however.13b shows how other angled surfaces can be constructed from sections of [1. 6.1] plane is a bit higher than for a [1. (b) Equilibrium crystal shape for isotropic surface energy. There is less surface covered by σ[3. Complete sets of planes are shown (without the vectors) in Figs.13b are sometimes called “vicinal” [1. 6. The first three steps are illustrated in Fig.14a for two planes. a sphere. 6.14b shows the Wulff construction for a crystal with isotropic surface energy. so all planes are equidistant from the central point. 6.7]) planes have a high surface energy. The close planes are the ones that form the crystal surfaces. draw vectors radially outwards.

with a displacement of the Wulff construction by σGB along the normal of the grain boundary.15a.3 Wulff construction with grain boundaries Figure 6. 4.3.15a would overlap slightly. 4. and the precipitate would be nearly flat and spread widely across the boundary. however. 4. so these {hkl} surfaces are included in the favored shape of the precipitate. 6. If σGB were zero.1] planes. 6. Other structural rearrangements at the grain boundary might be expected for this unusual case. A precipitate shape with more surface area parallel to the grain boundary is more favorable because more grain boundary energy is eliminated. Here there is low surface energy of the {hkl} planes in contact with the matrix phase. The curved surfaces have the same energetics as for the homogeneous nucleation of the precipitate. Figure 6. and the equilibrium shape is altered as shown at the bottom of Fig. so eliminating it effectively reduces the surface energy cost for forming the precipitate at a grain boundary. If σGB > 2 σX1 . 6.Energy 160 [1.5 is a clear example.1] plane is 2 times larger than for [1. σX 1 σX 1 σGB σGB σX 1 t Fig. These crystallographic features on grain boundary allotriomorphs are fairly common – Fig. With images like those of Fig. the spheres in Fig.0] planes. it is possible to assess the relative energies of the different surfaces between the precipitate and matrix.15 a hkl σX 1 σh k l b (a) Displacement of surface energy vectors σX1 by the amount σGB in the direction normal to the grain boundary.15b shows another modification that is expected in general.6. An important issue is that the grain boundary energy σGB is unfavorable. but the width of the “lenticular” precipitate allows more elimination of grain boundary area. The limits are informative.14d. (b) Crystallographic preference of the surface energy for hkl planes. The resulting shape is the lens discussed in Sect. In this case the crystal is bounded entirely by [1. the precipitate would be a sphere as expected for homogeneous nucleation where no grain boundary is present. the surface energy of forming a precipitate would be negative. showing effect on precipitate shape.14e shows the case where the specific surface energy of the [1. 17:29:19 . 6. If σGB were slightly less than 2 σX1 . The Wulff construction for isotropic surface energy is modified as shown in Fig.15 shows an interesting modification of the Wulff construction that is useful for understanding the shapes of precipitates that nucleate at grain boundaries. and nucleation would occur slightly above the critical temperature for the bulk free energy.0] planes. 6. √ Figure 6.5. and the resulting lenticular precipitate shape. and the grain boundary [51].

(a) Assuming the two ions to be point charges of +e and −e.3 6. 12 r r (6. neglecting the energy of repulsion. Their internuclear separation is 0.Problems 161 Problems 6. but then why is it no better than a rule of thumb? The equilibrium separation of the K+ and Cl− ions in KCl is 2.1 Consider the stability of a molecule of three identical atoms.8 eV.2 6. (b) The ionization potential of potassium is 4. including overlap effects. will alter the result of part b. A “rule of thumb” is that the stability of a crystal structure goes as the inverse of the density of states at the Fermi surface.032×1012 dyne/cm2 .34 eV.122) a0 π a0 (a) What is the electron density (electrons per Å3 ) at a point 0. Find the dissociation energy.6 A A − 6 .79 Å. Why might this be true.4 6. Using the density for potassium (0. Calculate the bulk modulus of potassium and compare to its actual value.82 g/cm3 ) and a molecular weight of 39. or a linear molecule? (b) By what percentage is the energy lower for the preferred structure? The actual bulk modulus of potassium metal at room temperature is 0. calculate the potential energy of attraction of the two ions. 6.121) (a) Which structure of three atoms has the lower cohesive energy: an equilateral triangle. The 4f electrons of the rare earth elements tend to reside close to the nucleus. calculate its electron density. (d) Make a semiquantitative estimate of how a proper normalization. where the atoms interact by this Lennard–Jones central-force potential φ(r) = 6.37 Å from the nucleus of an isolated hydrogen atom? Now suppose that two hydrogen atoms are brought together to form an attractive bond with 1s wavefunctions. and the electron affinity of Cl is 3.74 Å. (b) What is the electron density midway between the two nuclei? Do not go to the trouble to renormalize the molecular wavefunction.7 Arguing with a Wulff construction. N/V. assuming each potassium atom contributes one electron to the electron gas.42 eV. With this information only.5 6. and are screened by the outer 6s (and 5d) electrons. what will be the equilibrium shape of these crystals? 17:29:19 . (c) Explain in words why there is a difference between the results of parts a and b. What is the energy due to repulsion of the ions? Use the following 1s electron wavefunction of a hydrogen atom for this problem    3/2 1 r 1 ψ(r) = √ exp − (6. draw an approximate electron density of states for the elements from Ce to Lu. (c) The measured dissociation energy is 4.

123 and 6. but a vicinal surface as shown in Fig. and the density of steps on it is tan θ/a. k.16 Figure for Problem 6. This problem calculates the surface energy for surfaces tilted by the angle θ from the flat surface.. Show that it is a graph of γ = 2 cos(θ − φ) . (b) The specific surface energies of (1.13b has an energy that is proportional to the number of steps per unit area.0.125) (d) Identify 2 in Eq.) (b) Use the trigonometric identity cos(θ − φ) = cos θ cos φ + sin θ sin φ .0) planes are equal.1) planes and (1.124. and all other specific surface energies are at least twice this value.123) γ (θ ) = γ0 cos θ + a where a is the lattice parameter of the simple cubic crystal.16 (i.124) to find a relation between γ0 and γ1 /a in terms of tan φ.16 shows a generalized Wulff plot for the surface energy of parts a and b. 6. 6.8. (6.162 Energy 2G θ φ t Fig. (6. this ratio of γ0 to γ1 /a). 6.e. (6. (a) Show that the specific surface energy γ can be written γ1 sin θ . l) crystal plane is proportional to h2 + k2 + l2 .1. 6. (e) For the case shown in Fig. showing two rotated circles that intersect the origin.125 from the terms in Eqs. will the equilibrium shape of a crystal have exposed surface with surfaces other than the {100} family? 17:29:19 . 6.8 Wulff construction for a vicinal surface Suppose that a flat surface has a surface energy of γ0 . 6. (a) √ The specific surface energy of an (h. (c) Figure 6. (Hint: The fraction of flat surface can be written as cos θ.

2. 7. as does the vibrational entropy. this  increases with temperature.1) is an equation for entropy of breathtaking generality.4 this is often a small effect because not many electrons are available for these excitations. with kB as the Boltzmann constant. Configurational entropy in the point approximation was used throughout Chapter 2. are then adapted to include effects of vibrational entropy. 17.7 Entropy Without entropy to complement energy. but at high temperatures the electronic excitations may interact with phonons. Chapter 7. which counts the ways of finding the internal coordinates of a system for thermodynamically equivalent macroscopic states. thermodynamics would have the impact of one hand clapping.5. “What do we count for . calculated previously with configurational entropy alone. The epitaph on Boltzmann’s monument shown in Fig. The electronic entropy of a metal increases with temperature. and Sect. 6.1 S = kB ln  . (7. We used this idea extensively in Chapter 2. when the number  counts the intervals of volume explored as atoms vibrate. The other major source of entropy is vibrational entropy. For metals there is a heat capacity and entropy from thermal excitations of electrons near the Fermi surface. Critical temperatures of ordering and unmixing. The nub of the problem is the number . when the number  enumerates the ways to arrange atoms on the sites of a crystal.1. 17:28:36 . Here it is modernized slightly. On the other hand. This chapter. shows how the configurational entropy of chemical disorder or magnetic disorder can be calculated more accurately with cluster expansion methods. and how do we count them?” Sources of entropy are listed in Table 7. the method to calculate  does not depend on temperature. 7. Physical questions are. but as discussed in Sect. and some were discussed in Chapter 1. and its origin is explained in Sect.1 Static and dynamic sources of entropy It is often reasonable to separate the internal coordinates of a material into configurational ones and dynamical ones. As an example. and phonons interact with other phonons as discussed in Chapter 24. 7.4 accounts for magnetic entropy in essentially the same way.

and when many states are available the electronic entropy is large.1 The Boltzmann monument in Vienna. 7. although the configurational variables have different equilibrium values at different temperatures.1 Sources of entropy in materials Matter Nuclei Electrons Electron spins Energy Nuclei Electrons Electron spins Structural configurations Sites for the nuclei (the electrons will adapt) Sites for electrons in mixed-valent compounds Orientational disorder (magnetic disorder) Dynamics Vibrations (phonons) Excitations across Fermi level (electronic heat capacity) Spin waves (magnons) Configurational entropies of atoms or spins undergo changes during chemical ordering or magnetic phase transitions. although at most temperatures of interest in materials physics the nuclear spins are fully disordered and their entropy does not change with temperature. respectively. and the vibrational excursions of atomic nuclei are larger. The constant k is related to kB by the factor 2. and are presented in Sect. such as normal modes of vibration. Again. 2 Temperature also drives electronic excitations to unoccupied states.1 Configurational entropy was largely understood by Gibbs. Nuclear spins undergo ordering transitions at low temperatures. but care must be taken when counting them if the configurations of spin disorder are already counted.Entropy 164 S = k. are excited more strongly by thermal energy. In essence. who presented some of the combinatoric calculations of entropy that are used today [22]. new local variables are added to the list of composition and long-range-order parameters to describe more precisely the atom configurations on lattice sites.3026 if log denotes log10 . Photograph courtesy of Clemens Pfeiffer cc-by 3. Dynamical entropy grows with temperature as dynamical degrees of freedom of a solid. The calculation of  is more difficult when there are partial correlations over short distances. the entropy from dynamical sources increases with temperature because with stronger excitations of dynamical degrees of freedom. Our notation uses  instead of W. 7. log W t Fig. but cluster approximation methods have proved powerful and accurate [52]–[54].2. 17:28:36 . temperature does not alter the combinatorial method of calculating  for the configurational variables. Spin excitations are another source of entropy. the 1 Electronic entropy can also have a configurational component in mixed-valent systems.2 With increasing temperature more phonons are created. Fundamentally. too.0. Table 7.

10 presented a thermodynamic analysis of the order–disorder transition in the point approximation. For example. This rule is not helpful for understanding the thermodynamics of phase transitions.2. 7. 7. For a phase transition. Furthermore. 7. and only a few antisite atoms are present. and not on its structure. 3 This is frequently called a “phase space. or (3) a hypothetical case when the coordination number of the lattice goes to infinity. The domains have opposite sublattice occupancies. We see that the numbers of A-atoms and B-atoms on each sublattice are equal. is the  for Eq. This assumption is best in situations when (1) the temperature is very high. which assumed that all atoms on a sublattice were distributed randomly. it is possible to improve on this assumption of sublattice randomness by systematically allowing for short-range correlations between the positions of atoms. The domain boundary is marked by the curved line. (2) the temperature is very low. Equal numbers of A. 17:28:36 .4 7. This volume.2 Short-range order and the pair approximation 165 system explores a larger volume in the hyperspace3 spanned by position and momentum coordinates.1. It might be tempting to re-define the sublattices within each domain. normalized by a quantum volume if necessary.7. 7. 4 The “Kopp–Neumann rule” from the nineteenth century states that the heat capacity of a compound is the sum of atomic contributions from its elements. Incidentally. For more interesting temperatures around the critical temperature. what is important is not the total vibrational entropy so much as the difference in vibrational entropy between the two phases. By this rule. so the atoms are indeed randomly distributed on the sublattice. but this gets messy when the domains are small. allowing for a large value of L. the curved line is an “antiphase domain boundary” (APDB).2 Short-range order and the pair approximation Section 2.5.2 Alloy with two domains of chessboard order. so the LRO parameter L = (R − W)/(N/2) = 0. the Kopp–Neumann rule has inconsistencies when picking an atomic heat capacity for carbon.and B-atoms are in each domain. the vibrational entropy of a solid phase depends only on its chemical composition. for example. for example. there is obviously a high degree of order within each of the two domains. as discussed in Sect. a deficiency of the point approximation for ordering is illustrated with Fig. Nevertheless. The standard approach to this problem is to use short-range order A on α B on β A on β B on α t Fig.” not to be confused with geometric properties of crystallographic phases.

In the pair approximation. Figure 7. start with the assumption that the pairs really are random.3 B 4 A b a (a) Four numbered bonds to a central atom. we start with an arbitrary first-nearest-neighbor pair in the lattice (as opposed to starting with an arbitrary lattice site in the point approximation). This gives =K αβ αβ zN 2 ! αβ αβ .3) The correction factor. where an atom in the subscript is on the sublattice in its β β superscript. 2.3. Unfortunately. however.2 has a high degree of “short-range order.4) K≡ N N 2! 2! 17:28:36 . NBα . A–B. NAB . When calculating a configurational entropy of pairs. 7. turns out to be z−1  β β NAα ! NBα ! NA ! NB ! . the structure of Fig.2) NAA ! NAB ! NBA ! NBB ! A correction factor K was added to give the correct configurational entropy in the expression pair Sconfig = kB ln  . The entropy would be obtained from the combinatorics of placing one of four pairs on the locations of all the bonds in the crystal (a total of zN/2 bonds as shown in Fig. NBB }. NBA . K.7b). The problem is that the first two bonds put an A-atom at the center. 7. We need to ensure that the point variables (such as the central atom in Fig. (7. an inconsistency arises as illustrated with the help of Fig. B–A or B–B pair. so there are four equal possibilities for each bond.3a) are accounted for consistently.2 are of the A–B or B–A type. the SRO can be parameterized with the pair αβ αβ αβ αβ variables {NAA . it is not appropriate to treat the bonds as randomly selected. but for the pair approximation (and for all higher-order clusters). Figure 7. NB } are still needed. 7. (SRO) parameters. For a disordered solid solution of equiatomic composition. It is a problem that did not arise in the point approximation. For reference. labeled with a number. whereas the second two put a Batom at the center.3b shows a typical random selection of atom pairs for the four bonds. and consider its probability of being an A–A. The point variables {NAα . (b) Random selection of the types of atom pairs across each bond. we need a correction factor that gives the correct configurational entropy for a fully disordered alloy [55]. This is obviously inconsistent. (7. NA .” For an alloy that develops two sublattices.Entropy 166 A 1 B A 2 A 1 4 B 3 2 B 3 t Fig. 7. Since most pairs in Fig.3a shows four bonds. such as the total number of A–B pairs. 7. which prevents inconsistency when evaluating the point variables. (7. these four pairs are expected to occur at random.

The combinatorial factor of Eq.9 in Eq. L = 0.13) and with a similar expression for the “wrong” B-atoms on the α-sublattice. A better description of the order in this alloy is obtained with a short-range order (SRO) parameter. Using the pair variables. (7. or (more properly) s = 0.14) which counts all A–B pairs. 7.6) 2 2 2 The correction factor ln K is easy to calculate with Eq. 7. This would still yield L = 0 for the alloy depicted in Fig.75 using Eq.3 can now be obtained by using Eq.2 Short-range order and the pair approximation 167 It is not too difficult to derive K. since the process of eliminating the inconsistency discussed with reference to Fig. the number of “right” A-atoms on the α-sublattice. (7. (7. 17:28:36 .5) ln  = 2 2 2 8 8 8 zN 8 zN ln  = ln + ln K = ln 4 + ln K .6 from the pair variables alone.6. however. s αβ s≡ αβ NAB + NBA . 7.3 forces some correlations between the pair variables. (7. reducing the apparent randomness of the alloy.12) which is the correct configurational entropy for an equiatomic random solid solution.7) N ! 2      N N N N N N ln K = (z − 1) 4 ln − −2 ln − . s = 0. Nevertheless. zN/2 (7. In the pair approximation we can still define a LRO parameter L as before.   √ pair Sconfig = kB zN ln 4 − (z − 1)N ln 2 . 7. A–B.7. 7. 5 It is the entropy of the point approximation for a fully disordered solid solution. for which the site occupancies are. (7. is αβ R= αβ NAA + NAB z (7.9) The ln  of Eq.  4 z−1 N ! ln K = ln 4 2 .2.87 if correction is made for the lost bonds at the surface. 7. (7. pair Sconfig = kB N ln 2 . equiatomic alloy.2 is evaluated with the Stirling approximation   zN zN zN zN zN zN ln − −4 ln − + ln K . the LRO parameter is L = 2(R − W)/N. giving pair Sconfig = kB ln  . For this alloy.2. independent. in fact. 7. for example.10) (7. without regard to sublattice. but it should suffice to show how it works for a random.4 for a disordered equiatomic alloy.5 Notice that the correction factor K is negative.8) 4 4 4 2 2 2 ln K = −(z − 1)N ln 2 .14. For the alloy of Fig. B–A. reducing the combinatorial result of Eq. and B–B.11) (7. there are zN/8 pairs of each type A–A. 7. 7. where all four point variables equal N/4. 7. This is expected.

Instead of enumerating atom positions on sites.8 L 0.18.15) αβ αβ kB T NAA NBB This result is valid at temperatures above Tc .0 t Fig.4 0.0 SRO.5 2. This dashed curve corresponds to a situation when sublattice formation is suppressed. LRO 0.74. it uses the concept of 17:28:36 . however.4 shows results from the pair approximation for the SRO and LRO versus temperature for B2 ordering on a bcc lattice (see Fig. the SRO persists. Figure 7. (7. Even though LRO is lost above the critical temperature.10. the SRO parameter follows the LRO parameter. generally over short distances.0 2. An important feature of SRO is that it does not vanish at high temperatures. By the method of Sect.4 0. Also shown is the SRO below the critical temperature from Eq. showing that the SRO decreases monotonically with temperature.2 S 0. Below the critical temperature. 7. the SRO tracks the change of the LRO parameter with temperature. 7. The SRO is nonzero above the critical temperature of approximately 1. the equilibrium ratio of the numbers of pairs is   αβ αβ NAB NBA 4V = exp + . perhaps owing to kinetics as discussed in Chapter 22. 14.1.15.0 0.3 Local atomic structures described by clusters A “cluster expansion” is a method to account for atom arrangements on a lattice.Entropy 168 1.0 Temperature (4V/kB) Long-range order parameter L.3). At temperatures below Tc . 2. 2.5 1.6 0. which rises a bit more steeply than for the point approximation in Fig. 7. and short-range order parameter s versus temperature for B2 ordering on an equiatomic bcc alloy in the pair approximation.0 1. Tc .5 3. The equilibrium SRO at high T can be calculated by considering the interchange of an A–B pair and a B–A pair into an A–A and B–B pair.

etc. The first step for a binary alloy is to assign the “spin number”.17) The pair correlation function is obtained similarly. well past the fifth neighbor.n = (σ0 )(σn ) = (+1)(σn ) ξ2.7. consider pair variables with the help of the ordered structure in Fig.5 4 3 4 5 2 1 2 4 1 0 1 3 2 1 2 4 Chessboard ordering with atoms on numbered nearest-neighbor sites. correlation functions. At the other extreme.n } do a good job of describing the atoms at all sites of the structure.0 = (+1)(+1) = +1 ξ2. B–B.4 = (+1)(−1) = −1 ξ2. For example. no matter their separation from the atom at the origin. For example.3 = (+1)(+1) = +1 ξ2. (7. to each atom.1 = (+1)(−1) = −1 ξ2. where the numbers correspond to nearest-neighbor positions with respect to the site labeled “0”.5 = (+1)(+1) = +1 All correlation functions have the maximum value of ±1 for this structure with perfect order. with σ = +1 for an A-atom and σ = −1 for a B-atom. (7. and their values depend on the atomic structure.7 Adding the fractions of A–A. More detail is provided by pairs. 17:28:36 . the simplest “point correlation function” is a type of average of the atom concentrations on one site. A–B. If the A-atoms are white. Correlation functions are constructed as products of these spins.18) 6 The same result is obtained if the B-atoms are white because both signs are switched in each product. triangles. squares.16) ξ1 = (+1)(1 − c) + (−1)(c) = 1 − 2c . and the {ξ2. 7.2 = (+1)(+1) = +1 ξ2. 7. σ . 7 In a fully random alloy. The range and complexity of the correlation functions increase with the level of the approximation.3 Local atomic structures described by clusters 169 t Fig. the same analysis is valid for all pairs. the pair correlation functions are6 ξ2.5. and B–A pairs ξ2 = σA cA σA cA + σB cB σB cB + σA cA σB cB + σB cB σA cA ξ2 = (+1)(1 − c)(+1)(1 − c) + (−1)(c)(−1)(c) + 2(+1)(1 − c)(−1)(c) ξ2 = (1 − 2c)2 . The hierarchy of these clusters depends on the lattice type. The point correlation function is constructed as an average of the +1 for all A-atoms and –1 for all B-atoms ξ1 = σA cA + σB cB . consider a random solid solution. (7.

7. so in general ξn = (1 − 2c)n . In fact. The cluster expansion method is most useful when a short-range description of the atomic structure accounts for the property of interest. An important concept is that any configuration of atoms on a crystal lattice can be described by the {ξj }.20 may be truncated after a few terms. Alloys with similar correlation functions {ξj } are expected to have similar properties.17 and 7. 7. A rigorous way to select the best clusters has proved elusive. F. we can obtain the interaction free energies. at least those at low or moderate temperatures.4 Thermodynamics with cluster approximations For the same V. by “calibrating” Eq. but is beyond the present scope [57]. in the limit as z → ∞. so we might expect to evaluate a property with contributions from the different correlation functions.” {fj }.8 Since all properties of an alloy originate with its atom configurations. 7.18. fj . where all {ξj } are ±1. using known values of F from alloys with known correlation functions {ξj }. Notice that the Tc is less sensitive to the order of the approximation as the coordination number. (7. increasing the order of the approximation causes Tc to decrease. or any ground state property of the alloy F= ∞  fj ξ j . since both the energy and the entropy are different in the different approximations. Table 7.2 presents some results and shows some trends.5. 8 The proof is challenging [56]. z. is larger. The explanation is not obvious. In essence. so we accept it.20) j assuming all “interaction free energies. a set of n interaction free energies can be obtained from a set of n different alloys with known F. For the free energy.20. 7. 7. however. are known. 17:28:36 . the point approximation predicts the correct Tc . (7. For a random alloy. and the {ξj } can describe all atom positions in the alloy. so it is hoped that correlation functions involving short distances (small j or small clusters) are sufficient. It is often expected that longer-range correlations in atom positions are less important for alloy thermodynamics.) The inversion method is not too difficult. the {ξj } can account for alloy properties. (It is generally straightforward to calculate the correlation functions for an alloy structure. and the calculated F may be adequate.Entropy 170 Compare Eqs. If the cluster expansion is truncated after a few terms. In this case the series in Eq.19) These {ξj } are obviously quite different from the previous case of the chessboard ordered structure in Fig. for each additional site added to the cluster comes an additional factor of ξ1 = 1 − 2c.

In the limit of very large clusters.11. and has been in widespread use since the late 1970s [54. Eq.4 Thermodynamics with cluster approximations 171 Table 7. 7.g. 12. the hump in the free energy versus composition curve vanishes. What happens is that the populations of higher-order clusters are themselves biased towards unmixed structures.11 (or 2. for example. 7. there is less effect of composition variations on the free energy. 60]. 7. and a straight line connects two solid solutions of different compositions. and the handling of this problem is the key to the method. using weights that depend on how many subclusters are present in the large. and it continues to decrease as the cluster approximation extends to higher order [61]. is based on free energy versus composition curves of Fig. A systematic “cluster variation method” to do this was developed by Kikuchi [58]. independent cluster (e. One measure of the success of these higher-order approximations is their convergence on the actual critical temperature for ordering. Composition alone is therefore insufficient for characterizing the state of the alloy. 17:28:36 .7212 0. 59]. for example. by Monte Carlo methods. it is natural to extend the pair approximation in a hierarchy of approximations that goes from point to pair to larger clusters such as squares on the square lattice. for example.8222 0. but the analysis of accuracy is itself a specialty. “How reliable is an unstable free energy function for calculating equilibrium phase diagrams?” This question is avoided in part by treating spinodal decomposition as a kinetic process with a modified diffusion equation. Variables for the populations of larger clusters. for example. 2. This viewpoint helps to overcome a conceptual problem with the F(c) curve of Fig.6). In this case the pair variables are overcounted. In essence. these curves from the point approximation show a characteristic “hump” at temperatures below the critical temperature for unmixing. such as first-nearest-neighbor square variables. The systematics to do this are elegant and straightforward [58.2 Critical temperatures [zV/kB ] Lattice Hexagonal Square Triangle Simple cubic bcc fcc z Exact Pair Point 3 4 6 6 8 12 0.8163 0.4.6062 0.10.4). At intermediate compositions.7522 0. it is necessary to correct for the overcounting of all subclusters within the largest cluster. Discrepancies of 1% or so are possible with modest cluster sizes. and the F(c) curve in the point approximation is naïve.5673 0.9142 1 1 1 1 1 1 For further accuracy.8690 0. consistent with their concentrations. as in Sect.7944 0. In the pair approximation. as determined.7. The phase diagram for unmixing. the magnitude of this hump is decreased considerably.8222 0.5062 0. are placed randomly on the square lattice. With these internal degrees of freedom accounting for much of the unmixing enthalpy. Fig. An important conceptual point is addressed by cluster approximations for the problem of chemical unmixing by spinodal decomposition.. where all A-atoms are on one side of the cluster and B-atoms on the other side.

7. For temperatures very close to Tc  1 Tc +δT SP (Tc + δT) − SP (Tc ) = CP (T) dT .22) which emphasizes how the heat capacity is a derivative quantity of the entropy. S = kB ln . the singularity is milder. The heat capacity is therefore highly sensitive to changes in the entropy near Tc . It can have infinite height. Tc Tc (7. the heat capacity has a mathematical singularity at Tc . and the larger is S = kB ln . Tc These δ-function and logarithmic singularities can be broadened in temperature by kinetic effects that become severe near Tc . • For a first-order phase transition with a latent heat (and a jump in entropy of Eq. In fact.e. Second-order phase transitions have special challenges.5 Concept of vibrational entropy We can understand the essence of vibrational entropy for phase transitions with Eq.22 gives the latent heat at Tc . the material is at a point in this space that gives all positions {qi } and momenta {pi } of its atoms.. It is proportional to a Dirac δ-function. A function like − ln |1 − T/Tc | has the appropriate properties. At any instant in time. The point moves around as the atoms vibrate. (7. and integrating it in Eq. 7. 1.172 Entropy Box 7. 7.1. 17:28:36 .21) SP (T2 ) − SP (T1 ) = T T1 or the equivalent relationship between SV and CV . but must be rather narrow in temperature so it does not integrate to a finite value over an infinitesimal range of temperature. The larger the volume explored during vibration. the greater the number of ways  of finding the system.23) CV (T) = −NkB 0. since it requires an accurate difference in entropy of two phases near Tc . Figure 7. • For a second-order phase transition without a latent heat.5). In fact. the heat capacity has a singularity that exists over an infinitesimal range of temperature. but has an infinite height and finite area.6.6 depicts a multi-dimensional “hyperspace” spanned by the momentum and position coordinates of a material with 3N atoms. Predicting the magnitude of a latent heat is challenging. i. while making reference to Fig.4945 ln 1 − + constant .1 Heat capacity Heat capacity has been an important source of information about the entropy of solids through the relationship  T2 CP (T) dT . 7. as discussed further in Chapter 20. None of the cluster approximation methods discussed so far is capable of giving the logarithmic form for the heat capacity near Tc . Onsager’s exact solution for ordering on the 2D Ising lattice with 1nn interactions [62] gives the heat capacity near Tc T (7.

Assume each vibrational mode is a normal mode that can be treated independently of all the other normal modes of the material. q3N–3 ~1/ωjα ~1/√kα t Fig.5... however. 7.. 17:28:36 .. qβ max ks α ωα (7.. but different atom masses. and the same potential energy at the turning points of its vibrations where it is instantaneously stopped. (That is.7. At any instant in time. (a) Phases α and β have the same atom masses.. = 1 1 ks α q2α max = ks β q2β max . p3N–3 p1. a mode j with spatial coordinate qj explores a range of qj that is proportional to the inverse of the frequency ωj . In this case. the normal modes persist forever. The changes in chemical bonds alter the forces between atoms.6. The ranges of travel in position coordinates and in momentum coordinates depend on the strengths of the interatomic springs and on the atomic masses.E. p1. each atom has the same kinetic energy as it passes through the equilibrium point of its vibrational cycle. so changes in springs and masses cause changes in vibrational entropy. 7.6 a ~1/ωjβ ~1/√kβ b Multidimensional space of all momentum and coordinate variables for all atoms of a material. the material is at a point in this space..25) √ where the relationship ω = k/m for harmonic oscillators was used. p3N–3 α β α ~1/ωjβ ~1/ωjα β q1. 2 %2 ωβ ks β qα max = = .1 Changes in springs Suppose two phases α and β have the same atoms. 7. (b) Phases α and β have the same interatomic force constants. and do not transfer energy among themselves.24) (7. and the masses are equal if the same atoms are in both phases. but these phases have different structure and bonding. as shown from the equality of potential energy P.. q3N–3 ~ √mβ ~ √mα q1. Assume the two phases are at the same temperature. but different interatomic force constants. {ωi }.) For a fixed amount of thermal energy.5 Concept of vibrational entropy 173 We obtain differences in vibrational entropy by finding the change in the volume explored in the hyperspace of Fig. so they vibrate with different frequencies.

2 Changes in masses Now consider the alternative case where the masses of the atoms are altered. The volume in hyperspace explored by thermal vibrations is the product of all the ranges of coordinates explored ⎞ ⎛ 3N ω ⎜ j jα ⎟ ⎟ ⎜ (7. Nevertheless.28 is that the vibrational entropy is larger if the springs between atoms are weaker. pβ max mβ ωα K. If the A–B pairs have a deeper interatomic potential well (e. for the same temperature.29) (7.6a and Eq. 7. but the interatomic springs are kept the same. The potential energy depends only on the compressions of the springs. and the vibrational entropy decreases.26) ωjβ ωjα   ωjα β . and the vibrational frequencies are smaller. 6.30) 9 For differences in entropy. the excursions along the momentum coordinate. From this one mode. (7.6b shows how. so A–B bonds are energetically favorable. = (7.g. the difference in vibrational entropy is9     ω0 ω0 β Sj − Sjα = kB ln − kB ln . which are unchanged.27 gives the contribution to entropy from one normal mode. depend on mass.8). Changes in the force constants between atoms are expected if the atoms are rearranged to change their chemical bonding. so the interatomic springs are stiffer. In other words. we expect the vibrational entropy of chemical ordering to be negative. 7. 7.27) Sj − Sjα = kB ln ωjβ Equation 7. The maximum excursions of the vibrating atoms are unchanged because the energy at the turning points of a vibration is all potential energy.. 7.6a. at high temperatures it is k T/. Equality of the kinetic energies in the phases α and β gives p2β max p2α max = 2mα 2mβ $ ωβ pα max mα = = .E. pmax .Entropy 174 This is illustrated in Fig.5. Figure 7. For example. the curvatures at the bottoms of these wells are expected to be larger. compared with A–A and B–B. In a threedimensional solid with N atoms there are 3N − 3 3N normal modes. suppose 0 < 4V = eAA + eBB − 2eAB . and a larger spatial excursion in the α-phase. The atom vibrational frequencies are higher. Fig. the velocities of the atoms through their midpoint position are altered by the change in mass.28) Sβ − Sα = kB ln ⎜ ⎟ . B 0 17:28:36 . 3N ⎠ ⎝ωjβ j What we learn from Fig. the value of ω can be neglected. (7. Instead. which shows a lower frequency for the mode ωjα in the α-phase than ωjβ in the β-phase.

labeling the first. second. The number  becomes the number of ways to distribute m indistinguishable phonons over N + m possible sites (equal to the probability of distributing N indistinguishable bars over N + m possible sites). The product of these independent probabilities is (N + m)! / N!. The trick is to consider the three vertical bars and the four circles as distributed at random over the seven positions in each row. etc.7a illustrates five possibilities for distributing four phonons among four oscillators. Figure 7. too. or oscillator.. the interatomic forces are usually changed.6 Phonon statistics The independence of phonons is an important feature of the “harmonic model. i. Figure 7.7b.27   ωα Sβ − Sα = 3kB ln . and the mth phonon can be placed in N + 1 ways. Equation 7. so Eq.” for which there are many useful and powerful mathematical results. illustrated with Fig. Obtaining the entropy from the volume explored in phase space gives the same result as Eq. where there are a number of ways .32 is sometimes useful for intuition. 7.33 needs a factor of m! in its denominator. (7. 7. such as three phonons in the first oscillator. Consider a set of oscillators N. and the interatomic spring constant ks was assumed the same for both phases. The overcounting is the number of ways of distributing m phonons over m sites. 7. 7. 2 mα (7.7. Suppose there are a number m of phonons that are free to distribute between these different oscillators. all of the same energy ε. There is. The top configuration in Fig. N! m! (7.6 Phonon statistics 175 √ where again the relationship ω = ks /m was used. the second has one less. The bottom configuration places one phonon in each oscillator.30 Sβ − Sα =   mβ 3 kB ln . which is m!. We calculate the phonon entropy from Eq. to arrange the m phonons in the N oscillators. but phonons are indistinguishable. so there is no limit to how many of them can be present in each mode.e. an overcounting because we put the phonons down individually. however. one in the second. and that is the case of isotopic substitution.31) ωβ but also. The result is the binomial coefficient10 = (N + m)! . from Eq.32) for the case of unchanged spring constants. and none in the other two. and assume that these oscillators can exchange energy.7a inspires a trick for calculating  combinatorially. but if an atom is substituted for another chemical species of different mass. There is a case where Eq.7a has all four phonons in the left oscillator. 7.1. 7. 7.32 is quantitative. 17:28:36 .33) 10 The first phonon can be placed in N +m possible sites. The middle three are other possibilities. N +m−1 possibilities. This is often important for the isotopes of hydrogen. Phonons are bosons. 7.

7. (b) Abstraction of part a. circles denoting phonons.7 a b (a) Five configurations of placing 4 (m) phonons (circles) into 4 (N) oscillators (containers). 7.Entropy 176 t Fig. with vertical lines demarking oscillators. Substituting Eq.33 into 7.1. and using the Stirling approximation of ln(x! ) x ln x − x .

(7.34) Now consider the entropy per oscillator. Sosc /N. n ≡ m/N . and the occupancy of each oscillator. Sosc = kB (N + m) ln(N + m) − N ln N − m ln m .

35) N . (7. Sosc = kB (1 + n) ln(N(1 + n)) − ln N − n ln(nN) .

Sosc = kB (1 + n) ln(1 + n) + (1 + n) ln N − ln N − n ln n − n ln N . (7.36) N .

(7.37) N . Sosc = kB (1 + n) ln(1 + n) + (1 + n) ln N − (1 + n) ln N − n ln n .

38 with the analogous equation for fermions. the combinatorics give  = N! / [(N − m)! m! ]. Sosc = +kB (1 + n) ln(1 + n) − n ln n .38) N It is interesting to compare Eq. is the same as the configurational entropy of mixing of atoms on a crystal lattice (Eq. (7. For fermions. 2. Fermion entropy. 7. where m electrons are distributed over N states of energy ε.23). incidentally. where each lattice site can hold only one atom . for example.

and periodic boundary conditions. described in more detail in Sect. The Born–von Kármán model assumes that nuclear masses are connected by massless electron springs. Solving its equations of motions gives the normal modes of vibration of 17:28:36 .38 and 7. which are typically assumed to respond instantly to nuclear displacements (Born–Oppenheimer approximation).39) N Equations 7.1. and n can exceed 1.7 Lattice dynamics and vibrational entropy The mass of an atom resides in its point-like nucleus. assumes a crystal with translational periodicity.39 have two differences – the signs are reversed. and the forces between nuclei are transmitted through lightweight electrons. 7. giving a set of ordinary second-order differential equations of motion. The model. (7. Smix = −kB (1 − c) ln(1 − c) + c ln c . whereas 0 ≤ c ≤ 1. 24.

) Normal coordinates offer a sweeping simplification of the vibrational dynamics. exist in each normal mode. as is appropriate for harmonic phonons. Zi = 3N . with dispersion relations between frequency and wavevector.45) (7. 1. 1 − e−βεi (7. T) = +βε e i −1 −βε − ln e i − 1 = ln(1 + n) . n(εi . 7. This remains true in quantum mechanics. 11 The Planck distribution is the same as the Bose–Einstein distribution when the chemical potential is zero. each normal mode does not transfer energy to the other modes.46) (7.41) 1 − e−βεi where Eq. 7.43) and the phonon entropy (vibrational entropy) is found by differentiating with respect to T  3N   βεi −βεi Svib = kB + βε . For harmonic solids. n(εi .   1+n . Sosc . (7. e−βεi /2 ZN = . βεi = ln n (7. 7.44) − ln 1 − e e i −1 i Using the Planck distribution11 for phonon occupancy versus temperature. so we start with the thermodynamics of the quantum harmonic oscillator.19 F = −kB T ln Z.42) i from which the phonon free energy is calculated by the prescription of Eq. T). although integral numbers of excitations. phonons.47) it is straightforward to show that Eq. and is unchanged as more phonons are added to the normal mode. The partition function for a harmonic solid with N atoms and 3N independent oscillators is the product of these individual oscillator partition functions. Fvib =  1 εi + kB T ln 1 − e−βεi . The partition function for a single harmonic oscillator of frequency ωi = εi / is Zi = ∞  e−β(n+1/2)εi . In a harmonic solid.40) n e−βεi /2 . the energy cost for each additional phonon is εi = ωi (ωi is the angular frequency of the oscillator). These solutions are wavelike. Phonon modes are considered as independent harmonic oscillators.7 Lattice dynamics and vibrational entropy 177 the crystal. a normal mode persists forever. (7. Without damping. (Classic and useful texts on the Born–von Kármán model include [63]–[65]. 17:28:36 .40 as a geometric series times the constant “zero-point” factor exp(−βεi /2). (7.7. 2 3N 3N i i (7. 1 . 7.44 reduces to Eq. and β ≡ (kB T)−1 .41 was obtained by identifying Eq.38 for the vibrational entropy per oscillator.

using Eqs. 7. 7. For a three-dimensional solid. 3Ng(ε)dε phonon modes are in an energy interval dε. Eq.48) which becomes at high temperatures ZN =  m  . 7. kB T 3Ng(εj )ε εj j=1 . 1 − e−βεj j=1 (7.46. 7.49) Again.45. 7. a useful expression for the phonon entropy of a harmonic material at any temperature can be obtained from Eq. (7.38) ∞ Svib (T) = 3kB g(ε) .47. g(ε).48 (cf. e . For a DOS acquired as digital data in m intervals of width ε (so εj = jε).Entropy 178 It is useful to work with a phonon density of states (DOS). the partition function can be computed numerically ZN = m  −βεj /2 3Ng(εj )ε .

43 and 7.5 1. series expansions of Eqs. n(ε) + 1 ln n(ε) + 1 − n(ε) ln n(ε) dε .0 t Fig.5 0. Fig.50) 0 where g(ε) is normalized to 1 and n(ε) is the Planck distribution (Eq.0 − ln(E ) + 1 0. 7.38 to approximations for up to four terms of a standard high-temperature expansion (terms are presented in the figure). the following approximation works well for the entropy of one oscillator mode with energy ε = ω 3. can be useful. 7. 7. with E = ω(kB T)−1 .5 − ln(E ) + 1 + E 2/ 24 2. 17:28:36 .0 − ln(E ) 1. At high temperatures. For thermodynamic calculations. For Sph (T).8 − ln(E ) + 1 + E 2/24 − E 4/ 960 0 2 4 6 Temperature (hω/kB) 8 Temperature dependence of entropy calculated for one mode of energy ω.8 compares the exact expression of Eq. (7. 7. valid at high temperatures.0 Exact Phonon entropy [kB/mode] 2.45) for the temperature of interest.44. 7.

7.8 Bond proportion model 179 .

.ω (T) = kB n(ε) + 1 ln n(ε) + 1 − n(ε) ln n(ε) . Svib.

α and β.12 When comparing the vibrational entropies of different alloy phases. 7. 7..53) 0 The important point about Eqs. A challenge is to determine how Svib depends on the configurations of atoms. 7. and the experimental results were ambiguous. This is challenging.48 and 7. Svib. can be obtained readily from Eq. these two phases have the same free energy (i. It proves necessary to have rather accurate information on phonon frequencies or heat capacities to assess the importance of vibrational entropy. 17:28:36 .ω (T) kB ln ω (7.50 is that the only material parameter relevant to the thermodynamic partition function in the harmonic approximation is the phonon DOS.e. k T  B +1 .52) The constant factor of kB 1 will cancel when taking differences of phonon entropy. but differ in energy scaling by 10%. In the classical limit. so a more accurate critical temperature can be obtained for the phase transition. until recently accurate computations of phonons were impractical.51) (7.8 Bond proportion model If two phases differ in their energies by E and entropies by S. so although there were many suggestions over many years that vibrational entropy could have a significant effect on the thermodynamics of unmixing or order–disorder transitions. a 1% accuracy in the difference in logarithmic-averaged phonon spectrum is desirable. Suppose that Tc has been evaluated for the difference in configurational entropy alone.and β-phases have the same shape.1) 0. if Eq. 12 For example. Calling it Tcconf gives E = Tcconf Sconf . Now if. the critical temperature is changed to Tcconf+vib . determined by the condition E = Tc S. the change in vibrational entropy is Sph = 3kB ln(1. methods are more mature for evaluating E and Sconf for configurations of atoms on a lattice. This change in vibrational entropy is almost half of the maximum possible change in configurational entropy of an order–disorder transition of a binary alloy (which is kB ln 2/atom). (7. a difference in vibrational entropy Svib is added. and this has been a challenge until relatively recently. that for most problems g(ε) must be known with high accuracy. this new critical temperature is related to the critical temperature with configurational entropy alone as Tcconf+vib (Sconf + Svib ) = Tcconf Sconf . Today. Measuring or calculating the phonon DOS is central to much of our understanding of the vibrational entropy of materials. 7. F = E − Tc S = 0) at the critical temperature Tc .53 is used for a case where the phonon density of states curves of the α.50 β−α Svib ∞ = 3kB (gα (ε) − gβ (ε)) ln(ε) dε . though.3 kB /atom. for the same E. It turns out. g(ε). A handy expression for the high-temperature limit of the difference in vibrational entropy between two harmonic phases.

and with more risk the stronger bonds can be associated with the stiffer springs. and both the bond strength in t Fig. (b) More realistic model of atoms and springs. with each atom attached to a heavy mass.13 We begin with the “Einstein model. This adapted Einstein model partners well with the point approximation that was used for configurational entropy in Sections 2. The disordered phase has more bonds that are energetically unfavorable.9 conveys the nature of this approximation. A comparison is shown in Fig. We adapt the Einstein model so that the stiffness of each spring in Fig. 7. a disordered phase with smaller interatomic spring constants. the bond proportion model is rarely adequate. even those in its 1nn shell.9a depends on its neighbors. 7.10. By Fig. This complicates the problem. 7.10 for the bond strength and bond stiffness of a central A-atom by considering its four 1nn atoms. however. is called the “bond proportion model. the oscillator coordinate of any atom is independent of the vibrations of all other atoms. and counting these bonds. 13 Unfortunately.” It has a long history [66]–[71].9b. This reasoning is enough to explain qualitatively the effects of vibrational entropy on the critical temperature.6 and 2. 7.1 Local approximations for bonds and springs The vibrational entropy of a phase transition depends on changes in bond stiffnesses (force constants). and therefore has weaker springs between atoms. increasing its relative stability.” which considers the motion of each atom as an independent simple harmonic oscillator.9 a b (a) Einstein model. because the tension in each spring then depends on the positions of two atoms instead of one. Local atomic configurations alter the bond stiffnesses of atom pairs.6. has the larger vibrational entropy. as in Fig. Assigning a fixed stiffness to a particular type of chemical bond. which can be used to develop the Born–von Kármán model of lattice dynamics. 7. for example. 7. where a disordered phase at high temperatures transforms upon cooling to either an unmixed or ordered structure. and this is the usual case for real alloys. 24. Two neighbors are A. It is of course more realistic to assume each atom is attached to its first-nearest neighbors (1nn) by springs. so a natural first approach to estimate the vibrational entropy of a phase transition is to count the different types of bonds between neighboring atoms.9 discusses such effects parameterized by bond lengths and materials chemistry. 17:28:36 . k. including effects of temperature. Figure 7.6a.Entropy 180 7. for quantitative predictions of the vibrational entropies of alloy phases. and how they change. In the original Einstein model of 1907 [72].9a. With some risk. The critical temperature for disordering is therefore expected to decrease when vibrational entropy is added to the configurational entropy. the accounting can be confined to first-nearest neighbors (1nn). 7. and Sect.8. The Einstein model assumes each atom is attached by a spring to a rigid point as in Fig. Chemical effects are also discussed in Sect. and two are B.

7. A hexagonal lattice with z = 6 requires a further correction. For a fixed n = cN.14. This is consistent with our approach of adapting the Einstein model for vibrational dynamics to the thermodynamic point approximation. For two-dimensional dynamics. In this approximation.7.49 for the hightemperature limit of the vibrational partition function. 7. since there cannot be three normal modes per atom in two dimensions. This correction is discussed below. This problem was first illustrated with Fig. is Z1site = e−βeA + e−βeB . There is a new feature with the vibrational frequencies. ZN = Z1site (7. N! kB T 3N g(ωj ) ω e B .10 a A A A 2eAB + 2eAA A B B A 2ωAB + 2ωAA B b (a) Point approximation for the bond strength of the A-atom in the center. each atom vibrates 14 Equation 7. (b) Analogous approximation.55 plus Eq. 75] addressed this problem.56 needs products of all the frequencies. {γij }.56) e A ZN = (N − n)! n! ωj j=1 To evaluate the last factor in Eq.14 Using the adapted Einstein model of Sect. 7. Assuming all site occupancies are independent. Mahanty and Sachdev [73] and Bakker [74.8 Bond proportion model 181 A B t Fig. the partition function for the vibrational frequencies. Z1site . 7.1.10a and the bond stiffness in 7. only one term of the binomial expansion is selected  m  −βe N−n −βe n . 2. the number of normal modes is two per atom.55) This ZN can be evaluated by the binomial expansion of Eq. we invoke a physical correspondence between nearest-neighbor pairs and interatomic force constants. i − j. 7.54) where β ≡ 1/(kB T). and showed for some crystals how the product of the {ωj2 } equals the product of all interatomic force constants. (7. divided by the product of all atom masses. the total partition function of the alloy with N sites is N .8. (7. 7.2 Bond proportion model and unmixing on the Ising lattice Returning to Eq.10b are proportional to these numbers through parameters such as eAB or ωAB . and solved by dividing the total bond strength or stiffness by 2. 17:28:36 . 7. This works well for the square lattice where dividing the coordination number z by two for the overcounting gives two modes per atom. Fig. between atom pairs. there is an issue of overcounting of bonds and spring contributions when performing sums over all atoms in the lattice.56. 7. however.8. the partition function for a single site. which is constant in a closed system. where the spring stiffness depends on the average first-nearest neighbors. however. adapting the Einstein model for vibrational frequencies. 2. not the individual frequencies.

The correspondence works naturally for a linear chain. For a simple cubic lattice with z = 6.e. ηNBB .58) where dividing by 2 avoids double counting. each atom has z/2 neighbors (bonds are not overcounted). for which each atom is associated with one. 2 (7. 7. B–B. and z/2 = 3 springs (normal modes are not overcounted). Likewise. two. The vibrational factors in Eq. we correct the mode counting with the ratio η = 2D/z. scaled to conserve modes. i and j. giving   −βe N−n −βe n N! e B e A ZN (c) (N − n)! n!       kB T ηNAB kB T ηNAA kB T ηNBB × . with a frequency ωij that depends on the 1nn pair of atoms. Eq. Their exponents for the numbers of modes are replaced by the numbers of bonds of types A–B. ωAB }. so NAA = (1 − c)2 zN . the average number of A-atom neighbors of an A-atom is (1 − c)z. and using the Stirling approximation ln(x! ) x ln x − x. When the coordination number z is not the same as 2D. ωBB . Obtaining the Helmholtz free energy from the expression F(c) = −kB T ln Z(c).57 gives . i. A–A. NBB = c2 zN/2 and NAB = c(1 − c)zN (where A–B and B–A bonds were added together). for example. ηNAA . where D is the number of dimensions.Entropy 182 independently in three dimensions with three independent springs to its neighbors. ηNAB . or three interatomic springs. (7. square lattice.57) ωAB ωAA ωBB To obtain NAA for a random solid solution in the point approximation. which is 1/2 for an fcc crystal. 7.56 now include only the frequencies {ωAA . or simple cubic lattice..

where .59) F(c) = Econf (c) − T Sconf (c) + Svib (c) . (7.

zN . Econf (c) = eAA + 2c(eAB − eAA ) + c2 (4V) .

2 Sconf (c) = −kB c ln c + (1 − c) ln(1 − c) N . zηN .

Svib (c) = kB LAA + 2c(LAB −LAA ) + c2 (LAA +LBB −2LAB ) 2 2 . .

ωAA ωAB ωAA ωBB (7. 7. 17:28:36 . Sconf (c).62 factors were defined such as LAB ≡ ln(kB T/(ωAB )).15 This kB LAB is the vibrational entropy of an oscillator of frequency ωAB . In Eq.45.62) (7. configurational entropy. 15 In the classical limit of the Planck distribution. n(ε . 7. and vibrational entropy of the random alloy.62 show that Econf (c) and Svib (c) have identical quadratic forms in c because their energies or interatomic force constants depend in the same way on the nearest-neighbor bonds. Equations 7. ωAB ωAA kB T Svib (c) = kB ln + 2c ln + c2 ln DN .61) (7. Eq.60 and 7.38 in the classical limit. this L AB AB is the logarithm of the number of phonons of frequency ωAB .63) Here Econf (c). 7. T) of Eq. and Svib (c) are the configurational energy.60) (7.

positive when V < 0. 7. 2. The energy and entropy terms of Eqs. with the result kB Tc = 1 c(1−c) 4 |V| z  . Energy [kT/atom] E(c) 1. Econfig (c) for unmixing.4 0. and ωAA < ωBB . promotes unmixing at intermediate concentrations at low temperatures. i. representative Svib (c) 2 < ω ω (Eq. is found by setting ∂ 2 F/∂c2 = 0. 2.2 0.63). The free energy without vibrational entropy.2 −0..5 Stot(c) 1.8 0.2 2.4 0.6 Concentration (c) 0. It is straightforward to do this with Eqs. V < 0 (Eq.61.5 without vibrational entropy). 7.2 b 0. The critical temperature for chemical unmixing.11b shows curves of the total free energy versus composition for different ratios of vibrational frequencies.60 through 7.5 ωAB2 > ωAA2ω B B2 0. 2 =ω • When ωAB AA ωBB .8 1. Fconf (c). + 2D ln ωAA2ωBB (7.6 ωAB2 = ωAA2ω B B2 0. Figure 7.61). 7. showing effects of different ratios of ωAB AA BB For chemical unmixing. than A–B pairs).8 Bond proportion model 183 1.4 0.5. The logarithmic singularity of the derivative of Sconf (c) always forces a downwards slope away from c = 0 and c = 1.63 are √ graphed in Fig. The middle curve in Fig. 7.0 F = E − TS Free energy [kT/atom] Entropy [k/atom]. 7.0 1.0 (a) Sconf (c) for random solid solution (Eq.64 we can deduce that 2 .8 Concentration (c) 1.0 ωAB2 < ωAA2ωBB2 −0.e.24 this means that the average of A–A and B–B pairs is more favorable. 7..6 0.36 without vibrational entropy.0 t Fig. as may be plausible if eAA > eBB . The energy term. It was assumed that ωAB = 0. and increased when • The Tc is reduced by vibrational entropy when ωAA ωBB > ωAB 2 ωAA ωBB < ωAB .63. Eq.60). has a negative 17:28:36 . 7.7.4 0. their energies are more negative.0 0. Tc . consider the case where V < 0 (from Eq.11a.11 a 0.11b shows F(c) for this condition.64) ωAB From Eq. the previous result of Eq.77 ωAA ωBB . 7. condition at c = 0. 7.e. 7.0 0.64 provides Tc = z|V|/kB for c = 0.2 0.60. and 7. and S (c) = S (c) + S (c).0 Svib(c) Sconf (c) 0. (b) Free energy for zV/(kT) = 1 (critical when ωAB AA BB tot conf vib 2 to ω ω . 7. i.0 0.

8. Svib (c) is a quadratic function of c (Eq.63).513 ωAB > ωAA ωBB ).5. β}. Figure 7. B}. but this may change with temperature. much like • When ωAB 16 Econf (c). each having N/2 sites for a crystal with N sites. NAα = β β NB ≡ R. A on α B on β A on β B on α t Fig.12 identifies the α. suppressing disordering. by conservation of atoms and conservation of sublattice sites. 17:28:36 . NB }. 17 A negative entropy of mixing has been observed in alloys of V–Pt at surprisingly low concentrations of Pt [76]. characteristic of spinodal decomposition in the point approximation. this alloy is expected to become a random solid solution if the configurational entropy is dominant. reducing Tc . the vibrational entropy term −TSvib (c) effectively diminishes Econf (c). in exchange for B-atoms on the α-sublattice. where R denotes atoms that are on the “right” sublattice.3 Bond proportion model and ordering on the Ising lattice For chemical ordering. where for example NAα is the number of A-atoms on the α-sublattice. Dashed and solid lines represent two interpenetrating square sublattices {α. For a specific ordering transition.Entropy 184 curvature in c below Tc = z|V|/kB .61 near c = 0.18 driving some mixing of A-atoms onto the β-sublattice. At high temperatures. This is conceptually convenient.12. Finite temperatures favor some disorder in the structure – placing a few atoms on the wrong sublattice gives a big increase in the configurational entropy. and NA = NBα ≡ W. disordering cannot occur at any temperature at c = 0.24. 16 A simple approximation is that the strengths of the bonds and spring constants are proportional.and β-sublattices. NA . NBα .17 7. the entropy of mixing quickly becomes negative for intermediate compositions. 7.12 Ordered square lattice with two species {A. 7. we use an equiatomic A–B alloy on the square lattice shown in Fig. but it has weak predictive power. A few misplaced A and B atoms are shown. 7. If ωAB AA ωBB (approximately 0. giving S (c) vib the same shape as Econfig (c). 2 = ωAA ωBB . 2 < ω • When ωAB AA ωBB . 2 > ωAA ωBB . 7. Figure 7. Tc is raised and the whole unmixing phase boundary moves • When ωAB 2 exp(−2/3) > ω 2 to higher temperatures. 18 The entropy has a logarithmic singularity in the derivatives of its sublattice concentrations analogous to the concentration dependence of Eq. we consider the case V > 0 of Eq.12 motivates the definition of four types of sublattice concentration variables β β {NAα . For an equiatomic alloy. so the energetics tends to maximize the number of A–B pairs at low temperatures. For yet larger 2 with respect to ω ωAB AA ωBB . 2.

70 works naturally with η = 1.65 is used with the conservation of sites (R + W = N/2) to remove all instances of R.67) Using Eqs. 7. To determine the numbers of pairs NAA .65) This L can range from −1 to 1. The case of L = 1 corresponds to perfect order where R = N/2 and W = 0. 8 zN = (1 + L2 ) . The partition function depends on the LRO parameter. giving Z(L).68) (7. 7.10.8 Bond proportion model 185 denoting “wrong” atoms. irrespective of the presence of A-atoms or B-atoms on neighboring sites. 7. For the square lattice Eq. N/2 NAB = (R2 + W 2 ) (7.  ωAB e−β eAA  (7. 7. the sublattice concentrations are used to define a “long-range order” (LRO) parameter.70 are replaced with Eqs.e. 7. The disordered alloy has as many right as wrong atoms on each sublattice. the partition function has a combinatoric factor from the random occupancies on two sublattices (giving two binomial probabilities) and three factors that depend on the numbers of bonds  Z = (N/2)! R! W! 2        kB T η NAA −β eBB kB T η NBB e  ωAA  ωBB   η NAB kB T × e−β eAB . i. NBB . assume that the atoms are placed randomly on a sublattice.10.. L. the number of vibrational modes must be consistent with the number of pairs. To evaluate NAA . 2. where D is the dimensionality.68 and 7. Similar to the assumption of a random solid solution. and multiply by the probability of finding a wrong A-atom in each of its z neighboring sites on the β-sublattice (which has N/2 sites) NAA = NBB = RW z . 7. 7. As in Sect. so the exponents in Eq. 13.. W. 17:28:36 .12 is that the first neighbors of each atom on the α-sublattice are on the β-sublattice. 4 NAA = NBB = (1 − L2 ) NAB (7.69. This is the “point” or “Bragg–Williams” approximation. but for other lattices it is necessary to avoid incorrect counting of the 3N normal modes by setting η = 2D/z. and Eq. an important point evident from Fig.7. NAB in Eq.69) In the high-temperature limit for phonons.g. L.70) As for Eq.70. R = W. for the alloy [11. NAB . N/2 (7. so L = 0 for a disordered alloy. 7. NAA . the NAα A-atoms have no preference for any site on the αsublattice.65 through 7.57. NBB .66) z . 2. e.2). This approximation makes it easy to count the bonds between the different species of atoms. and vice versa (Sect.67 zN . pick one of the R A-atoms on the α-sublattice. N/2 (7. 77] L≡ R−W .

7. In 2 =ω the absence of vibrational entropy.19 The result is kB Tc = zV  .1 ω • In the case where ωAB AA ωBB . F(L) ≡ F(L) − F(L = 0) is zN 4V 8 N −T kB [−(1 + L) ln(1 + L) − (1 − L) ln(1 − L)] 2   ωAA ωBB 2 DN kB ln . Eq. 2 < ωAA ωBB . so Eq. −T L 2 4 ωAB F(L) = −L2 (7. V. 7. 19 This assumes that the ordering transition remains second order. and −TSvib . which decreases continuously from L = 1 to L = 0 at Tc . 7.46 by the extra term from vibrational entropy. 7. 2. 2 = 1.72 for all T. This suppression of Tc seems to be the usual effect of vibrational entropy. giving a transcendental equation for L(T)     2 ωAB 1+L 2LzV = kB T ln + LD kB T ln . vibrational entropy suppresses from physical intuition. The trick is to assume L is infinitesimally small. Tc .73) The configurational entropy is always important in a chemical order–disorder transition. ωAB AA ωBB . the well-known result of Bragg–Williams theory (Eq. −TSconf . Subtracting the free energy of the disordered solid solution.. 17:28:36 .72 can be linearized.72) This differs from the usual Bragg–Williams solution of Eq. There is no analytical solution to Eq. the free energy of ordering. 7. and Tc = ∞. from Eq. 2 ωAB D 1 + 2 ln ωAA ωBB (7. if ωAB AA BB disordering.73 we find 2 >ω • If ωAB AA ωBB the Bragg–Williams value of Tc is reduced. Although difficult to expect • The opposite effect is possible if ωAB 2 /ω ω < exp(−2/3). i.49).Entropy 186 The free energy is then F(L) = −kB T ln Z(L). This term is linear in L. is rescaled monotonically with Svib – the effective V can be considered as altered by a temperature-dependent term. so it can be considered as a temperaturedependent rescaling of the interatomic exchange energy. which is true for Eq. The curve of L(T) from Bragg–Williams theory. respectively. perhaps significantly. When both vibrational and configurational entropies are important.71 are identified as the changes upon ordering of Econf . for a simple cubic lattice Tc is reduced to 87% of its value without vibrational entropy. 7.e.73 reduces to Tc = zV/kB . In equilibrium ∂F(L)/∂L = 0. but it is possible to obtain an analytic solution for the critical temperature.71) where the three lines of Eq.72. 2. 1−L ωAA ωBB (7.

A change in interatomic distance of +10% could cause the interatomic force to be lost entirely. 80].20 Counting the proportions of atom pairs is a first step in comparing the vibrational entropies of two different structures. only 1nn bonds are transferred between different structures [66. ω ∂V (7. Consider a disordered solid solution with a distribution of interatomic distances.1 Phonon frequencies and bond lengths The bond proportion model just described. Quantitative analyses of the effects of bond length on bond stiffnesses were performed by Sluiter et al. The calculation of these distances is not simple because the relaxations in position for each atom depend in turn on the relaxations of its neighbors.74) where ω is the phonon frequency and V is volume. In the “bond-stiffness-versus-bondlength” approach. (7. owing to the different atomic sizes.187 7.9 Bond-stiffness-versus-bond-length model 7. It is also observed that only the stretching (longitudinal) forces between atoms have a sigificant dependence on the bond length.9.75) There is no universal value for a Grüneisen parameter. Eq. The Grüneisen parameter γ is defined as γ ≡− V ∂ω . Recognizing that V ∝ r3 and ω ∝ where ks is an interatomic force constant 6γ = − r ∂ks . This is not a problem for axial force models. There is some subtlety to the concept of transferrability of force constants because different symmetry operations for different crystal structures make it difficult to transfer tensorial force constants from one structure to another.9 Bond-stiffness-versus-bond-length model 7. but it is also important to consider how bond stiffness changes with bond length. shorter and longer than optimal. ks ∂r √ ks . on Al–Li [78] and by van de Walle and Ceder on Pd–V [79]. the interatomic force changes by –12%. 17:28:36 . 20 This same characteristic of the interatomic potential also gives rise to quasiharmonic theory. and the stiffness of a bond is sensitive to its length. The bond lengths between pairs of atoms are different in different structures. The bond bending forces are smaller and less dependent on bond length. 7. This is a major effect. with the goal of predictive power without major calculations or measurements on every material. but typically γ 2. where interatomic forces weaken with thermal expansion. which counts the different types of atom pairs. has a deficiency that is easy to understand.75 predicts that if the interatomic distance changes by +1%. In this case. Nevertheless. which involve a central-force “stretching” force and a “bending” force. this effect on phonon frequencies can be estimated. and axial force models often give adequate descriptions of lattice dynamics. They tested the idea of transferring interatomic forces from one material to another.

F(0) = 0).2 Extending the bond proportion model For a chemical ordering transition. and by simplifying the three curves of Fig.13 3 2 1 V−V 0 4 3 2 1 Pd−Pd 0 3 2 1 0 Pd−V −1 2.13. show the longitudinal (stretching) interatomic forces between V–V. there is a range of about 10% in the interatomic separation where the force constants vary from their maximum value to zero.74). compositions.9. Pd–Pd. 7. and lattice parameters. and when calibrated for specific elements by ab initio calculations. Some results. The bond-stiffness-versus-bond-length model is useful for semiquantitative arguments.. The free energy difference F(L) of Eq. 7. and there is only one bond length.e.73) of the bond proportion model. 7. 7. and Pd–V 1nn pairs of atoms for different structures. Reproduced. The effects of bond length on bond stiffness of Fig.5 2.71) and the critical temperature (Eq. 7. 7.13 by 17:28:36 .8 2. from [79]. Also approximately. all 1nn bonds are between A–B pairs.0 ˚ Bond length [A] Stretching terms of the nearest-neighbor force constants as a function of bond length of Pd–V alloys. In this structure. This is a complex problem owing to the different relaxations in the different local environments. Each point corresponds to one type of bond in one of a set of fcc crystal structures of different volumes. 7. We can get to the essence of the problem by using the same Grüneisen parameter for all bonds (Eq. Approximately.73. as discussed following Eq. 7. and is zero for L = 0 (i. The points tend to fall rapidly with bond length. presented in Fig. Van de Walle and Ceder performed density-functional theory calculations of bond stiffnesses for a set of fcc-based Pd–V structures with different interatomic distances. the stretching stiffness has a linear relationship with bond length.13 are large. It is the disordered solid solution where the distortions of bond lengths occur. with permission.7 2. we use the bond-stiffness-versus-bond-length relationship to improve the free energy (Eq.9 3.Entropy ˚ 2] Stiffness [eV/A 188 t Fig. as expected when the interatomic forces soften with expansion. so a Pd–Pd bond cannot be expected to have the same stiffness in any local environment of a random solid solution. for example. Consider again the case of B1 ordering (NaCl prototype) of an AB alloy on a simple cubic lattice.71 is with reference to the disordered solid solution. 7. this model can be used for comparing phonon entropies of different alloy phases.6 2. 7. for example. Copyright (2000) by the American Physical Society.

but grows with disorder as the number of R–W pairs increases as 1 − L2 . 7. For small δV/V 0 ln(ωij ) ln(ωij0 ) − γ δV .77) To the free energy of Eq. Such effects may scale with the electronegativity difference between A. Assuming γ = 2. 21 In the less common case where the lattice parameter of the disordered alloy is smaller than the ordered. 7.73 becomes kB Tc = zV  .79 is +0. For example.13 shows that the bond stiffnesses for V–V and Pd–V pairs have considerable scatter about the linear fit.9.78) which vanishes when L = 1. ωBB . this effect of bond stiffness versus bond length is independent of the effect on Tc from stiffness differences of the different pairs of atoms.9 Bond-stiffness-versus-bond-length model 189 assuming the bond stiffnesses are zero at the same bond lengths for A–A. V where the superscript 0 denotes a reference value for the ordered structure.79) Consider a typical case where the disordered phase has a lattice parameter 1% larger than the ordered phase (so δV/V 0 0.03).21 Because we assumed the same Grüneisen parameter for all bonds. Some of this can be accounted for by a dependence of bond stiffness on chemical composition. In a more complete treatment these effects are interrelated. all three frequencies {ωAA . The electron screening of the A-ion cores is thereby altered.3 Chemical effects on bond stiffness A closer examination of Fig. 7. causing changes in the density of electron states at the Fermi level. 2 ωAB D 1 + 2Dγ δV + ln 2 ωAA ωBB V0 (7. V0 (7.7. The critical temperature is suppressed substantially when the average bond length in the disordered solid solution is 1% longer than for the ordered phase. V0 (7. the second term in the denominator of Eq. 7. These chemical effects are probably less significant for alloys of isoelectronic elements. T c would be raised. and the density of electron states at the Fermi level. and A–B pairs. In this case. ωAB } change with volume as   δV 0 (7.76) ωij = ωij 1 − γ 0 .and B-atoms. the addition of B-atoms to a host crystal of element A will fill or deplete the electronic band structure of A. changing the bond stiffnesses of A–A pairs. 7. B–B.36.71 we add the term − TSGruen = kB T (1 − L2 ) DN γ δV . 17:28:36 . The critical temperature of Eq.

BAB. is essentially the same as the bond-stiffness-versus-bond-length model. BBB}. EPI). Problems 7.” which assumes independent phonons with altered frequencies. so show that the product of all these factors equals αβ (zN/2)! /(zN/2 − NAA )!. BAA.9. Of course we could have filled these bonds in any order αβ with A–A pairs.) 7.1 The combinatoric entropy becomes more tricky to evaluate for larger clusters of atoms.Entropy 190 7. so there are zN/2 − 1 ways to put αβ it down. and assign the first A–A pair to one of them. however.4 Temperature and anharmonicity The harmonic model explains many physical phenomena. Each pair goes down independently of the others. but the triplets overlap between sites. and we need to divide by this factor. The quasiharmonic model. The second pair cannot go into the first bond.3 Derive Eq. AAB. (Hint: Here is an approach to counting the ways of assigning pairs to each bond. (Hint: It may be convenient to consider how many ways a new atom can be added to the end of a finite chain. which gives the temperature dependence of short-range order in the absence of long-range order. of which there are zN/2 in the crystal. αβ αβ αβ A marvelous simplification at the end occurs because zN/2 − NAA − NAB − NBA − αβ NBB = 0.3. The existence of phonons alters the energies required to create additional phonons (phonon–phonon interaction.3 and Chapter 24. which is the same result obtained in the point approximation. especially at low temperatures when the phonons are approximately independent of each other. and independent of other sources of dynamical entropy. ABB. Start with an empty set of bonds. 10. and the electron excitations are altered by vibrational excitations (electron–phonon interaction. The last A–A pair can go down zN/2 − (NAA − 1) ways. 7.2. PPI). BBA. The A–B pairs go next. 17:28:36 . so we overcounted by the factor of NAA !. At high temperatures.15.) 7. Prove this. It is often extended to higher temperatures as the “quasiharmonic model. 7.2 Derive the combinatorial factor in Eq. the different sources of dynamical entropy can interact in the important ways discussed in Sect. ABA. Be brave and continue through the B–A and B–B pairs. At each site we might expect to have eight possibilities for a triplet.2 and 24. discussed in Sections 8. An understanding of vibrational entropy at high temperatures is still emerging. The result is that the random solid solution has a configurational entropy of kB ln 2 per atom. Consider all eight possible nearest neighbor triplets of atoms on a linear chain for a binary alloy: { AAA. but they start with only a number of bonds αβ (zN/2) − NAA to occupy. There are zN/2 ways to put down this first pair.

pure element B. at what temperature is the vibrational entropy equal to the highest value of the configurational entropy from part b when ω = 400 K? (e) The temperature of part d is a bit low for the high-temperature expression for n(T).39 to compare the vibrational entropy and configurational entropy of an alloy with a composition c and a characteristic phonon frequency ω. we first use three alloys to calibrate the interaction entropies of first-neighbor pairs. . . (a) Show that the mathematical forms of Eqs.0693 kB /atom? Same question for 0. 7. and what is this value? (c) Show that the Planck occupancy factor becomes n(T) = kB T/ω at high temperatures. from which we can calculate the entropy of disordered alloys. with S = 0. and the other has ω = 440 K kB .5 This problem uses the cluster expansion method to evaluate the entropy of a disordered solid solution in one dimension.38 and 7.40 kB /atom. . don’t forget that there are three times as many vibrational modes as the number of atoms.) (g) Two materials have different characteristic phonon frequencies – one has ω = 400 K kB . an ordered equiatomic A–B alloy (having structure .95 kB /atom.693 kB /atom? 7. (b) What composition has the largest value of configurational entropy. 7.39 are the same for small n and small c. What is the effect (up. (a) Using σ = +1 for A-atoms and σ = −1 for B. n(T) = kB T/ω. at a temperature of 800 K. no change) on your result of using the full expression for the Planck occupancy? (f) For ω = 400 K.38 and 7. (You can use the high-temperature approximation of part c. . ABAB. calculate the vibrational entropy. At what temperature will their difference in vibrational entropy exceed 0. with S = 0.4 The following questions ask you to use Eqs.191 Problems 7. ) with S = 0. For questions about vibrational entropy. down. and compare it to the maximum value of the configurational entropy of a binary alloy. The three samples with known atom arrangements and known entropies are: pure element A. calculate the correlation function ξ0 = σn0 . (d) Using the simple approximation for the Planck occupancy at high temperature. In this problem.75 kB /atom.

ξ1 = σn . = 1 for the empty lattice.

ξ2 = σn σn+1 . for the point variables.

and ξ3 = σn σn+1 σn+2 . for first-neighbor pair variables.

80) where the three Sj were given to you above.i si . ξ1 . for all three alloys. (Use only ξ0 . for triplets of neighboring atoms. (d) What is the mathematical form for the correlation functions ξ0 = 1. You have already calculated the required nine elements of ξj. and ξ2 . but the si are unknown.) (c) Invert the matrix of part b. and obtain the three sj . (7.i in part a – now put them in the correct places in the matrix. not ξ3 . (b) Set up the 3 × 3 matrix equation Sj = ξj.

ξ1 = σn ..

ξ2 = σn σn+1 ..

. and ξ3 = σn σn+1 σn+2 .

for a disordered solid 17:28:36 .

D.60. available here: http://thesis. and can be useful as a method for estimating the age of minerals. 7. and preferably further.25.1. thesis.61. 7. 7. make a graph of Tc for unmixing versus R for an equiatomic alloy. assume that atoms of different isotopes differ only by their nuclear mass. What is the fractional change of Tc over this range? Give a physical explanation for why Tc changes as the A–B bonds become stiffer than the average of the A–A and B–B bonds. and x = 0. 7.9 to 1. (b) Starting with the partition function of Eq. (a) At what concentration of Pt in V will the addition of an additional Pt atom cause zero change to the total entropy (the sum of configurational plus vibrational)? (b) How does this concentration change if the amount 0.caltech. (b) Show that the zero-point energy of these two phases can be lowered by isotopic segregation to one phase or the other. and why it appears only in the factors involving temperature and vibrational frequency. The configurational entropy of chemical mixing increases with the addition of Pt to V (at least for Pt concentrations c < 0.5).7 7.PDF) The vibrational entropy decreases when a Pt atom is added to bcc V by the amount 0. (a) What is the physical reason why the first factor with the factorials is squared? (b) Equation 7. Use appropriate partition functions for the harmonic oscillators. where 0 ≤ x ≤ 1? Using the three sj obtained from part c.8 7.41. including the correlation functions of part a. Eq. x = 0. derive Eqs. Lucas in his 2008 Caltech Ph. in place of the three 17:28:36 . Consider equal amounts of two phases in equilibrium at temperature T.7 to 1. Be sure that your range of R extends from 0.35kB /(Pt atom) (comparing the vibrational entropy of the pure elements and the Pt–V alloy at 300 K).9 (Hint: Much of this is worked for you. (d) Usually the R defined in part c is in the range from 0.Entropy 192 solution of B-atom concentration x. In this problem.3.edu/3032/2/ Lucas_final_cor.70 needs to be modified at low temperatures.35 kB /(Pt atom) is doubled? Isotopic fractionation occurs in materials over geologic times.6 7. by Matthew S. and 7. Do heavy isotopes prefer the phase with the stiffer or softer phonon spectrum? (a) Explain the role of η in Eq.library. 7. what are the entropies of disordered solid solutions of concentrations x = 0.57. 7.63. (a) Using high-temperature expressions for vibrational entropy.57. This question refers to Eq.75? 7.50. show that there is no preference of the heavy or light isotopes for the phase with stiffer interatomic force constants. Consider the total free energy or energy to answer these two questions.70. * 2 /(ω (c) Defining R ≡ ωAB AA ωBB ) as the ratio of vibrational frequencies.

such as a change in chemical order. B–B. would this violate the third law of thermodynamics? 7. even though they have the same R = 1. compared with what would be expected from the alloy properties at low temperature? (b) Suppose two disordered equiatomic alloys differ in their thermal expansion and Grüneisen parameter. ii. Show that the vibrational entropy becomes independent of temperature as T → 0. A typical Grüneisen parameter is 2.5 and a larger linear (not volume) thermal expansion of 2. and a typical coefficient of linear (not volume) thermal*expansion is 2.0. iii. (a) For a temperature change of 500 K. would there be a difference in the vibrational entropy at T = 0? If so.5 × 10−5 /K. and A–B bonds. but the second has a larger Grüneisen parameter of 2.10 The arguments in this problem are typical of the “quasiharmonic approximation” for nonharmonic behavior at elevated temperatures. Suppose the ratio of vibrational frequencies R is R ≡ ωAB AA ωBB ) = 1. If there were a change in the numbers of A–A.) i.0 × 2 /(ω 10−5 /K. Suppose the first alloy has the properties given above.193 Problems frequency factors with form kB T/ω. what is the effect of thermal expansion on the critical temperature for ordering in an equiatomic alloy. How much more vibrational entropy is expected of the second alloy compared with the first for a temperature increase of 500 K? 17:28:36 . the vibrational part of the partition function is dominated by the zero-point contribution of half-occupied states. Show that.05.05. as T → 0. (Do not confuse this with the Planck distribution.

Recently there have been rapid advances in high-pressure experimental techniques. whereas temperature is complemented by entropy. as opposed to calculating all different sources of entropy. The situation is reversed for computational work. depends on the thermodynamic intensive variables T and P. and how phase diagrams are altered by temperature and pressure together. 8. Nevertheless. The electronic energy accounts for most of the PV contribution to the free energy. it is useful to compare gases with solids to see how the thermodynamic extensive variable. often driven by interest in the geophysics of the Earth. Chapter 8 begins with basic considerations of the thermodynamics of materials under pressure. Volume changes can also be induced by temperature. at least high pressures that induce substantial changes in volume. it is difficult to achieve pressures of thermodynamic importance. (8.1 Materials under pressure at low temperatures The behavior of solids under pressure. V. and some such as diamond can be recovered at ambient pressures. The chapter ends with a discussion about using pressure to investigate kinetic processes. and the meaning of an activation volume. as opposed to temperature. The thermodynamic variable complementary to pressure is volume. For experimental work on materials. 8. and there is a brief description of how electron energies are altered by pressure.1. so experiments at high pressures can point directions for materials discovery.1) 17:28:44 .8 Pressure Historically there has been comparatively little work on how phase transitions in materials depend on pressure. It is comparatively easier to calculate the free energy of materials with different volumes. is more complicated than the behavior of gases. Sometimes the changes in interatomic distances caused by pressure can be induced by chemical modifications of materials.1 Gases (for comparison) Recall the equation of state for an ideal gas comprising noninteracting atoms PV = NkB T . and the concept of “thermal pressure” from nonharmonic phonons is explained. New materials are formed under extreme conditions of pressure and temperature. whereas high temperatures are obtained easily. The use of pressure to tune the electronic structure of materials can be a useful research tool for furthering our understanding of materials properties.

1 is replaced by P + a(N/V)2 . and the force between them may also go as 1/V. 8. 8. The quantity V in Eq. This can be considered as a surface tension that pulls inwards on a group of gas molecules.2. • An attractive interaction between the gas molecules tends to increase the pressure a bit.1 is replaced by V − Nb. where b is an atomic parameter with units of volume.8. The quantity P in Eq.) The Van der Waals equation of state (EOS) is   N2 . if the correction went simply as 1/V. 8.1 Materials under pressure at low temperatures 195 Nonideal gases are often treated with two Van der Waals corrections: • The volume for the gas is a bit less than the physical volume it occupies because the molecules themselves take up space. The quadratic dependence of 1/V 2 is expected because the number of atoms affected goes as 1/V. (Also. it would prove uninteresting in Eq.

(8.9) At lower temperatures. 8. V2 .3) with the definitions b2 . (8.26a/b shown in Fig.1). 8. such as kB T = 0.4) a V V≡ .7) 27 b2 8 a kB Tc = . respectively. with P ∝ T/V (Eq.2 can be converted to a dimensionless form P= T 1 . the volumes V1 and V3 can be interpreted as the specific volumes of a liquid and as a gas.1 shows the Van der Waals EOS of Eq. but with varying temperature. V (8. We can ignore V2 because it is unphysical – at V2 an increase in pressure causes an expansion (and likewise. Nevertheless. a Figure 8. More interesting behavior occurs at low temperatures. (8.2 works surprisingly well for the gas phase when the parameters a and b are small and the gas is “gas-like. the material shrinks if pressure is reduced). We find the pressure that defines 17:28:44 .1a. P + a 2 V − Nb = NkB T . the same pressure corresponds to three different volumes V1 . 8. for which the volume is Pc = Vc = 3 b N .2 for a fixed a and b. This is point “C” in Fig.1b. It can be shown that below the critical pressure and critical temperature P≡P 1 a .8) 27 b a two-phase coexistence between a high-density and a low-density phase appears.5) Nb b (8.6) T ≡ kB T .” Equation 8. (8. − V − 1 V2 (8. 8. and V3 . At high temperatures the behavior approaches that of an ideal gas.2) Equation 8. (8.

(8.Pressure 196 0.2 V2 V3 –0. plotted with rescaled variables of Eqs.0 V1 20 0. for comparison with the characteristics of a solid. and is worthy of more study than given here. the solid has a finite volume. i. P ∝ T/V for large T.1.6. V.10) μ3 = μ1 + N V1 The integral must be zero if μ3 = μ1 . 1/N P dV. 8.1 –0.8. and this “Maxwell construction” defines the pressure of the horizontal line.2 Solids (for comparison) The ideal gas behavior shown at the top of Fig. the Van der Waals equations of state for all gases are the same. and varies for different gases.9 Pc Vc 3 = .26a/(bkB ). Table 8.1. Along a P(V) curve.7.2 0. the change in chemical potential is .. μ3 = μ1 . The areas above and below the horizontal line in Fig.1b  1 V3 P(V) dV .4 50×10−3 0. often around 0. 8.4–8. (b) Maxwell construction for T = 0. This instability of the Van der Waals EOS below a critical temperature can be used to model a pressure-induced liquefaction.1a.1 –0. but it gives the essential behavior. for example. V1 and V3 from the condition that the chemical potentials of the gas and liquid are equal in equilibrium.1b must therefore be equal. 8. The approach has problems with quantitative details.25 to 0.3 t Fig. Starting at a chemical potential of μ1 at the point V1 in Fig. 8. NkB Tc 8 (8. 8. At P = 0. 8.2. 8. 8. 8.3. for example.e.. i.e. Some characteristics of a generic gas are presented in Table 8.2 0 2 4 6 υ 8 0 10 b 0 2 4 υ 6 8 10 (a) Isothermals of the Van der Waals equation of state Eq.25 0.5 = t 40 0.11) Rescaled appropriately.1 17:28:44 . 8.1 8/27 ρ ρ C 0. or the liquid–gas phase boundary at constant pressure.4 30 0. This is approximately true in practice. is never appropriate for a solid.3 0. although the dimensionless ratio is lower than 3/8. A dimensionless ratio can be formed from Eqs.1 a 10 0.

the energy Eel increases as the square of δ.1 Materials under pressure at low temperatures 197 Table 8. 6. On the other hand. Like a spring. Equation 8. dP . 2 where δ is the fractional change in volume. where typical materials deform as springs. and not individual strains or stresses of Sect. since the bulk modulus increases with compression.1 Pressures and temperatures of gases and solids Pressure Temperature Stresses Typical pressure Gas Solid P>0 T>0 Isotropic 1 atm = 0.8.13) Eel = B δ 2 .13 can be handy if the elastic energy is needed. chemical bond energies. the range of elastic behavior is typically δ ∼ 10−2 . For a rod-shaped material loaded along its long dimension. In more detail. and it is possible to achieve δ ∼ 0. The challenge is that such compressions require pressures in excess of 20 GPa (2 × 105 Atm) for most inorganic materials. most materials will deform plastically. Figure 8. and will not return to their original shape. This gives an elastic energy density of order 107 J/m3 . such as by surrounding it with a liquid under high pressure (sometimes called “hydrostatic loading”). (8.12) dV is typically a few times 100 GPa. however. for example. It is possible to relate the interatomic force constants to the macroscopic elastic constants. suppose a material is loaded on all surfaces simultaneously.4. and the macroscopic response of the material is still that of a spring. These springs are loaded in different directions when different stresses are applied to a material. but all strains are linear with stresses. comparable to differences in energies of different crystal structures. the slopes of the curves are seen to be steeper with pressure. but this requires delicacy in accounting for the symmetry of the crystal and the geometry of the problem.1 or more. but not all.1 MPa P > −Pcoh T≥0 Anisotropic 10 GPa = 105 atm shows that on the scale of familiar pressures in gases. a solid is essentially incompressible.1 eV/atom.2 is a plot of pressure versus volume. and the elastic energy per unit volume under uniform dilation is 1 (8. Effects B ≡ −V 17:28:44 . which give tensorial “springs” between atoms as explained in Sect. 6. in this case for metallic sodium. This is small on the scale of many. or of order 10−3 eV/atom.3. making the curves seem like vertical lines at familiar pressures of hundreds of atmospheres. B. The elastic constants originate from second derivatives of the interatomic potentials. The elastic energy is of order 0. Beyond this small elastic limit. More familiar are small compressions of solids and elastic behavior. The bulk modulus of a solid. In this case it is possible to have compressions in all dimensions. Familiar pressures lie low on the y-axis.

Curves are offset by the effect of entropy.Pressure 198 Pressure (atm) 40.8 t Fig.000 30.3b).5 0 0. the solid expands with temperature.000 373 20.3a include the elastic energy. of temperature are also seen – as expected.0 V/Vo 1.5 1.5 0 F [kcal per mole] E [kcal per mole] 0 0 30 0 25 0 20 0 15 0 10 0 5 1. although less so when the pressure is higher. giving pressure versus volume at three different temperatures. after [81]. The equilibrium volume is instead at the minimum of the free energy. The curves in Fig. 8. Without the entropy. as labeled in K.9 1. the equilibrium volume of the crystal is not at the minimum of the energy. the three curves are extrapolated to negative pressures.0 200 –1. Reproduced.5 100 150 –1.1 –0. but also the phonon energy in the solid.0 V/Vo 1.5 V/Vo “Isothermals” for metallic sodium. (b) Free energy.5 250 –2. versus volume for metallic sodium and different temperatures. with permission. with permission after [81]. Reprinted. which includes the entropy of the solid at elevated temperatures (Fig. F. Somehow 17:28:44 . and have some reduction in curvature as the elastic constants soften with temperature. 8.2 0. 8. which is possible until the cohesion of the solid is lost.1 1.0 0.000 0 t Fig. Finally.9 1. Typical curves of E versus V for sodium metal are shown in Fig.000 273 10.2 (a) Energy.0 300 0. only the elastic energy would be minimized at temperature. The dashed line shows the equilibrium volume at the different temperatures in K. causing a vertical shift with temperature. E. versus volume for metallic sodium and different temperatures in K.3. 8.8 1. These curves are approximately parabolic owing to the elastic energy. 0. 8.3 a 0.2 b 0. At elevated temperature. F = E − TS.

21) (8.17) ∂V ω ∂V T T    kB V ∂ω ∂svib. In general the entropy will change with volume.” Pth . Confining the analysis to high temperatures. (8. the volume dependence of the entropy can be found through Eq. The dominant source of the entropy is from the phonons. however. This is called the “thermal pressure. Pth1 = The (dimensionless) mode Grüneisen parameter γ was defined as     V ∂ω ∂ ln ω γ =− =− .52.199 8.ω (T) to the entropy     kB T +1 . (8.15 is the “thermal pressure. 7. For a crystalline solid without magnetism or significant chemical disorder. E. 8. A single phonon mode makes the contribution svib.” Pth .24   ∂F . (8. (8. so thermal expansion alters the free energy.16) svib. Eel .2 Thermal pressure. 1.22) .ω (T) = −kB . (8. the source of this entropy is vibrational. V Pth1 V = γ kB T . increasing the energy.19 is the thermal pressure contributed by one phonon mode to kB T γ . From Eq.14) P=− ∂V  T   ∂E ∂S P=− +T .20) (8. 8. (8. The second term in Eq.2 Thermal pressure.15 was discussed in Chapter 6 as the elastic energy. a step beyond the harmonic model the entropy contributes a pressure to expand the solid against its elastic constants. 8.ω (T) = kB ln ω The only parameter in this equation that could depend on V is the phonon frequency     1 ∂ω ∂svib.18) ∂V V ω ∂V T T   kB ∂svib. Harmonic phonons do not change their frequency with the volume of a solid.19) ∂V V T When multiplied by T. ω ∂V T ∂ ln V T 17:28:44 (8. It accounts for how chemical bonds in a solid are distorted with changes in volume.ω (T) =− . so we need to go beyond the harmonic model to understand Pth .15) ∂V T ∂V T The first term in Eq.ω (T) = γ . but at finite temperature we should consider how the free energy depends on volume. 8. a step beyond the harmonic model At absolute zero temperature we can relate the pressure to the volume dependence of the internal energy. Eq.

Eq. 8. by adjusting the pressure. 8. 10.” This is an effective mnemonic.Pressure 200 which is the fractional change in phonon frequency per fractional change in volume. equal numbers of atoms in the two phases. P) . 8. (8. For a real solid with 3N phonon modes. and differs for different phonons and for different chemical elements. so it is tempting to attribute Pth to “the pressure of the phonon gas.3 Free energies and phase boundaries under pressure 8. of course. although dishonest as a physical explanation.24) assuming.23 is better understood as the “quasiharmonic model.23) j Compare Eq. 8. (8.3. the Gibbs free energies of two phases α and β are equal Gα (T.1. we can bring back the equality of Eq. They are of the same form.24. We seek the relationship dP/dT that gives the change in pressure required to maintain equilibrium for a change in temperature. P) = Gβ (T.1 Clausius–Clapeyron equation At a phase boundary.23 to the ideal gas law of Eq. . This number γ is not a fundamental constant.21 is modified as Pth V = kB T 3N  γj . However.3 and Chapter 24. a change in temperature alone will cause the phases to have unequal free energies. 8. Starting at an equilibrium phase boundary. The development of Eq.” described in Sect. but typically γ ∼ 2. 8.

⎞ ⎞ ⎛ . P) .25) 0 = d Gα (T. P) − Gβ (T. (8.

⎛ .

It is usually impractical to obtain calorimetric data on heats of phase transformations at high pressures.28) (8.24). P) ∂ Gα (T. P) ⎠ dT + ⎝ ⎠ dP . 1.29) where L = TS is the latent heat of the phase change.29 is the “Clausius– Clapeyron equation. P)−Gβ (T. Equation 8.   L 0 = − dT + Vα − Vβ dP . and     0 = − Sα − Sβ dT + Vα − Vβ dP . for constant n and T. ∂ Gα (T. P)−Gβ (T. T L dP  . 17:28:44 . but the 1 Because dG = μdN − SdT + VdP. ∂G/∂P = +V.26) 0= ⎝ ∂T ∂P P T The partial derivatives of G equal −S (Eq.” and is handy in high-pressure experiments. (8.27) (8. = dT Vα − Vβ T V1 (8.

(8.31 and 8. (8. 8. First with respect to temperature           ∂Sβ ∂Vβ ∂Vα ∂Sα − − dT + dP . (8. (8. but they have an analogous relationship between pressure and temperature along the phase boundary.201 8. (8. We need to take partial derivatives of Eq. 8. 17:28:44 .2 Ehrenfest’s equations The Clausius–Clapeyron equation is appropriate for finding phase boundaries of first-order transitions that have a specific heat. the isothermal compressibility.” useful for interpreting the slope of P–T phase boundaries for secondorder transitions. respectively.31) dT VT (ββ − βα ) and then with respect to pressure           ∂Sβ ∂Vβ ∂Sα ∂Vα 0=− − − dT + dP .3.     1 ∂V 1 ∂S β≡ =− .27 with respect to T or P. The subscripts α and β are useful to denote two different states of equal free energies.3 Free energies and phase boundaries under pressure latent heat L can be obtained from phase boundaries and volume differences between the phases.3 Characteristic pressure of a solid Suppose now we confine ourselves to consideration of the thermodynamics of a single phase. Finding a phase boundary in a P–T diagram at high pressure is no small effort. a second-order transition requires an equality of slopes of the free energy.32) (8. however.34) V ∂T P V ∂P T   1 ∂V . 8. It is still useful to compare states of the phase that have equal free energies when these states are at different combinations of T and P. = dT κ β − κα (8.35) κ≡− V ∂P T   ∂S CP ≡ T .30) 0=− ∂T P ∂T P ∂T P ∂T P CP β − CP α dP = . L.33) where the coefficient of thermal expansion. and how it changes with T and P. Second-order phase transitions are continuous and have no specific heat.33 are sometimes called “Ehrenfest’s equations for second-order phase transitions.3. ∂P T ∂P T ∂P T ∂P T ββ − βα dP . Instead of an equality of the free energies at the phase transition.36) ∂T P Equations 8. and the heat capacity at constant pressure are.

000 K. dT V ∂V T (8. materials behavior moves into unfamiliar territory.27 Sα − Sβ dP . the bulk modulus B originates at the level of the electrons in the solid. Importantly. The bond integral a|H|0. 8.19 gives (∂S/∂V)T for one oscillator mode. phase diagrams show phase changes over temperature ranges of 1. This also suggests that at pressures of 100 GPa and above. Sometimes the explanation is quite simple. Changes in crystal structure with pressure are probably the rule rather than the exception. 8. pressure favors structures that minimize the volume per atom. and for a solid of N atoms and 3N modes  dP dT  = V 3N kB  γj .45 of the tight-binding model. Calculating this pressure might be done with an approach explained after Eq. 6. Equation 8. it is often useful to approximate γ 2. = dT Vα − Vβ     dP ∂S = . Electronic energies are not always possible to understand with a simple isotropic model. bcc iron is driven to hcp iron under a pressure of 10 GPa. and why pressure alters the energies of electrons and their chemical bonds.40) dT V ∂V T K This type of scaling is also typical of how phase boundaries shift with pressure. (8. and V is of order 10 cm2 /mole.13. For materials like iron that have multiple crystal structures of similar energies of formation. We can understand a general feature of chemical bonding at high pressures with Eq.1.Pressure 202 following the previous analysis of the Clausius–Clapeyron equation. however. For example.4 Chemical bonding and antibonding under pressure We delve deeper into why pressure causes phase transitions in materials by considering the electronic structure. we can use our knowledge of phonon entropy to predict dP/dT if we rearrange Eq. so pressure ranges of 5 GPa are typically required to observe similar effects. Approximately.38) (which is a Maxwell relationship that we could have invoked immediately). V (8. Ultimately. at least when the pressure is sufficiently high. Hence the magnitude of the value listed in Table 8.37) (8. Such extreme conditions for materials are not possible with temperature changes alone because they exceed boiling temperatures.39) j As mentioned previously. 8. giving     dP ∂S 5 MPa = ∼ .

so the tails of the wavefunctions |a. becomes larger as the atoms are pushed together.

and |0.

overlap more strongly in the region of the perturbation H. The energy split between the bonding and 17:28:44 .

meaning that a negative lobe of one wavefunction overlaps a positive lobe of its neighbor. Notice that the light gray overlaps other light gray especially strongly for the two cases at left (the σ -bonds of the eg orbitals). The energies of the different 2 Although the present treatment uses only a bonding level and an antibonding level. and the latter is a π -bond. which causes a more rapid increase in the energy difference between the bonding and antibonding states.4. the orbitals shown in the two cases at left are called eg . For reasons of symmetry. Notice that the light gray overlaps the neighboring dark gray. the wavefunction x2 − y2 has its maxima and lobes along the x-axis and y-axis. We now take the viewpoint of a solid-state chemist by considering the shapes of bonding and antibonding molecular orbitals in a crystalline solid. The three cases at right have lobes along diagonal directions. 17:28:44 .4 Chemical bonding and antibonding under pressure 203 antibonding levels therefore grows larger with pressure. The former is a σ -bond.8. the eg bonding is more sensitive to the decrease in volume. where molecular identity is lost as the atoms of separate molecules approach each other closely. whose lobes are less direct in their overlap. so the overlap is not so strong. zˆ that are used to define the orientation dependence shown at the bottom of the figure. If we compress the cubes in Fig. which point directly at each other. The phase factor for the wavefunctions of electrons at neighboring atoms (which gave the factor cos(k · a) in Eq. A common consequence is that insulators become metals at high pressures. This even occurs with some molecular solids. • Rotating about the axis of the bond.4 are • The cube axes lie along the directions xˆ .2 this viewpoint can be extended to a periodic solid by multiplying the atomic orbitals at each atom position {r} by phase factors  eik·r .) What often proves interesting is how the individual electron energy levels are altered differently by pressure. whereas the t2g states at right have one nodal plane that contains the bond axis. Some items to note in Fig. the two eg cases at left will overlap much more strongly than will the three t2g at right. The effect of pressure on the bonding can now be understood intuitively. For example. which have five characteristic shapes in an isolated atom as shown in Fig.4 (with labels at the bottom). • The two cases at left show strong overlaps of the lobes of the wavefunctions. • The antibonding cases are shown at the top. as in Sect. too. yˆ . Does water become metallic at a sufficiently high pressure? (This may require some thermal excitations of electrons. consider d-electrons. The energy change with volume is smaller for the t2g states at right. 8. This causes a node in the wavefunction perpendicular to the bond axis. 8. This boosts the electron density between the atoms. 8. 6.4 shows how the lobes between atomic d-wavefunction overlap each other to form bonding or antibonding orbitals. the eg states at left have a continuous density of overlap. • The bonding cases are shown at the bottom of Fig.4. whereas the wavefunction xz has lobes along the diagonal of the x–z-plane. a characteristic of antibonding states. 8.2 For an example. and t2g orbitals are in the three cases at right. 6.45) causes the entire band of energy states to widen in energy with increasing pressure. For a simple cubic lattice. Consider the example of a simple cubic lattice where atomic orbitals are placed at neighboring sites. between the atom centers. Figure 8.

as expected when the wavefunction is multiplied by a factor of −1. the widths of the different energy bands and their center shifts will change with pressure. (b) More general case.) This can lead to crossings of levels as shown in Fig. 8.. As volume decreases.5a. showing their characteristic symmetries and geometries when centered on a simple cubic lattice. usually differently for the different energy bands.Pressure 204 z y x t Fig. spanning from the bonding curve to the antibonding curve. When we consider the spread of electron levels into bands.5b will be altered. 8. since bonds have an orientational dependence that favors some near-neighbor atom configurations over others. with unequal energies for atomic levels. e. antibonding on the top. Notice changes of gray for antibonding. are involved. 8. 8. (This is more obvious when different electron states. This energy spread increases with pressure. 8. the different levels need not start at the same energies at large separations. there is no overlap of the lobes of the wavefunctions on different sites. 8. A level crossing may induce a change of crystal structure. 8. of course.5b. The details of the level crossings in Fig. 3p and 3d.5 to have some vertical spread in energy.r 2 x2 .5 a Distance (lattice parameter) Antibonding eg Bonding eg b Distance (lattice parameter) (a) Expected energies of bonding and antibonding levels for the d-electrons of Fig. the bonding levels split away from the antibonding. and detailed 17:28:44 . When the atoms are far apart (large lattice parameter). Antibonding t2g Antibonding eg Energy Energy Antibonding t2g Bonding t2g Bonding t2g Bonding eg t Fig. our picture is in need of several refinements.g. levels are shown in Fig.4. First. Perhaps the simplest way to extend the argument of this section is to consider the sharp lines in Fig. Although we have justified why we might expect different electron levels in a solid to respond differently to pressure. Bonding levels are on the bottom.4 3z 2 .y2 xz yz xy The five atomic d-electrons. and the effect is stronger for the eg than the t2g .

with neighbors on the face centers.44) E ≡ 0 E10  1 δ≡ ϒ1 E10 − ϒ2 E20 B (8. Changing the lattice parameter and recalculating the electron energy levels is not a particularly difficult thing to do. Without pressure. which changes the energies of the two electronic levels. It includes • Temperature. Assume the two electron energy levels change linearly with pressure. which is something like a Grüneisen parameter. but it is for electronic energies. the diagonal lobes of the t2g orbitals make σ -bonds. or with other orbitals (such as p-electrons) can be understood similarly.5 Two-level system under pressure The analysis of the previous Sect.41 and 8. and these bonds are more sensitive to pressure than are the eg orbitals on the fcc crystal. not phonon energies. Here the degree of freedom is the choice of electronic level. 17:28:44 (8.5 Two-level system under pressure calculations of electron energies are needed to assess the effects. once the DFT machinery is up and running. Fortunately.43) − E20 (8. 8. the “crystal field” approach of the present section often proves valuable for understanding changes in bonding under pressure.4 had no role for temperature. In the fcc structure.42 is: E = E0 + Pδ . The example worked out with Fig.205 8. The different problem of d-orbitals on a facecentered cubic (fcc) lattice can be worked out the same way. with thermal energy kB T per degree of freedom of the solid.45) the energy difference of Eqs. It is perhaps handy to imagine each atom as having two levels.41) E1 = E1 1 − ϒ1 V B     V P = E20 1 + ϒ2 .42) E2 = E20 1 − ϒ2 V B Making a few definitions to simplify appearances E ≡ E1 − E2 (8. Nevertheless. E1 and E2 . 8. 8. although with different rates     V P 0 0 = E1 1 + ϒ1 .46) . these energies are E10 and E20 . 8. Examples with other crystal structures. • Pressure. (8. (8. Here we consider a simple model for a material under pressure and temperature that can be analyzed thoroughly. • Energies of two electron levels. this can be done fairly directly with modern density functional theory (DFT) codes.4 placed d-orbitals on a simple cubic lattice. These effects are parameterized by ϒ.

0 1. the transition would be abrupt. Now consider a property of this two-level system.47) Finally. What is important for thermodynamics are the Boltzmann factors for the fractions.Pressure 206 1. the fractions of the two levels are normalized f 1 + f2 = 1 . a larger difference in ϒ1 and ϒ2 would be needed to get a transition at the same pressure. the two levels may cause an atom to have two different specific volumes. 8. (8. (For example. are shown in Fig.6 ∆E = 10 kT δE = 5 kT/P 0. 17:28:44 (8. For the same differences in energy of the two levels at P = 0 (same E10 −E20 ). a higher temperature broadens the pressure range of the transition.5 3. too.0 0.6.6 0.5 1. such as an equation of state V(P). and the low-pressure level became energetically unfavorable with increasing pressure.4 ∆E = 5 kT δE = 10 kT/P 0. and f1 (P) would be a step function falling from 1 to 0 at the single pressure where the levels cross. The difference in energy levels is important.49) . taken as the average of properties of the two levels. f1 .0 ∆E = 10 kT Fraction f1 0. In addition.2 ∆E = 5 kT δE = 5 kT/P t Fig.5 Pressure 2.0 Fraction of the low-pressure level for the two-level model at different temperatures and pressures. f .) V(P) = f1 V1 (P) + f2 V2 (P) . For these examples kB T was assumed relatively small compared with the other energies (although not overwhelmingly so or the examples would be near T = 0). The parameters were set so ϒ1 > ϒ2 . of the two levels e−E1 /kT f1 = −E /kT = e−E/kT .48) Some calculations of the fraction of the low-pressure level.8 δE = 10 kT/P 0. 8. the range of pressure of the transition is narrower under these conditions of large E10 − E20 and large ϒ1 − ϒ2 (compare the two middle curves). If there were no effect of temperature.0 2.0 0. Parameters E and δ are as labeled on the curves. With a larger energy level difference. f2 e 2 (8.

from [82]. Reproduced. The fractions of the two levels vary with pressure according to Eqs.7 0 5 10 15 20 25 Pressure [GPa] 30 35 Volume–pressure data obtained from synchrotron X-ray diffraction measurements on Pd3 Fe under pressure (symbols).47 and 8. and after a little manipulation V(P) = .88 0.86 t Fig.48. with permission.94 0. 8. 8. 8.8.50 with different V1 (P) and V2 (P) (line).96 0.92 0.9 0.5 Two-level system under pressure 207 1 0.98 V/V0 0. and the fit of the data to the Weiss-like equation of state of Eq. Copyright (2009) by the American Physical Society.

we consider how the effect of pressure on entropy can alter the fractions of the two levels. 8. and make a smooth transition to V2 (P) as predicted by the fractional change shown in Fig. Through the magnetic transition. If each level has a fixed entropy the relative fraction of the two levels is the result of the energy difference alone.7.51) (8.50 can be used to model Invar behavior that is driven by either temperature or pressure. f2 (8. Eq. Now suppose the entropy of a level varies with volume as given by a Grüneisen parameter and suppose the Grüneisen parameters are different for the two levels. f2 e 2 f1 = e−E/kT eS/k . 8.53) . It can be shown that Eq.52 becomes f1 = e−E/kT eS/k e3(γ1 −γ2 )P/B .50) This two-level model can be used for the Weiss theory of Invar. f2 17:28:44 (8. One way to interpret this result is to graph two equations of state V1 (P) and V2 (P). 8. Start on the low-pressure curve V1 (P). In this model.47 if S is constant. the loss of magnetic moment with temperature causes the magnetic atoms to have a smaller volume.52) which is essentially the same as Eq. V 1 2 1 + eE/kT (8. 8. Finally. 1 E/kT (P) + e V (P) . a material with zero thermal expansion. Although the Weiss theory is phenomenological. modified by a simple correction e−F1 /kT f1 = −F /kT . this contraction counteracts the normal thermal expansion.

and now includes a pressure term   E + PV . 8. The probability ratio is set by a ratio of Gibbs factors. It is typical to interpret experimental data on activation energies with Eq. The vibrational modes for an atom in this level do not stiffen so strongly under pressure. (8.208 Pressure where the factor of 3 is needed because three modes of vibration are available to each atom. the jump rate reflects the Boltzmann factors as in Eq. T) = 0 exp − kB T where V is an “activation volume. The V is comparable to an atomic volume. The basic idea is that there is an equilibrium between the intitial state and the “activated state. for example. 8. If the system is in a steady state. • An activation volume is the change in volume of the system that must be produced by a thermal fluctuation if the jump process is to occur.6 Activation volume In activated state rate theory as in Sect.5.” An activation volume V is perhaps best understood by analogy to an activation energy. Pressure will favor the level with the smaller Grüneisen parameter.56) (P. (8. Section 3. (8. Performing measurements of jump frequencies under pressure is a way to obtain information about how atoms are rearranged during the transient state of a jump. 3. so the probabilities of these states are in the ratio of their Boltzmann factors.54) where E is an activation energy. kB T 17:28:44 (8. so there is the same rate of transition from initial to activated state as from activated state to initial state.” or transition state. and the transition rate is (P.55) where G is an activation free energy that includes the E.. so its vibrational entropy remains larger than the other. the jump frequency depends on temperature as (T) = 0 e−E/kB T . 3. • The activation energy is the change in energy of the system that must be produced by a thermal fluctuation if the jump process is to occur. An activation volume gives insight into how atoms are rearranged during a kinetic process.5 explained the origin of this expression (cf.54. This condition of equilibrium between the initial and activated states can be generalized for equilibrium in a system under a constant pressure and temperature.98). Eq.54 rearranged as ln ( (T)) = ln ( 0 ) − E . 8. For atomic diffusion. activation volumes are positive because surrounding atoms must open a passage for the diffusing atom. T) = 0 e−G/kB T .57) .

2 × 10−10 m2 /N. It is often true that the effective γ is not known.2 K. starting at atmospheric pressure.3). is compressed adiabatically in a container with good thermal insulation. fcc γ -cerium undergoes a volume collapse to fcc α-cerium at room temperature (see Fig. This is expected if the prefactor is proportional to a phonon frequency. and −V −1 (∂V/∂P)T = 1. giving the activation energy. What is the density of the solid near the interface? 8. 8.35.6 × 10−4 /K. the material below a height l in the column is solid. The specific volume of ice is greater than that of liquid water by 9×10−5 m3 /kg. vertical column of a material is kept at constant temperature at all heights. (8.58) For a fixed T. 17:28:44 . is heated in a container that maintains constant volume. At what pressure will the ice melt? 8. Measurements at multiple temperatures and pressures can help sort out the different contributions. the solid–liquid interface shifts upwards by 0.1 The melting temperature of ice depends on pressure. and κ is the isothermal compressibility of Eq. In this case. 8. A similar rearrangement of Eq. measurements at one temperature cannot determine separately the two terms in the parentheses () in Eq. for a fixed T. At what pressure will the ice melt? (b) Ice at −2 ◦ C. starting at atmospheric pressure. When the temperature is T0 . At what temperature will the ice melt? (c) Ice at −2 ◦ C. In cases where V is small. (a) Ice is compressed isothermally at −2 ◦ C. 1. ln ( (P)) = ln ( 0 ) − PV E − . T) = −γ κ − (P. The heat of fusion is 10 kJ/mole.000 kg m−3 .22.209 Problems A graph of ln ( (T)) versus 1/kB T gives a slope of −E. Problems 8. One explanation of this effect includes an electronic transition. and the density of the liquid is 1. giving the activation volume.56 gives. but the material above l is liquid.2 A tall. When the temperature is increased to T0 + 0. and the phonon frequencies increase under pressure.3 Under pressure. This graph gives a slope of −V. The heat of fusion of ice at atmospheric pressure is 330 kJ/kg. the second term on the right-hand side is a constant when ln ( (P)) is plotted against P/kB T.5 m. kB T kB T (8. T) .05 kJ/(kg K).59.59) ∂P kB T T where γ is the Grüneisen parameter of Eq. Thermal expansion and compressibility constants for ice are V −1 (∂V/∂T)P = 1. starting at atmospheric pressure. The heat capacity of ice is 2. 8. an improved analysis should consider how the prefactor depends on pressure. 8. the gas may be a nearly free electron band composed of 6s electrons. An analysis gives     V ∂ (P. where one 4f -electron at each cerium atom transfers to a free electron gas (in more detail.

60) ∂S N.61) (8. N) to obtain dE.210 Pressure but please assume a simple electron gas for this problem). alters thermal expansion. (8. 8. ϒ. (b) At an intermediate volume.5 (a) Explain why the parameter for the shift of electron energy with volume.N 17:28:44 (8. and using the definitions       ∂E ∂E ∂E =T . ∂V S.30 to calculate the increment to the linear coefficient of thermal expansion. = −P . Why would there be a two-phase mixture of α.and γ -cerium instead of just one of them of intermediate lattice parameter? 8. there are two coexisting fcc phases of cerium with distinct lattice parameters. (b) bcc structure (hybridize the three t2g and two eg ). use the approach of Eqs.N make use of the equality of the mixed partial derivatives ∂ 2E ∂ 2E = .6 Differentiate the fundamental relationship E(S. =μ .4 Following the approach of Sect.4.V ∂N S.V ∂V S.N ∂S V.62) . α = ϒ/(3BVT). How does the effect depend on the sign of the applied pressure? (b) For a compressed solid. Suppose that the bonding between the 4f -electrons has an interatomic potential of a Lennard–Jones potential. 8.28–24. 24. V. ∂V ∂S ∂S ∂V to derive the Maxwell relationship     ∂P ∂T =− . You may assume that all levels have equal energy at large volumes. graph how the bonding and antibonding orbitals for the five types of d electrons change with volume in the (a) fcc structure. (a) Explain why this electronic transition might occur with decreasing volume. 8.

and the second is a venerable statistical mechanics model of diffusion that has components used today in many computer simulations of diffusion. There are other risks in using the diffusion equation when atom motions occur by the vacancy mechanism. (9. When materials with different diffusivities are brought into contact. It shows where the atom is located. Chapter 9 ends with two other topics of diffusion – one is atom diffusion that occurs in parallel with atom jumps forced without thermal activation. their interface is displaced with time because the fluxes of atoms across the interface are not equal in both directions.1 used the kinetic master equation approach to obtain the diffusion equation. and this is rarely true as an alloy evolves during a phase transformation. Solutions to the diffusion equation were presented.1.9 Atom movements with the vacancy mechanism Chapter 3 derived the diffusion equation with the assumption of random atom jumps. Figure 9. Other phenomena such as stresses and voids may develop during interdiffusion. but the reader was warned that these solutions require a constant diffusion constant D.  6 . 9.1) i=1 17:29:48 . but used averaged jumps – it did not consider fluctuations in the flow of atoms between bins. where a mobile vacancy rearranges atoms in its wake. after n jumps Six jump vectors are added to obtain this distance. after six jumps.1 Correlation factor Section 3.1 is a picture of the process in two dimensions. and such a bias can also be created by chemical interactions between atoms. so any assumption of random walk may be seriously in error. An applied field can bias the diffusion process towards a particular direction.1 Random walk and correlations 9. In general. the vacancy frequently resides in energetically favorable locations. This chapter explains the nonrandomness of atom jumps with a vacancy mechanism. Furthermore. The method considered random jumps of individual atoms. with respect to its start at the origin. R n = R n  ri . in an alloy with chemical interactions strong enough to cause a phase transformation. Here we focus on the motions of an individual diffusing species. and these nonrandom characteristics occur even when the vacancy itself moves by random walk.

9. . which is the number of (off-diagonal) elements across the square structure of Eq. (9. • The second sum (two of them) from one off the diagonal has n − 1 terms with indices that differ by 1 (note that ri · ri+j = ri+j · ri .4 are now written with the cosine of the angles.. + 2 i=1 1  ri · ri+n−1 . • The nth sum (two of them) from n − 1 off the diagonal has one term with indices that differ by n − 1 (these are the single elements in the upper right and lower left corners).” which is a scalar: R2n = We calculate the square of R   Rn · Rn . • The single sum along the diagonal has n terms with indices that differ by 0. Note the roles of the variable j in Eq. . (9. θ. r3 · rn + . j=1 i=1 17:29:48 (9. Writing out in detail the product of two forms of Eq. 9.3. and j ranges from 1 to n − 1. the “mean-squared displacement.5) . The dot products of vectors in Eq. .. Writing these sums explicitly R2n = n  i=1 ri · ri+0 + 2 n−1  ri · ri+1 + 2 n−2  i=1 ri · ri+2 + .. which identifies the difference in indices between two vectors ri and ri+j . • The third sum (two of them) from two off the diagonal has n − 2 terms with indices that differ by 2. r2 · rn + r3 · r1 + r3 · r2 + r3 · r3 + . between the vectors and their lengths R2n = n  i=1 ri2 + 2 n−j n−1   ri ri+j cos θi. (9. .3 into sums of terms along diagonals from upper left to lower right. 9. and all off-diagonal elements are in the double sum. .3) i=1 We keep the first sum separate from the others. . . r1 · rn + r2 · r1 + r2 · r2 + r2 · r3 + .i+j . but we sum the others by making a double sum R2n = n  ri2 + 2 i=1 n−j n−1   ri · ri+j .Atom movements with the vacancy mechanism 212  n . so these two sums are equal).2) + rn · r1 + rn · r2 + rn · r3 + .4) j=1 i=1 where the first sum contains the n diagonal elements of the square form of Eq. 9. rn · rn Writing the terms in this square form makes it easier to identify terms with equal differences in their indices.. .1 R2n = r1 · r1 + r1 · r2 + r1 · r3 + . . We regroup the big sum of Eq. 9. • . The variable j also accounts for the reduced number of terms in the sums that are further off the diagonal.3. 9.4. .

4 if the directions of successive jumps are completely uncorrelated.0 Six random jump vectors. either the time average or the ensemble average ri · ri+j .0..9. a subsequent jump has no bias in orientation with respect to the previous jump. 9.1 Random walk and correlations 213 6 5 2 3 4 R6 t Fig.1 1 0.e. 9. summed to give R6 . i.1 Random walk result The random walk result is easily obtained from Eq. In this case. Box 9.

9. The first sum is simply a sum of n identical terms R2n = n r2 + 2r2 n−j n−1   cos θi. An ensemble average would average the double sums in Eq.9. jump distance. but we assume their equivalence in what follows.1. There are subtle issues in how to perform the averaging. 3. (9. Proving the equivalence of the ensemble average and the time average is not easy in general.7) R2n = D τ . and diffusion coefficient in three dimensions. recognizing 6n/ as a total time for diffusion. Using the relationship between jump rate.42. 9. τ R2n = n (9. An alternative average is a time average of many jumps of a single species. A thermodynamic average is an ensemble average. D = a2 /6 (Eq. however. (9. is zero for all j > 0.i+j .9) j=1 i=1 We have been considering a single chain of jumps as depicted in Fig. as is usually expected when diffusion occurs on a crystal lattice. there is no special 17:29:48 . Our interest. Now assume that all jumps have the same length.16) 6D . is in macroscopic behavior. r. Finally. 3. (9. If the system is in steady state. which averages quantities from a large number of identical systems. We work with the time average of a system that is in steady state (so there is no phase transformation underway.8) consistent with Eq. for example). which averages many jumps of many atoms.6) assuming all n jumps have the same length ri = a. giving the result for a random walk R2n = n ri2 .

because all n − j terms in this sum are the same R2n = n r2 + 2r2 n−1  (n − j) cos θj . is defined as n−1 2  (n − j) cos θj . 9.10 is no longer needed (i is the number of the jump in the sequence). not absolute time. the jumping species is expected to lose memory of its initial jump.14 average to zero (which they will if the jumps are equally likely left.” f . and ignoring the off-diagonal elements that give the cos θj terms. (9.12 becomes the result for a random walk where the mean-squared displacement is R2n = nr2 . so cos θj → 0 for large j. that atom motions are usually correlated. This amounts to adding all elements along the long diagonal in Eq. The time-averaged angle between a pair of jumps cannot depend on whether the first jump of the pair is the very first jump at t = 0 or the nth jump at a later time.10) j=1 i=1 where the overlines indicate time averages.14 can be generated by long division of 1 + cos θ1 by 1 − cos θ1 f = 1 + cos θ1 1 − cos θ1 . (9. however. up. The sum over i in Eq. 9.15) Equation 9. (9. 9. so they usually depend on the direction of previous jumps.12 has a very simple form. It turns out. down. What is important in averaging the jump angles is the time between jumps. so f =1+2 ∞  cos θj .3. and sometimes f  1. The system loses memory of the direction of a previous jump. out). In the case of a truly random walk.11) j=1 We rewrite Eq.i+j with cos θj and average it R2n = n r2 + 2r2 n−j n−1   cos θj . in. which is expected because each subsequent jump has the average correlation cos θ1 along the direction of the previous jump.11 as R2n = fnr2 . 17:29:48 . As n → ∞. In these cases f < 1. n→∞ n f = 1 + lim (9. j  n. 9.15 gives f = 1 and Eq. 9. (9. 9. Finally. Eq. 9.12) where the “correlation factor.Atom movements with the vacancy mechanism 214 meaning of t = 0. (9. right. the series of Eq. We therefore replace cos θi. but the time for losing memory is the same for jumps that happen earlier or later in the jump sequence. where the cos θj terms in Eq.13) j=1 After a large number of jumps.14) j=1 We next make use of the relationship cos θj = (cos θ1 )j . however. so R2n is smaller than expected from random jumps.

Radioactive atoms were the first robust example of tracers. 9. 6m (9. 216 cos θ1 = (9. so the vacancy must have jumped to the right. we watch the vacancy because the vacancy jump sequence sets the probabilities.2 Correlation factor for a tracer atom on a hexagonal lattice From Eq. Assume the vacancy executes a random walk on the lattice. and composition. For the tracer atom to move.2 –1 First jump of tracer atom (gray. 17:29:48 . lattice. 9.2 shows all possible ways that the tracer can move into a neighboring site within three jumps of the vacancy. and the tracer to the left.16) 1 2 −1 2 1 2 (−1) + 2 + 3 + 3 (−1) . times the probability of the vacancy jump sequence (1/6)m cos θ1 = m max  n=m Nsq cos θsq . 9. Here we work the example of the motions of a tracer atom1 shown as a gray circle in Fig. 9. What is the average cosine of the angle between the next jump of the tracer with respect to its original jump to the left? In working this calculation for the tracer. 9.18) 1 This means that an atom is “tagged” in some way to follow its motion amid numerous similar atoms. Figure 9.1. Light arrows are vacancy jumps. heavy arrows are for the tracer. The probability is (1/6)m for any specific sequence of m jumps of the vacancy.2).15 we are left with the task of calculating cos θ1 .2. so the previous tracer jump was to the left. so its probability of exchange with any one of its six neighbors is 1/6. It is assumed that the vacancy and tracer have just interchanged positions.9. 6 6 2 6 2 6 47 cos θ1 = − . The average cos θ1 is calculated as the sum of the values of cos θsq for the sequences (in the right column of Fig. it must jump into one of the six sites of its first neighbor shell. Assume there has just been an exchange of sites between the vacancy and the tracer. times the number of sequences Nsq . at center) after specific sequences of jumps of a vacancy (square in initial position).1 Random walk and correlations 215 Nsq cosθsq Jump sequence m 1 1 2 2 – 1 2 2 + 1 2 –1 3 3 t Fig. This depends on the diffusion mechanism.17) (9.

and its correlation factor can be much less than 1.15.3 Correlation factor for a vacancy in a heterogeneous alloy We just showed how atomic diffusion with a vacancy mechanism is correlated. This practice will avoid the error of including sequences such as shown in Fig.1 Correlation factors for tracer atoms with vacancy diffusion Lattice t Fig. Consider the black atom in Fig. which is not to be included in the average of cos θ1 .64.Atom movements with the vacancy mechanism 216 Table 9. Notice that lattices with higher coordination numbers z tend to have larger values of f . Sometimes this approximation is used: f 1 − 2/z. Using Eq. even though the vacancy executes a random walk. cos θ2 . which contributes to the average of cos θ2 . 9.. and with a much higher probability of exchanging sites with the vacancy. but not dramatically so. This differs from the more accurate result of f = 0.3. The third jump of the vacancy involves the second jump of the tracer. In an alloy the vacancy motion becomes correlated too.78145 0. which shows two arrows for the tracer.56006 Three jumps of a vacancy (light arrows). 17:29:48 . the correlation factor is evaluated as f = 0. These are all less than 1. 9. different chemical species will have different atom–vacancy interchange frequencies. Table 9. it is useful to draw arrows for both the vacancy and the tracer atom jumps. with the first involving the jump of the tracer (heavy arrow).1 presents correlation factors for tracer atoms on a number of lattices. i.6531 0.56 because we stopped our analysis at only three jumps of the vacancy.4 as a different species from the white.1. assuming the vacancy executes a random walk.72719 0. 9.78146 parallel to c 0.2 2 In practice. This means that the sequence pertains to the second jump of the tracer. 9. The frequencies of exchange are ω1 and ω2 .e.3 f fcc hcp hcp bcc sc dc 0. When enumerating sequences of vacancy jumps. 46694 0. as in Fig. 9. not cos θ1 . and ω2  ω1 . 9. and for thermally activated jumps these frequencies become increasing different at lower temperatures.2.78121 normal to c 0.5 0.5 Honeycomb Square Hexagonal 0.

4. with a different interchange frequency ω2 for the solute atom (black) than the majority species (white). becomes small as ω2  ω1 .19) (9.25) which is substituted into Eq. This is not unusual for diffusion problems in an alloy. the solute at left is most likely to jump with probability p2 . and all four jumps should be considered for calculating the average cos θ1 of the vacancy. 9. Notice that f is proportional to ω1 . but the other three jumps have the probability p1 ω2 . ω2 ω2 ω1 cos θ1 −1 + 4 .4 ω ω1 1 Vacancy interchanges with first neighbors on a square lattice. −ω2 + ω1 cos θ1 = . Usually both species must be able to move for macroscopic diffusion to 17:29:48 .21) (9.26) (9. ω2 + 3ω1   −ω2 + ω1 3ω1 cos θ1 1− . 9.15 for f to obtain f = 1 − 1 + 4 ωω12 1 + 1 − 4 ωω12 ω1 . the frequency of vacancy exchanges with the slower-moving majority species. There is a competition between the four neighbors for jumping into the vacant site in Fig.22) (9. Since ω2  ω1 .24) (9.27) The correlation factor for the vacancy. ω2 + 3ω1 ω1 p1 = . ω2 (9. 9. ω2 + 3ω1 p2 = (9.20) The average of cos θ1 is the sum of these probabilities times the cosines of the angles between the original vacancy jump (to the right) and the next jump of the vacancy cos θ1 = p2 (−1) + 2p1 (0) + p1 (+1) . ω2 ω2   3ω12 1 cos θ1 ω1 − ω2 + 3ω1 − .217 9. and its mean-squared displacement after n jumps. f 2 ω2 .1 Random walk and correlations ω1 ω2 t Fig. (9.23) (9.

12 gives R2n = 0.40) 4V ≡ eAA + eBB − 2eAB .5 shows an antiphase domain boundary in the middle. (9.28) which averages the chemical preferences of the A-atoms and B-atoms. vacancies are expected to become trapped at antiphase domain boundaries. 9. but three of them are bonded to two or three B-atoms.6 as an overall effect on the kinetics of a disorder→order transformation. Unfavorable pairs are indicated with lines. 9.5 shows an A-atom that is likely to exchange sites with a vacancy in an ordered alloy. Eq. If one species remains unmoved on its lattice sites. Eq. rapid motions of the other species around it cannot lead to chemical rearrangements [83. occur. the vacancy is “trapped. Alternate jumps of the vacancy are in opposite directions. running from top to bottom. The thermodynamics of ordering depends only on the parameter 4V (c.5 B A A B A A A B A B a B A B B B B b B B A B A A A B A A B B B A B B (a).and B-atoms switch lattice sites. A closer examination of the local atomic configuration in Fig. In the limit where ω1 /ω2 = 0. 9. At low temperatures. 9.” and will exchange sites numerous times with this A-atom. and similarly for the solute atom. 7. and are less likely to move than the central A-atom (it is bonded to three A-atoms and zero B-atoms). so cos θ1 = −1 and f = 0. the vacancy and solute atom exchange are the only events that occur. 2. After numerous jumps of the solute atom. without considering the cause of these differences. Kinetic processes depend on the individual atom preferences. or in regions with local disorder. The temperature dependence of trapping is shown in Fig. All neighbors of the vacancy are Aatoms. Figure 9. in which unlike pairs are preferred.Atom movements with the vacancy mechanism 218 B B t Fig. and we can assess these effects on the temperature scale of the critical temperature.29) . Especially at low temperatures. (b) Frequent vacancy interchanges are expected with an A-atom that is unfavorably bonded to its neighbors in an alloy that favors A–B pairs. and neither the solute atom nor the vacancy has moved a macroscopic distance. An antiphase domain boundary (see Fig.4 Correlation factors for ordered alloys The previous section on correlated vacancy motion used a simple parameterization of jump frequencies. 17:29:48 (9.1. Correlations of vacancy motion can be caused by the same interatomic interactions that are responsible for phase transitions.. however.2) separates two ordered regions where the A. 84].f. 9. for which the parameter U is defined [85] U ≡ (eAA − eBB )/4V .

Assuming A-atoms prefer the α-sublattice (and B the β). 9.25) vs. 2. and a lower correlation factor f for the A-atom (although this needs to be considered in parallel with those A-atoms that start on the β-sublattice).4 200 Interchange o 1 2 3 4 5 4V in units of kT Number of vacancy jumps per atom to change a two-dimensional square lattice from a disordered solid solution to a partially ordered alloy (with a Warren–Cowley short-range order parameter of 0. negative contribution to cos θ1 in Eq. as a fraction of the critical temperature. Consider the B2 ordered structure or the chessboard structure of Fig. This reversal becomes even more likely at low temperatures where chemical preferences are larger with respect to kB T.9. there is no vacancy trapping for the physically unrealistic interchange mechanism. An A-atom that started on the α-sublattice and moved to the β is more likely to reverse its interchange with the vacancy than an A-atom that started on the unfavorable β-sublattice. owing to the exponentially decreasing probability of exiting the trap. At lower temperatures vacancies tend to spend most of their time in the strongest traps. H. For comparison. 9. 9. and the higher degree of order gives fewer unfavorably placed atoms to exchange sites with the vacancy. This is approximately the same scaling as the critical temperature for ordering. Even neglecting the trapping of vacancies in ordered structures.6 U = 0.5) because for low z a strong trap can be constructed with fewer atoms. and the chemical preferences are dominated by one atom over the other. Bakker adapted the random walk problem to account for different 17:29:48 . Fig.8 U=0 100 20 0 t Fig.1 Random walk and correlations 219 300 Jumps per atom U = 0.15. and the efficiency of ordering by atom interchange does not depend strongly on temperature. the correlation factor f becomes small in ordered alloys at low temperatures.16 where each first-nearest neighbor (1nn) is on the other sublattice.g.6 shows that the effects of vacancy trapping occur at much higher temperatures when |U| > 0. 4V [86]. lattices of low coordination number are more susceptible to vacancy trapping at antiphase domain boundaries (e. the probability to reverse an atom–vacancy exchange depends on the starting sublattice of the A-atom. Figure 9. The strengths of traps for vacancies at antiphase domain boundaries increase as (z − 1)(eAA − eAB ) because there are more unfavorable bonds to a trapped atom in lattices with larger coordination number z. Nevertheless. where adjacent pairs of atoms interchange positions without the use of a vacancy. An increased probability for reversal leads to a large.

Heterogeneities like the one of Fig. Some issues for diffusion in a chemically inhomogeneous material are • Mobility differences between different chemical species. and independent diffusion equations can be used for both chemical species.2. using some of the methods of Chapter 22 and a pair approximation for chemical interactions. The first two issues are illustrated with Fig. interatomic interactions. light gas (top) and slow. and appropriate solutions to both diffusion equations are the error functions of Sect. 17:29:48 .16). in principle. showing increased and decreased total atom density near interface. If we sum the concentrations from the two error function solutions. 3. (a) Initial condition at t = 0 – a sharp interface between the two gases.7b. and consider the effects of inhomogeneous composition on the diffusion coefficient. calculate a diffusion coefficient that works well for a homogeneous material in steady state. where f is the correlation factor. light gas and the heavy. Note the peculiar density variation near the original interface. 9. (c) Sum of concentration profiles.2 Phenomena in alloy diffusion The previous section showed that the prediction of a diffusion coefficient from atomic jump frequencies. by the substitution of → f . we find the total concentration of Fig. The initial conditions are sharp concentration cutoffs at x = 0 for both gases. 9.7. we can. (b) Independent diffusion of fast. Fast Slow a 0 x b 0 x c 0 x Sum t Fig.7c.2. 3. which shows two ideal gases that mix by diffusion. The fast. 9. slow gas are brought into contact at t = 0.01 or less.5 are a concern for alloys. and the geometry of the lattice is more complicated than expected from the simple expression D = a2 /6 (Eq. heavy gas (bottom).Atom movements with the vacancy mechanism 220 sublattices of an ordered alloy [87].7 Composition profiles for diffusional intermixing of two ideal gases. An ideal gas has no chemical interactions between point-like atoms as they move through each other. however. • Density conservation (or not). • Chemical interactions. but Bakker found that at half the critical temperature for ordering it is reasonable to expect correlation factors of 0. These two error function solutions are illustrated in Fig. 9. Now we go beyond the issue of a homogeneous alloy. Nevertheless. 9. 9. This analysis is beyond the scope of this book.

the increased atomic density is not possible. (a) Original composition profile from Fig.2. In this case.9. displacing an amount of volume needed to account for the missing atoms (minus any voids that have formed).8c. A brass rod was wrapped with 17:29:48 .8b and 9. 9.” Another possibility is the annihilation of crystal sites as in Fig. but the overall effect is to move the bar to the right by an amount that accommodates the extra atoms.2 Phenomena in alloy diffusion 221 a 0 x 0 x 0 x b t Fig. Suitable markers should be materials with negligible diffusivity at the temperature of interest. 9.8c. there will be additional crystal sites created near the original interface. especially since the sample is typically quite hot. the extra atoms that have moved to the right must cause the bar to grow to the right. 9.8b. although to a lesser extent.8 c Resolution of atom concentration problems for a bar of crystalline material. where the entire bar of material crawls a bit to the right.8c.8b. the markers will move to the left. but we can design an experiment to keep track of the original interface. especially on a crystal lattice that has a fixed number of sites per unit volume. deficit of matter causes Kirkendall voids on the left. and no reactivity with either of the chemical species. (c) Excess matter forms more crystal on the right. Measuring the “marker velocity” is an important way to understand differences in the diffusivity of two chemical species. with its very high melting temperature. as shown in Figs. or small holes to the left of the original interface. For the case of Fig. These are called “Kirkendall voids. It is rarely practical to measure such tiny motions in the laboratory reference frame. (Tungsten. 9. owing to the greater mass transport to the right. The classic marker velocity experiment was performed by Smigelskas and Kirkendall with a diffusion couple as shown in Fig. and these coalesce into voids.9 [88]. (b) Excess matter forms more crystal on the right. The atom deficit on the left of the original interface can be accommodated two ways. 9.8c. 9.7c will not do.) For the case of Fig. is often a suitable marker for metallic systems. the left end of the bar moves to the right. however. deficit of matter causes a loss of crystal on the left. For diffusion of two chemical species in a solid. the profile of Fig. 9. and both occur in nature. Most likely. vacant crystal sites are formed on the crystal lattice.7c. The idea is to put inert “markers” at the original interface between the two different materials at x = 0 and t = 0. This is also true for Fig. 9. 9. 9.1 Marker velocity Suppose the difference in atom mobility is accommodated as shown in Fig. Instead. 9. To the right of the interface.

note that the marker velocity is to the right if ∂cB /∂x 17:29:48 . ∂cA ∂cB =− .32) (9. ∂x ∂x (9.222 Atom movements with the vacancy mechanism Cu (electroplated) Mo (wires) Cu–Zn t Fig. there will be a marker velocity. which is a velocity.b ∂cA .30 gives ∂cB A DA ∂c ∂x + DB ∂x .36) (9. and electroplated with a thick layer of copper around it. 9. c ∂cB (DB − DA ) v = ∂x . for the two chemical species in Fig. c v= (9. and after the diffusion couple was heated it was found that d decreased. The marker velocity is v=− JA + JB . the greater the volume added to the left. there is no marker velocity. 9. As a check. molybdenum wire. and the higher the marker velocity to the right. ∂x JA = −DA (9.30) where c is the total number of atoms/cm3 . as expected.31) (9.9 d Cross-section cut through a diffusion couple with Mo wire markers used for the Kirkendall experiment.7a. cA = c − cB . 9. The markers define a plane through which there is a diffusive flux. c (9.33) When the fluxes JA and JB in Eq. The interpretation is that Zn diffuses faster than Cu.37) If there is a difference in the diffusivity of two species in a diffusion couple.34) (9. 9. The ratio J/c has units of [atoms/(s cm2 )] [cm3 /atoms] = [cm/s]. Fick’s first law gives. The distance d between the Mo wires was measured. The greater the flux to the left. These relations between composition variables prove useful below c = cA + cB . so more atoms moved outside the markers than moved in.30 are equal and opposite. ∂x ∂cB JB = −DB . Any extra atoms moving to the left will add volume in proportion to 1/c.35) which upon substitution into Eq. The marker plane has a positive velocity (along xˆ to the right) if there is a greater flux of atoms moving through it to the left than to the right – the marker velocity is in the direction opposite to that of the dominant flux. which has units of cm3 /atom.

the flux of B-atoms must include the flux v cB of B-atoms that do not make diffusive jumps. DA and DB .2. The first uses the crystal lattice as a frame of reference.44) D c We now have two equivalent formulations of interdiffusion for a diffusion couple. coefficient.3 The interdiffusion coefficient is a composition-weighted average of DA and DB . 17:29:48 .” In this case we solve one diffusion equation instead of two.36 for v   ∂cB A DA ∂c ∂cB ∂x + DB ∂x + cB .40) ∂cB cA + cB ∂cB cB ∂cB cB − DA + DB . 9. (9.2 Interdiffusion coefficient For a diffusion couple of elements A and B.2 Phenomena in alloy diffusion 223 is positive and the B-atoms are the faster-moving species.9. ∂x c ∂x c ∂x c From Eq. and v is the velocity of the coordinate system in [cm/s]. 9. The total flux of B-atoms across the moving marker (here +v is to the right along +x) is JB = −DB ∂cB + vcB . but we need to know the marker velocity.35). 9. 5. the compositional weights are for the opposite species. 9. (9.42) c ∂x ∂c ˜ B .4. we can describe the interdiffusion dynamics with two diffusion coefficients. which move to the left by Eq. with two diffusion coefficients {DA .33. The net flux is dominated by B-atoms. The second uses the ends of the JB = −DB 3 This is much like the moving solid–liquid boundary considered in Sect. JB = −DB ∂x c ∂cB cA + cB ∂cA cB ∂cB cB + DA + DB . the marker velocity v and D.35.38) (9. Suppose the marker moves with velocity −v with respect to the far ends of the diffusion couple.41) ∂x c ∂x c ∂x c and with the cancellation of the two terms involving DB cB   DB cA + DA cB ∂cB JB = − . If the solute concentration is in [atoms/cm3 ].39) (9. In the moving coordinate system. It turns out that we can describe the problem ˜ an “interdiffusion with another equivalent pair of variables.34 and 9.43) J B = −D ∂x where the definition of the “interdiffusion coefficient” is D c + DA cB ˜ ≡ B A (9. two diffusion equations (Eqs. DB }. as appropriate for a flux. ∂cA /∂x = −∂cB /∂x JB = −DB (9. 9. ∂x Substituting the expression Eq. (9. It sweeps past B-atoms that are stationary. the product v c has units of [atoms/(s cm2 )]. but curiously.

The second formulation can be more convenient in practice because the marker positions are easier to measure than the position of the original interface. 9.45 are needed ∂η ∂x t ∂η ∂t x ∂c dη ∂c = ∂x t ∂η dx t ∂c dη ∂c = ∂t x ∂η dt x 1 . The alternative approach of using two diffusion ˜ and marker velocity v) is based on the difficult coefficients {DA . ˜ 9. we accomplish a related goal of ˜ extracting the concentration dependence of D(c).50) . some derivatives of Eq.46) In what follows. The deeper observation is that for interdiffusion across a moving interface. Sadly. In a concentration gradient. 2 t dη = 17:29:48 (9. and depend on the local composition. DB } (instead of D(c) practice of monitoring the diffusion process in a laboratory coordinate system. Another challenge comes from how chemical interactions give a concentration dependence to DA (c) and DB (c). 3. The mathematical trick for what follows is the “Boltzmann substitution” x η≡ √ . ∂η dη (9.3 Variable D(c) In an alloy of uniform chemical composition there is no diffusional flux. In the present section we do not solve the diffusion equation for a concentrated alloy. v} (Eqs. however.48) (9. and they are the only variables upon which η depends (see discussion of Eq. t1/2 1 x = − 3/2 .49) (9. ˜ there is no general way to solve a diffusion equation with variable D(c).47) (9. t dη 1 x dc = − 3/2 . We do not need to measure the velocity v of physical markers. t (9. 9.42).44 ond approach are the interdiffusion coefficient and the marker velocity {D and 9. It uses one diffusion equation (Eq. this is also true for c dη ∂η = ∂x dx ∂η dη = ∂t dt ∂c dc = .43). 3. the local fluxes vary with position. but in the course of the analysis of a diffusion profile we extract a position for a “Boltzmann–Matano” interface by a condition of conservation of solute. 2t 1 dc = 1/2 . A nonconstant ˜ D(c) violates our basic assumption of Sect. but for a one-dimensional diffusion couple.1 that all atoms jump independently without regard for their local environment. The variables in this sec˜ . 9.37). even when two chemical species have different diffusion coefficients. For this reason the partial derivatives of η are equal to the total derivatives.224 Atom movements with the vacancy mechanism sample for reference.45) It is handy to note that x and t are independent variables.2.

50 and 9. 9.55) The definition of η (Eq.58) Finally.47. Multiplying through by t helps to simplify −  2  1 x dc dD dc d c = + D .49   ∂ 1 dc 1 x dc = D 1/2 .9. 9. − 3/2 2 t dη ∂x t dη   ∂ dc 1 ∂D dc 1 1 x dc = 1/2 + 1/2 D . we return from the variable η to the coordinates x and t that are more appropriate for analyzing a measured concentration profile in a diffusion couple.53) Since the concentration profile depends on η (and not x and t independently). and the right can be simplified with the chain rule   d dc η dc = D . t = t . dη t ∂η2 dx (9. Setting equal increments dη on the two sides of Eq. 9.53 are more fundamentally written as derivatives with respect to η − 1 x dc 1 = 1/2 3/2 2 t dη t  dD dη dη dx   2  d c dη 1 dc + 1/2 D .56.56 are now concentration [atoms/cm3 ].45) is found on the left. The two derivatives with respect to x in Eq. − 3/2 2 t dη ∂x dη t ∂x dη t (9. and integrating from c = 0 to a c = c that is less than the maximum c0 for the diffusion couple   dc η . the xdependence of D also originates from its dependence on η. 9.52) (9. (9. (9. by Eq. which corresponds to the amount of time that diffusion has occurred in our 17:29:48 .57) − dc = d D 2 ∂η 1 − 2 c 0 η dc = D(c ) dc dc  − D(0) . ∂η c ∂η 0 (9. 2 t1/2 dη dη dη ∂η2 (9.51) ∂t ∂x ∂x Substituting the partial derivatives from Eqs. We do so by fixing a time. 9. and therefore may vary along x when there is a concentration gradient   ∂ ∂c ∂c = D .56) − 2 dη dη ∂η The units of this Eq.54) The two derivatives dη/dx are simply t−1/2 . 9. making it appropriate for further analysis of concentration profiles c(x).2 Phenomena in alloy diffusion 225 The one-dimensional diffusion equation is now written in a way that D could be a function of composition. (9.

i.60 1 dx ˜ ) = − D(c 2t dc c c x dc . 9. 9. when integrating to c = c0 1 − 2 c0 x dc = 0 . This implies that the areas “1” and “2”. D(c) can be evaluated with a slight rearrangement of Eq. Derivatives with respect to η are transformed to derivatives with respect to x at a fixed time t .62) . called the “Matano interface.0 1 c/c0 0. In particular. ˜ After finding the Matano interface. Using Eq. 9.e.8 0.4 0. the net integral is zero.2 2 t Fig. 9. because there is no variation in concentration in the diffusion couple at c = 0.10. From Eq. 9.60) 0 Figure 9.45 and noting that dc/dη = dc/dx ∂x/∂η = dc/dx t1/2 1 − 2 c x t1/2 dc = D(c ) dc 1/2 dc − D(0) t1/2 .59 is zero. are equal when we have chosen the correct Matano interface.0 x=0 Distance The integration of Eq. 9. 9. (9.61 tells us that. 0 17:29:48 (9. which corresponds to distances far from the interface 1 − 2 c dc . marked in the figure.60 we obtain.Atom movements with the vacancy mechanism 226 1. The ˜ first step in extracting a D(c) from a profile c(x) is to find the origin x = 0.59) 0 The last term in Eq. the correct x = 0. 9.6 0..” It corresponds to the marker position if there is no change in atomic density after the diffusion process. implying that the areas of regions 1 and 2 are equal if x = 0 is chosen properly. far to the left the concentration is c0 and dc/dx = 0.10 0. diffusion couple.10 shows a representative concentration profile from a diffusion couple. integrated along the vertical “c-direction” in Fig. dx c x dc = t D(c ) (9. The condition dc/dx = 0 is true at locations far from the interface where the diffusion profile does not yet reach. t dx c dx 0 (9.61) 0 Equation 9. however.60 over the full range of c gives a value of 0 (Eq.61). This can be done iteratively with experimental data.

Even without marker ˜ velocity information.62 to evaluate D(c) at different compositions c between 0 and c0 . Sometimes these are the same interactions that lead to phase transformations. the largest displacement of solute along x is at small c.10. 9. and it is quite common for new phases to form in the interdiffusion zone of diffusion couples (Sect.4 0.8 1.11b. (b) Reorientation of the functions in part a.3 Diffusion in a potential gradient In inhomogeneous materials.4. 17:29:48 . it is sometimes more comfortable to redraw the diffusion profile as shown in Fig.0 t Fig. Since the range of c integration is vertical. along with the tangent curve dx  needed for the analysis.6 nt Integral c' 0. 9. which tends to move the atom into these regions. 9. we see that in both cases the B-atoms diffuse faster than A-atoms. 9.62 and the (inverse) slope needed for ˜  ).11 a dx d c c’ – ge 0. The (negative) gradient of a potential is a force. The next section addresses such chemical issues in more detail. 9.10 with Fig. This is consistent with Eq. Here the tangent curve is used without inversion. In ˜ Fig. showing the integral of Eq.1.7b. and integration is along the horizontal axis. such as dence of D(c) differences in their jump frequencies or correlation effects. 9. A force on a mass causes acceleration.2 c' 0. and may have a lower chemical potential in these regions. with the range of integration The evaluation of a particular D(c dc shown. 9.0 (a) Use of the Matano analysis of interdiffusion.6 c/c0 0. big correlation effects (small f ) can originate with chemical interactions between the species. an atom will find some regions to be more favorable energetically. ˜ Knowledge of D(c) is valuable for estimating diffusion behavior.0 0.44 when DB is large because DB cA is large when cB is small.10). As discussed in Sect.3 Diffusion in a potential gradient Distance 227 1. D(c) can be used to provide the spatial extent of the interdiffusion zone. 5. The time t is assumed known from the experimental record. for convenience.4 x=0 Distance b 0.8 + x=0 n Ta c/c0 0. 9.0 0. 9.4 The concentration depen˜ can originate with mobility differences between the chemical species. Counteracting 4 Comparing Fig. indicating that D(c) is largest for small c. 9. 9. evaluation of D(c ˜  ) is shown in Fig.11a.2 0.9. ˜ so it is possible to use Eq. and can give some information about the diffusion profile. and an increase in the velocity of the atom.

9. The physical picture is shown in Fig. but the velocity of the particle is unknown. Figure 9. (b) Approximation of the process in part a. so it is easy to obtain the average velocity of the particle. we first consider the simpler problem of the effect of collisions on a particle moving under a general force. dx (9.12 V 0 ? 0 0 t tc 0 a tc b 2tc t (a) A collision occurs at time tc . however. which is the acceleration a during the time tc v= 1 a tc . which prove to work well when we consider the average effect of numerous collisions. where m is the mass of the particle. J. v. and the force F is obtained from the potential U as dU . F (9. Unfortunately.12a. since it bumps into them continuously.12b. this acceleration are the interactions of the moving atom with all its surrounding neighbors.” M M≡ v .12b reminds us that this average velocity is half of the maximum velocity. and Fick’s first law. and our particle may come away with positive or negative velocity. depending on the details of the interaction.64) 1 dU tc . 2 (9. these motions of atoms in a potential gradient are not diffusive behavior. Under a steady force the particle accelerates.63) The acceleration is F/m. the acceleration of the particle is as before. The details depend on the temperature and of course the motions of the other atoms in the interaction. illustrated in Fig.Atom movements with the vacancy mechanism 228 V V ? t Fig. Specifically. After the collision.65) F=− so v=− So far. We assume that it takes a brief time. and we relate them through the flux. Before addressing the trajectory of the atom in a chemical potential gradient. 9. with linearly increasing velocity. We therefore make some approximations. A collision occurs at time tc . 9. 2m dx (9. 17:29:48 . we assume that the velocity of the particle is reset to exactly zero after a time of acceleration tc . D.66) whereas diffusive motion is parameterized by a diffusion coefficient. Collisions are assumed to occur at regular intervals. The two types of dynamics are related. we do not know all the details of the collision. The motion of an atom in a potential gradient with collisions is parameterized by a “mobility.

The force on the electron is −eE. ∂μB kB T = . (9.77) ∂cB cB and JB = −MB kB T dcB . dx 17:29:48 (9. (9.67) and with Eq. or an ideal solution (9.76) μB = kB T ln cB + ln Vq .70)  1 J = −en eE tc 2m ne2 tc E. a flux of B-atoms is a concentration times a velocity (units are [atoms/cm3 ] [cm/s] = [atoms/(cm2 s]) JB = cB vB . 9. (9. is the product of the electron charge e.68) JB = cB MB FB . (9.3 Diffusion in a potential gradient Box 9.74) σ ≡ Ohm’s law takes the familiar form Again. The flux of current.2 Ohm’s law Ohm’s law for electrical conductivity is now obtained easily.72) ne2 tc . (9.78) . and the average electron velocity v J = env . dμB .73) J = −σ E . not distance.71) (9.69) JB = −cB MB dx where the chemical potential μB is the potential that generates the force on the B-atom. the electron density n. (9. so with Eq. 2m (9.75) ∂cB dx It is easy to work with the chemical potential of an ideal gas (where cB Vq = eμB /kB T ). J.229 9. The chemical potential depends inherently on composition.65 (9. J=− 2m Defining the electrical conductivity σ as  . so it is more useful to write ∂μB dcB JB = −cB MB . 9.66 (9.

This isotope effect can sometimes be used for isotope separation.79) DB = MB kB T . We can obtain a more realistic chemical potential from Eq.19 to describe diffusion in a solid solution with chemical interactions. A characteristic diffusion distance x = Dt. μB = N ∂cB μB = Using this Eq.19 (9.81) 2m The collision time is a tricky quantity to obtain. ∂x (9. ∂NB 1 ∂G .85)  The concentration dependences of the chemical potentials are calculated   ∂μA 4Vz cA cB + kB T ∂μB = = . 2. since it originates with details of interatomic interactions. the diffusion coefficient of √ deuterium.82) (9.83 with Eq. ∂cB ∂cA cA cB 17:29:48 (9. 2. 1 H. (9.84) . and add corrections later for more realistic expressions for chemical potentials. recalling that the chemical potential μB is the change in free energy per change in number of B-atoms ∂G .80) JB = −DB it is found that It is traditional to use this expression for DB . Comparing to Fick’s first law Eq.81 is useful for relating diffusion coefficients of different isotopes. appropriate for an ideal gas or dilute solution.83)  cB μB = z (eAB − eAA + cB 4V) + kB T ln 1 − cB and likewise for the A-atoms (cA = 1 − cB )  μA = z (eAB − eBB + cA 4V) + kB T ln cA 1 − cA  . 2 H. 3. so protium diffuses about 41% further than deuterium under the same conditions. (9. (9.3 The isotope effect The isotope effect on diffusion can be understood by combining Eqs. 9. 9. assuming that an isotope has identical chemical interactions with surrounding atoms. Nevertheless. the mass denominator in Eq. Likewise the factor of kB T is not what dominates D in a thermally activated process. (9.230 Atom movements with the vacancy mechanism Box 9.80 to obtain tc D= kB T . and 9.66.64–9. 9.86) .20 ∂cB . For example. is half that of protium.

87 (and 9. The chemical factor in the square brackets of Eq.89) transform to Eqs. there is no contribution from B-atoms that change sites with other B-atoms because this has no effect on the concentration profile. For an equiatomic alloy at the special temperature T  = −Vz/kB .9.87 and 9.89. There is no diffusive flux in any concentration gradient. 9. 9.49).89. 2.88) where MB kB T .87 and 9. We saw in Sect.3 Diffusion in a potential gradient 231 and substituted into Eq. 17:29:48 . since cA → 0 and DB apparently diverges in an unphysical way. 9.78 (and 9. suppressing the effective value of DB .87 and 9. 9. The chemical factor in Eqs. This temperature corresponds to the critical temperature of chemical unmixing (Eq. kB T dx (9. which is proportional to cA . The resolution to the dilemma is in the mobility.88 is less than 1. 9. Now consider an alloy with ordering tendencies so 4V > 0. 9. JB (T  ) = 0.89) (9. cB DB = (9.90.75 and its equivalent for A-atoms to give   dcB 4Vz cA cB +1 . We could add a correction such as MB → cA MB in Eq.” They can effectively increase or decrease the diffusion coefficient. The other limit. the chemical factor is [−1 + 1] = 0. so the atom movements do not occur in a disordered solid solution.24. and work with constant diffusivities in Eqs. 2. Eqs.89. so 4V < 0 (from Eq.90) The terms in the square brackets in Eqs.88 are called “chemical factors. cA MA DA = kB T .36).1. 9. JB = −DB kB T dx   dcA 4Vz cA cB JA = −DA +1 . As the concentration of A-atoms goes to zero in regions rich in B-atoms. For the flux of B-atoms.88 is greater than 1. 9. It is necessary for a B-atom to exchange sites with an A-atom if it is to contribute to JB . the chemical factor is [+1 + 1] = 2. 9. In calculating a mobility for use in a diffusive flux. so diffusion behavior does not apply. all B-atoms are in the same chemical environment. there is no contribution to JB . For an equiatomic alloy at a temperature T  = Vz/kB .87 and 9. 2. In this limit.80) when cB → 0.5 It is perhaps best to ignore the explicit temperature and concentration dependences of DB and DA in Eq. There are also important factors in MB from a vibrational attempt frequency and lattice dynamics. and heterogeneities in composition are leveled more rapidly. and this overwhelms the proportion- ality to kB T in Eq.87) (9. might seem to give a peculiar result with Eq. but other considerations are important. 9. so intermixing is enhanced. First suppose we have an unmixing alloy.89 and 9.88. and no leveling of heterogeneities in chemical composition. Consider the effects of chemical preferences on interdiffusion. as expected if unlike atom pairs are preferred. 9.4 that diffusion 5 The mobility usually depends exponentially on temperature as e−Q/kB T . 4V = eAA + eBB − 2eAB ). each surrounded by all A-atoms. MB . This particular T  corresponds to the critical temperature of ordering (Eq. 9. too. where cB → 1.

the free energy of the two subvolumes is (δV1 + δV2 )F(c0 ). Alternatively. although this is not a state of thermodynamic equilibrium for the temperature T. by the amounts δc1 and δc2 .6 In essence. It is risky to use Eqs.91) where the subvolumes δV1 and δV2 have compositions that deviate from the mean composition of the alloy. It is assumed that a fraction. or perhaps during the severe mechanical damage inflicted by high-energy ball milling. the concentration profile can be expressed in a Fourier series. 17:29:48 . For each sine wave in the series. f . we can match positive and negative subregions so that Eq.2. When the alloy is homogeneous (i.232 Atom movements with the vacancy mechanism in an ordered alloy. provided there is not too much change in the local atomic configurations in this short time. one enriched and the other depleted in solute. but driven towards mixing by ballistic atom movements.6 for calculating the response of a system to small departures from the steady state can be used to obtain a critical temperature for unmixing. Actually.. 90].90 at low temperatures – the alloy may be undergoing a phase transformation so that its diffusion coefficients are changing with time. δc1 = δc2 = 0). 12. for which traditional free energy versus concentration curves are relevant. Under such conditions. so that δV1 δc1 + δV2 δc2 = 0 . occur thermodynamically. over time the alloy reaches a state of equilibrium for the conditions of temperature and rate of forced ballistic jumps. 9. This section uses the model of Martin to treat the stability of concentration heterogeneities in an alloy driven towards chemical unmixing by thermodynamics. regions depleted in solute must account for the same amount of solute as the regions enriched in solute.4 Nonthermodynamic equilibrium in driven systems In a model for driven systems developed by Georges Martin [89. These ballistic atom jumps occur at random. especially at lower temperatures. the approach of Sect. even for temperatures above Tc there will be significant short-range order in the alloy. c0 . Examples of ballistic jumps are atom motions when an alloy is bombarded by energetic radiation such as fast neutrons. conservation of solute can be included with the method of Lagrange multipliers for the volume integral of the free energy density. This conservation condition can be obtained by matching pairs of small subvolumes.87 and 9. 1 − f . some atom movements occur by thermally activated processes. 9.91 is true. Conservation of solute requires any concentration fluctuation of positive sign to be compensated by a fluctuation of negative sign.88. of atom movements occur ballistically. 9.e.3. 9. which can overwhelm the effects of the chemical factor in Eqs. without influence of the local chemical environment of the moving atom. and a fraction. can have a small correlation factor f . The new feature of Martin’s model is a second type of “ballistic” atom movement that occurs without any thermodynamic bias. over short time intervals it may be useful to consider atom fluxes with a diffusion equation. Since it is a steady state. and atom movements will be altered. Nevertheless. where F(c) 6 As shown in Sect. 3. (9.89 and 9. however.

3.91      dF dF δF = δV1 δc1 − . Nevertheless.97) (9. 9.7 We make the same type of assumption as we did after Eq. i. dt = −4MδV thermo (9.98 shows that F is then a restoring force that will suppress a concentration fluctuation δc. so absolute rates of growth or shrinkage of a composition fluctuation are not found.9. In our present treatment.92) (9. δF.93) (9. With the composition fluctuation.6) is  2  d F δF = δV 2 (δc)2 .98) recognizing that the second derivative of F evaluated at c0 is a constant.95) = δV 2 δc dc2 c0 The change of free energy F(c) caused by the composition fluctuation δc can be either positive or negative. however. δc1 = −δc2 ≡ δc by Eq. δV1 = δV2 ≡ δV. so  2  δF d F δc .. 9. Following Eq.95 determines the thermodynamic tendency for the composition fluctuation to grow or shrink. there is no parameter such as surface energy or gradient energy to set the spatial scale of the fluctuations. 2. = −δV 2 dδc dδc dc2 c0  2  d F F = −4δV δc . dc c0 +δc1 dc c0 +δc2 (9. 3. is δF = δV1 F(c0 + δc1 ) + δV2 F(c0 + δc2 ) − (δV1 + δV2 )F(c0 ) . as discussed in Sect.     dF dF + δV2 δc2 . The 0 negative sign of Eq. (9. 9. the change in free energy.91. 17:29:48 . 9. d2 F/dc2 is a positive constant. the effective potential energy for a composition fluctuation (as opposed to an order parameter in Sect. dc2 c0 (9.e. (9. as expected near the critical temperature for unmixing.103 for the ordering kinetics.100) c0 7 In the case where the alloy of composition c is stable against unmixing. If the subvolumes are equal.100. the kinetic tendencies can be interpreted with the results of Sect.6. 3.4 Equilibrium in driven systems 233 is the free energy per unit volume. the higher derivatives of F(c) can be ignored.  dδc dt  dδc = MF .96) dc2 c0 The effective restoring force against a composition fluctuation is  2  dδF d d F F =− (δc)2 . δF = δV1 δc1 dc c0 +δc1 dc c0 +δc2 Using the conservation of solute of Eq. the rate of change of δc is proportional to F.94) For small δc1 and δc2 . The sign of Eq.99)  d2 F dc2  δc . 3. depending on the sign of d2 F/dc2 . (9.6.

using Eq. we calculate the critical temperature for spinodal unmixing. Before we add the ballistic jumps.22 with Eq. and were obtained by thermodynamic arguments plus the assumption that the rate of change of the composition fluctuation scales with the mobility times the deviation of the free energy from its value in thermodynamic equilibrium.234 Atom movements with the vacancy mechanism The subscript “thermo” indicates that all the arguments so far have been without ballistic jumps. since this is almost at hand. 2. For the free energy (per atom).23 . 2.

9. c(1 − c) dc2 (9. These occur without consideration of the chemical environment. so their contribution to the rate of change of the concentration can be obtained by making T very large in Eq. Eq. dt total c(1 − c) The critical temperature is obtained by setting (dδc/dt)total = 0 for small δc   (1 − f )M . which was just worked out (Eq.36. z (9. 2. 9.e. Substituting Eq. (9.103) which is the result of Eq.107) (9. giving kB Tc = z 4V c(1 − c) . generalized for any concentration c. 9.103). (9. When f = 0.105) dt ball c(1 − c) where a different mobility M  was assumed for ballistic jumps.102 into Eq. 17:29:48 .108) Equation 9. Eq.104   4M  δV δc kB T dδc =− . 21 and 22 in Martin’s paper [89]. To get the total rate of change of concentration.105     dδc M + f (M  − M) = (1 − f )M4Vz − kB T 4 δV δc . M  = 0.108 is equivalent to Eqs.108 reduces to the thermodynamic result of Eq.102) and setting d2 F/dc2 = 0.104 and 9.101) F(c) = c(1 − c) 4V + kB T c ln c + (1 − c) ln(1 − c) . Likewise. 9.103 8 A second way to obtain the critical temperature is to find the condition where (dδc/dt) thermo = 0 for small δc. 9.. This is proportional to d2 F/dc2 .103. kB Tc = z 4V c(1 − c) M + f (M  − M) (9. 2 The critical temperature is found by calculating d2 F/dc2 d2 F kB T = −z 4V + . 9.100 gives8     kB T dδc = 4MδVδc z 4V − . when the ballistic atom movements have no effect on the concentration fluctuation.106) dt total dt thermo dt ball Substituting from Eqs. 9. i. 9. (9. (9.104) dt thermo c(1 − c) We now add the ballistic atom movements to the thermodynamic ones. we add the rates for ballistic and thermal jumps (recall that f is the fraction of ballistic jumps)       dδc dδc dδc = (1 − f ) +f .

√ mi are needed to get correct vibrational 17:29:48 . y3N Y y3 y2 B A P S t Fig. The potential energy is a hypersurface in this space. Eq. When the atom species are different. (9.9 and we make the big supposition that the normal coordinates {yi } are all known. The other extreme of f = 1 or M = 0 gives a Tc = 0.) In intermediate cases.9. the normal coordinates are essentially position coordinates. It is of limited predictive power on its own.5. 9. which are independent of each other.5 Vineyard’s theory of diffusion 235 is again recovered. (This is intuitive – it means that there are only random atom motions. the prefactors energies and frequencies in what follows. When M = M  . At any instant in time. 9. 9. The crystal requires 3N coordinates {xi } to describe uniquely all atom positions. so unmixing does not occur. 9 When all the masses are the same (a pure elemental material). but its strength is for elucidating the issues and processes of atom motions by activated state processes. which ought to mix the alloy.13. We transform to normal coordinates √ yi = mi xi . The theory is based on the statistical mechanics of atom vibrations and velocities. 9. What it does that is new is address the vibrational degrees of freedom. showing saddle point P and a general point Y on the ridge S. It predicts the familiar activation energy for an atom movement in a potential energy landscape.108 reduces to the simple result kB Tc = z 4V c(1 − c) (1 − f ) .5 Vineyard’s theory of diffusion George Vineyard developed an elegant and general theory of diffusion [91]. from which it predicts the vibrational entropy of diffusion as a prefactor for the rate of atom jumps.109) showing that the critical temperature is directly proportional to the fraction of thermodynamic atom movements.13 y1 Potential energy hypersurface ({yi }). the critical temperature for unmixing is reduced by the effects of ballistic jumps. the crystal is at a single point in a hyperspace spanned by its {yi } coordinates.1 Degrees of freedom for atoms in a crystal Assume we have a crystal with N atoms with masses {mi }. Initial and final points of the crystal are A and B. 9. as depicted in Fig.

9. to the left of S. There is a saddle point on the ridge at P. A key feature of this hypersurface is that it has local minima at points A and B. A general shape of the potential energy hypersurface ({yi }) with some labeled points is shown in Fig.236 Atom movements with the vacancy mechanism Box 9. so most systems are concentrated near the point A and the integration is not sensitive to its upper limit. 2 0 √  ∞ π a3 2 −x2 /a .4 Gaussian integrals The following results are needed frequently in what follows  ∞ √ 2 e−x /a dx = π a . 9.2 Ensemble-averaged jump rate To get an average rate with statistical mechanics. A where ρ0 is a normalization factor. The potential well is expected to be deep. which correspond to the crystal before and after the diffusive jump of one atom. and • QA is the fraction of systems around point A. x e dx = 2 −∞ (9. It is = I .114) QA = ρ0 e−({yi })/kB T dV .13.13.113) where • I is the fraction of systems that reach the surface S per second. and are moving to the right in Fig. 9. we need to make an ensemble of many similar crystals. and the integration is over the volume of the hyperspace around A.112) and the crystal is expected to explore different regions on this hypersurface as all atoms vibrate and as one atom makes a diffusive jump. Obtaining QA is by a straightforward thermal average of systems in the potential well around A  (9. These two points are separated by a ridge in the potential marked by S. and define as the ensemble average for the rate of transitions from A to B. 17:29:48 . and crossing the ridge at any point Y will suffice to get the diffusive jump.110) (9. Our goal is to calculate the average rate at which systems make the transition from A to B.111) (9. −∞  ∞ a 2 x e−x /a dx = .5. QA (9.

normal to the surface S at Y. (9.124) dI = dS · y˙+1 y˙ 1 =0 {˙yi }=0.119) −∞  ∞ 3N 2 ρ0 e−(Y)/kB T = ρ1 e−(Y)/kB T e− i=1 y˙ i /2kB T d˙y1 d˙y2 . . 9.121) and from Eq.117) It is assumed that v2 is independent of location on the hypersurface of Fig.122) i=1 ρ1 = ρ0 . (9. . 9. but the ensemble averaging may still be successful. ∞ i=1 −∞ e−˙yi /2kB T d˙yi . and does not equilibrate as it moves. assume that y1 is aligned along the direction of dS. A conceptual difficulty arises if the diffusing atom moves ballistically over the barrier.13. . as shown in Fig.5.10 The density at Y of crystals having velocity v is ρ(Y. written as {˙yi }. 9.13  ∞  ∞ y˙ 1 ρ(Y. Some ballistic character is expected. 2 (9. .110 ρ0 = ρ1 3N ) . d˙y3N . 2π kB T .5 Vineyard’s theory of diffusion 237 9.116) 2 + · · · + m3N x˙ 3N . v = 2 m1 x˙ 12 + m2 x˙ 22 (9.5. (9.120) −∞ The independent normal coordinates {yi } are separated ρ0 = ρ1 3N  . . (2π kB T)3N/2 (9.123) 9. i=1 10 This is reliable if the system is in equilibrium. The transition through the point Y occurs by varying y1 . v) dv d˙y1 . 17:29:48 .118) The normalization factor ρ1 is calculated with the condition that the crystal must have some velocity. (9. . (9.9.. v) dv .3 Transition states Obtaining I requires attention to the rates of change of the coordinates {yi }.e.  i. (9. .115) Its square is twice the kinetic energy of all atoms in the crystal v2 = y˙ 21 + y˙ 22 + · · · + y˙ 23N . y˙ 3N ) . (9. y˙ 2 .4 Transition rate We next consider the contribution dI to the flow of systems across the surface S at point Y. v) dv = ρ1 e−(Y)/kB T e−v 2 /2k BT d˙y1 d˙y2 . First. d˙y3N . and the probability of the system at Y is ρ0 e−(Y)/kB T  ∞ −(Y)/kB T ρ0 e = ρ(Y. . The velocity of the independent normal coordinates is defined as v = (˙y1 .

Harmonic vibrations modify the potential energies11 at points A and P as ({qi }) = (A) + 3N  1 j=1 2 ωi2 q2i . the transition rate of Eq. Small displacements of harmonic vibrations occur as the crystal coordinates are near both points A and Y. and the oscillation frequency of a mass m on this spring is ω = k/m. It is a many-body expression that includes coordinates of all the atoms in the crystal.129) Equation 9. Something we do not know is the contour of the ridge along S. giving ) ρ0 dI = dS √ e−(Y)/kB T kB T ( 2π kB T)3N−1 . kB T S e−(Y)/kB T dS I .5 Harmonic vibrations We make further progress by considering harmonic vibrations of the atoms in the crystal. y1 . with all atoms vibrating about their positions of equilibrium. 9.130) 11 The potential energy of a harmonic oscillator of angular frequency ω is 1 ω2 q2 . All 3N coordinates are available for vibrations around point A.113 is $ . = QA 2π A e−({yi })/kB T dV √ 2π kB T.128. is used to constrain the system to be on the surface S. We expect that the ridge is high. which has the lowest potential energy along S.118 and 9.123  ∞ ρ0 2 −(Y)/kB T dI = dS e y˙ 1 e−˙y1 /2kB T d˙y1 (2πkB T)3N/2 0 3N  ∞ .129 is a ratio of two configurational partition functions.111 and 9.128) (9.125) i=2 0 The two integrals are evaluated with the help of Eqs. 3N ( 2πkB T) $ −(Y)/kB T kB T dI = ρ0 e dS . 2π $  kB T e−(Y)/kB T dS . (9. I = ρ0 2π S With Eqs. In what follows we replace the energy at a general point Y with the energy at point P.5. but there is one less coordinate for point Y – one of the coordinates. 9. 2 × e−˙yi /2kB T d˙yi .114 and 9.127) (9.110 as kB T and respectively. so there is a low rate for a crystal in the ensemble to attain this potential energy. = . 9. (9. Using Eqs.) 17:29:48 . We therefore approximate the transitions over the ridge as all occurring at the saddle point P. 9. (Recall that the energy of a 2 i i √ spring is 1/2kx2 .238 Atom movements with the vacancy mechanism where the integration over y˙ 1 is separated from the others. but there is a difference in the number of normal coordinates at these points. 9. (9.126) (9.

(9. 9.136) (9.140) . Substituting Eqs.133) (9. 12 π i=2 ωi E = (P) − (A) . 9.130 and 9. (9.5 Vineyard’s theory of diffusion 239 ({qi }) = (P) + 3N  1 j=2 2 ωi2 q2 i .110 which evaluate to the √ form 2πkB T/ω -3N √2πkB T −(P)/kB T e i=2 kB T ωi √ (T) = . 9. (T) =  2π 3N i=2 ωi $ (9.1.131 into Eq. The q1 from displacements along y1 is missing from the {qi }.129 $ (T) =  . 1 2 2 − 3N j=2 2 ωi qi /kB T dA kB T S e−(P)/kB T e .132 are Gaussian integrals of Eq.138) i=2 3N   ωi ln ωi i=2 -3N i=2 ωi -3N  i=2 ωi  . Svib ≡ SP − SA = kB ωi ωi i=2 Svib = e+Svib /kB = (9.134) 9.  1 2 2 2π . 2π e−(A)/kB T 3N 2πkB T i=1 ωi -3N 1 i=1 ωi −[(P)−(A)]/kB T e .134 for (T) allows us to write the diffusion coefficient of Eq. 17:29:48 . and {qi } around P. 7.137) Using Eq.131) where the small displacement coordinates around A are {qi }.6 Diffusion coefficient D(T) This expression 9. 9.139) (9.1) -3N f a2 i=1 ωi D0 = -3N  .132) The integrations in Eq. 9.8 for the vibrational entropy  3N      3N  kB T kB T ln ln − .9. (9.5.135) where (adding the correlation factor f of Sect.16 as D(T) = D0 e−E/kB T . e−(A)/kB T e− 3N i=1 2 ωi qi /kB T dV A (9. (9. 3.

This is necessary so that a temperature can be defined and the tools of statistical mechanics are applicable. for example. The Vineyard theory and its extensions assume a quasistatic process for the atom to surmount the activation barrier.2 The diffusion coefficient in one dimension with uncorrelated atom jumps between adjacent planes is: D = 1/2 a2 .1 (a) Explain in words (with diagrams and/or equations if helpful). The F includes both the activation energy and the activation entropy of the diffusive jump. (c) Explain the origin of a marker velocity in a diffusion couple. Such phenomena may be important at high temperatures. but the distance from the line is along one dimension.Atom movements with the vacancy mechanism 240 so the prefactor D0 is D0 = f a2 ω1 e+Svib /kB . Calculate in an analogous way the diffusion coefficient in one dimension when there are both jumps between neighboring planes (separated by a) of frequency 1 . 9. It is already a multibody treatment of diffusion. Problems 9.3 One hundred jumping beans are placed along the center line of a large floor at 6 in intervals. and jumps between second-neighbor planes (separated by 2a) of frequency 2 . one fast and the other slow. but a more detailed theory would consider other atom movements around a vacant site. 9.) 17:29:48 . why solid-state interdiffusion involving two species of atoms. The Vineyard theory advances our understanding of diffusion. This is the frequency of the phonon that moves the atom at A towards the saddle point at P. 12 π (9. differs from interdiffusion of two ideal gases. Molecular dynamics computer simulations are not subject to the same constraints. one might add coordinates for the neighboring atoms that need to be displaced to allow the diffusing atom to enter the vacancy [92]. Instead of one coordinate y1 that reaches a critical condition. and the sum of the squares of the distances divided by 100 is 36 in2 . Twelve hours later the distance of each bean from the center line is measured. Assume that the atom density [atoms/cm3 ] remains constant during interdiffusion.141) Recognizing the change in free energy between points A and P as F = E − TSvib D(T) = f a2 ν1 e−F/kB T 6 (9. (b) Explain the origin of the Kirkendall effect and the origin of Kirkendall voids. and can allow for ballistic motions of a vacancy across more than one neighbor in a quick sequence.142) where the attempt frequency is ν1 ≡ ω1 /2π. (This is a two-dimensional problem.

9.241 Problems (a) Calculate the diffusion coefficient of the jumping beans. (b) If the mean jump distance of a bean is 0. and the probability of one jump to the right and one to the left is pq. estimate the mean jump frequency of a bean. (a) Show that the probability for moving a total of n steps to the right out of a total of N steps is the binomial probability P(N.1 in. and the probability q of a jump to the left. The probabilities of a set of multiple jumps are products of p and q. there is the probability P of a jump to the right. for example the probability of two jumps to the right is the product pp.143) (b) Show that the average value of n can be expressed as n. n) = N! pn qN−n (N − n)! n! (9.4 In a one-dimensional random walk.

144) (c) Calculate n. = p ∂ (p + q)N ∂p (9.

. and use the same technique to calculate n2 .

In obtaining cos θ1 . show that more vacancy jumps tend to be strongly trapped when the lattice has a lower coordination number. (d) What is the mean displacement after N jumps? (e) What is the mean squared displacement after N jumps? (Hint: The function (p + q)N is the “generating function” for the various P(N.6 For equiatomic ordering alloys with the same T/Tcrit and |U|.7 Consider a random walk of an A-atom on an ordered structure such as a B2 or chessboard structure. The combinatorics of finding a trap in a largely disordered solid solution can be approximated combinatorially as 2−ζ ..5. for diffusion on a simple cubic lattice. f . (b) How would your result change if you considered the first four jumps? (Reasonable answers get credit. consider only the first three jumps of the vacancy. 9.145) It is sometimes more convenient to manipulate (p + q)N than the individual P(N. 9.14 from Bakker [87]. but you will also need the probability of finding a trap. 9. n) because of the binomial expansion (p + q)N = N  n=0 N! pn qN−n (N − n)! n! (9. n). z (at least for z ≥ 3). where each jump is to a nearest-neighbor on the opposite sublattice (α and β denote the sublattices).) 9.5 (a) Calculate the correlation factor. You will need to consider the relationship between eAA − eAB and both the normalized critical temperature and the probability of exiting a strong trap. and Q 17:29:48 . where ζ is the number of atoms needed to construct a strong trap like that in Fig. We use the definitions of Fig.) 9. where the notation P means the probability of a jump from the α to β.

9.148 when there are no differences between the sublattices? 17:29:48 . from β to α. note that cos θ1 = Q+ (+1) + Q− (−1) = −q .14 f =1+2 ∞  cos θj .151) and the correlation factor for an A-atom on the β-sublattice is fβ = show that 1 1 = f 2  1−p .149) (a) Show that the correlation factor is f = 1 − p − q + pq .150 equivalent to Eq. 9.146) p ≡ P− − P+ . 1 − pq (9. (9. 9. 9. (9.14. Define also q ≡ Q− − Q+ .) (b) If the correlation factor for an A-atom on the α-sublattice is fα = 1−q . (9. 9.148) j=1 To calculate cos θ1 . where n is labeled on the right.14 P+ P– Q– First P– P+ Second Definition of jump probabilities for random walk on an ordered structure with sublattices α and β. and work more cos θj terms with Fig.150) (Hint: Perform the mechanics of long division on Eq. 1+p 1 1 + β fα f (9. The variables + and − mean that the jump is in the same direction or reversed from the previous jump (examples are in Fig.14).153) (c) What is needed to make Eq.147) The correlation factor is analogous to that of Eq. 9.150. 9.Atom movements with the vacancy mechanism 242 β α β α β Zeroth Q+ t Fig. (9. (9. for example. Solid circles are starting points for the nth jump.152)  . 1+q (9.

) c(x) = c0 17:29:48 .8 For convenience.154) (a) Use an expression of this form to find the position of a Matano interface. (9.243 Problems 9.155) 2 (Hints: Use a mathematical software package for parts a and b. assume an interdiffusion profile for a fast and a slow species with the form c(x) = c0 tanh(e−x ) . ˜ (b) Calculate D(c) for a few positions at equal distances to the left and right of the Matano interface. Consider the inversion symmetry in x for part c. (c) How and why would you expect the results of part b to change if the c(x) were 1 + tanh(−x) ? (9.

17:29:48 .

Electronic and magnetic phase transitions can sometimes be understood with similar approaches as phase transformations involving atom rearrangements.PART III TYPES OF PHASE TR ANSFORMATIONS Part III describes the important and established families of phase transitions in materials. Chapter 17 describes some of the major phase transitions involving electrons and spins that occur inside materials. and microstructural and nanostructural aspects of phase transformations are covered in Chapter 16. All these phase transitions involving atom rearrangements are historical figures in the field of materials science. results from condensed matter physics offer new insights into phase transformations in materials. Chapters 10–16 describe structural and chemical phase transformations of materials that occur by movements of atoms. This is emphasized in Chapter 18. These include heterogeneous first-order transitions such as melting and precipitation. 17:29:51 . and spinodal decomposition and ordering that may occur homogeneously as second-order phase transitions. Martensite and other displacive phase transitions are the subject of Chapter 15. which ends by touching on quantum criticality. and new phenomena are often explained with reference to them. Likewise. Research on quantum phenomena such as superconductivity is often reliant on controlling the structures of materials. but some aspects of electronic or magnetic excitations are not classical.

17:29:51 .

and of course the equilibrium thermodynamics are the same for either melting or freezing. but is discussed in more detail in this chapter.” The glass transition was described in Sect.2. 1. 1.1) Schematic behavior is shown in Fig. this chapter describes some features of melting in two dimensions.10 Melting The free energy of melting was discussed in Sect.4 (repeated here) Sl − Ss = where the latent heat.1. a glass undergoes a softening akin to melting. called the “glass transition. is the equality of the free energies of the solid and liquid phases Fs (Tm ) = Fl (Tm ) . Finally. the criterion for melting. 1. Chapter 10 presents additional considerations about the enthalpy and entropy of the solid and liquid phases near the melting temperature Tm . 10. 5. and highlights rules of thumb. is El − Es L = . Tm Tm . Correlations between Tm and the amplitude of thermal displacements of atoms (“Lindemann rule”).11. such as the tendency for the entropy of melting to be similar for different materials. At a temperature well below the melting temperature Tm . (10. and for determining the melting temperature Tm . 1.1 Free energy and latent heat As described in Sect. Thermodynamic equilibrium between the solid phase and the liquid phase was the starting point for the discussion on alloy freezing in Chapter 5. and between Tm and the bulk modulus B are justified with simple models. but these correlations are semiquantitative at best.1. and discussed in Eq.3. which is quite different from melting in three dimensions. L.3.

2) (10. (10. 17:29:53 . With latent heat comes an arrest or slowing of either heating or cooling at Tm . The release or absorption of latent heat has a significant effect on how a material is heated or cooled through its melting temperature. Sl (Tm ) − Ss (Tm ) is similar for many materials. L ≡ Sl (Tm ) − Ss (Tm ) Tm .3) The sign of the latent heat is such that L is absorbed during melting. so from Eq. and released during freezing. As discussed below.3 the latent heat is expected to scale with the melting temperature. Without latent heat there is only a monotonic change in temperature with time. 10.

become mobile as they undergo a “pre-melting” at temperatures well below Tm . The situation can be similar to that shown and discussed in Fig. and the surface liquid grows rapidly into the bulk. 17:29:53 . When the solid forms.1. the latent heat raises its temperature and the temperature of the adjacent undercooled liquid. Ss }.1 In many cases of rapid cooling. At temperatures above Tm there is no need to nucleate the liquid phase. Lu–Pt) have a rise and fall of Tm from left to right. 5. however.” Sometimes the heat release during recalescence is enough to raise the local temperature to Tm . or certainly not to the extent it is possible for undercooling a liquid. A modest difference is that recalescence need not occur at the surface of a growing solidification front. • The rare earth metals from La to Tm have similar melting temperatures. many details are needed for assessing chemical trends for the thermodynamic functions of solids and liquids that determine Tm . The mechanism is not well understood. after accounting for the rate of heat removal. Superheating a solid above Tm does not seem to be possible. There are several important trends in Tm across the periodic table that can be assessed with trends in chemical bonding. 1 Arrests or kinks in cooling curves of T(t) are common identifiers of other phase transitions. the local temperature of the new solid does not rise above Tm . 10. Es . W) have the highest Tm . For realistic calculations. and this is a research challenge today. Sl .2 we see that a detailed calculation of Tm requires the functions {El . A thermal arrest during cooling is usually a robust and reliable way to determine Tm . This chapter presents some trends of melting temperatures of the elements in the periodic table. the liquid can be cooled below Tm before the solid phase has a chance to nucleate and grow. Y–Pd. The energies and entropies of solids vary with the nature of their chemical bonds and crystal structures.248 Melting The duration of the arrest may also be expected to scale with Tm . This process is called “recalescence. where the temperature remains constant for a time. 10. • The alkali metals and noble gases have very low melting temperatures. the elements with four s. but can occur when crystals nucleate inside the liquid phase. Variations in the energy and entropy of liquids are similarly expected. too. which have lower coordination than in the bulk.2 Chemical trends of melting Melting temperatures of the elements are listed and depicted graphically in Fig.and p-electrons (C.7e for one-dimensional solidification. but some tendency for the melting temperature to increase towards the end of the series. The elements with five d-electrons (Cr. Si. It is believed that the atoms at the surface of a solid. • The actinide metals from Ac to Es have a peculiar trend of Tm . 10. Mo. • The three rows of transition metals (Sc–Ni. with general relationships to bonding energy and entropy. During recalescence. however. Ge) have the highest Tm . From Eq. or at least the differences El − Es and Sl − Ss . • From left to right across the main group.

249 t Fig.1 shows this to be approximately true. The first two or three trends can be understood with the bonding and antibonding arguments of Sect. This argument can be used to explain the chemical trends of Tm for the transition metals with their ten d-electrons. with numbers in K.1 10. 10. the maximum cohesive energy occurs when the band structure is half full.5 describes the simple but semiquantitative Friedel model that includes some of the essential features of chemical trends of band filling. Figure 10. On the other hand.2. although in a broad band of states the highest bonding states will be similar in energy to the lowest antibonding states. From Lawson [93]. Nevertheless. and graphical depiction with diameters proportional to melting temperature. For a tight-binding electronic band structure. Problem 10. the midpoints with maximum bond energy are the C column of the periodic table for the main group elements. so all the bonding states are occupied and the antibonding states are not. The outer radii of the 17:29:53 . and the Cr column for the transition metals (assuming one s-electron in a band of its own). the 4f -electrons of the rare earth metals do not give this trend because they are somewhat contracted. 6.and p-electrons. The maximum Tm is therefore expected when the band structure is half full.2 Chemical trends of melting Melting temperatures of the elements. From this argument. lying inside the outer 6s-electrons. half the states in the band will be bonding and half will be antibonding. and for the main group elements with their eight s.

3. thermal expansion.3.2). however. (2) bulk modulus (curvature at bottom of the potential. an interatomic potential (r) departs from a parabola because (r) is skewed to larger r.Melting 250 rare earth metals do contract across the series from La to Tm.3 Two consequences of the skewed potential of Fig. and (2) a type of phonon softening (meaning a reduction in phonon frequency) caused by thermal expansion. Surprisingly.2 are obtained easily: (1) thermal expansion. In general.. (2) curvature. negative cubic contribution is added to the interatomic potential to give it skewness (x) = 1 2 kx − bx3 .2 shows implicitly that the interatomic potentials for many elements have the same ratio of cubic to quadratic components. What we learn from Fig. allowing some f -electron bonding in the later part of the series and a gradual rise in Tm . 3 When a cubic term is added to the quadratic potential to approximate a range near the bottom of the potential. Predicting properties of the actinides. (2) aj . (3) x0j . but this is oversimplified.2 is that this ratio of skewness to curvature is similar for many metallic bonds. 17:29:53 . 10. (3) interatomic distance (horizontal offset). That is.1 Energy. 10. Fig.3 Free energy of a solid At the melting temperature Tm . (3) horizontal offset. 2 (10. This obvious fact about parabolas motivates the rescaling of more realistic interatomic potentials of different elements by their (1) binding energy (vertical offset). 10. Sect. cf. The filling of the 5f -states in the actinide elements from Ac to Es gives a peculiar trend in Tm . 10. The electron energy levels in the actinides are strongly sensitive to the atom positions and coordinations. including melting.2 [94. 6. Both can be calculated approximately if a small. 95]. 10.2 One consequence is that a large number of crystal structures occur in actinide elements as functions of temperature or pressure.4) 2 This was once attributed to the strong angle. the free energies of the solid and liquid are equal. all functions yj = y0j + aj (x − x0j )2 can be superimposed into the same curve if we rescale them with the three constants (1) y0j . phonon softening All parabolas look alike.and distance-dependence of the f -electron wavefunctions. 10. These are the (1) vertical offset. Knowing the free energy of the solid is half the problem of understanding the thermodynamics of melting. remains a research challenge. this rescaling of different interatomic potentials places them on one universal curve shown in Fig.

10. Points were obtained from Mo.8 –0.6 –0.5 –0. +∞ − 1 kx2 /k T bx3 /k T B e B dx B dx xe 2 −∞ x e x= = −∞ . K. +∞ −(x)/k T .5 as . Reprinted figure with permission from [95]. (10. whereas even moments are  +∞ 2 x2n = x2n e−ax /kB T dx . all odd moments are zero. Cu. To calculate the thermal average x.4 –0.9 –1.7 is the convenience of evaluating moments of Gaussian functions.6) Z= −∞ −∞ To evaluate Eq. where x is the displacement from the minimum of the potential.−∞ . · (2n − 1) π 2n+1  17:29:53 kB T a n+1/2 .10) . (10.2 0 1 4 5 2 3 Scaled separation (a*) 6 7 8 “Universal” interatomic potential of Rose et al. follow the usual prescription for probabilistic averages .1 Scaled binding energy (E*) –0.0 –1 t Fig. Recall that for integration over ±∞. Copyright (1981) by the American Physical Society. (10. 10. +∞ −kx2 /2kB T 3 (1 + bx /kB T) dx −∞ e (10. Ba. (10.7) We therefore approximate Eq. 95] (solid curve). 10. assume bx3 /kB T is small and ebx 3 /k BT 1 + bx3 /kB T .5) Z Z where the classical partition function for normalization is  +∞  +∞ 1 2 3 e−(x)/kB T dx = e− 2 kx /kB T ebx /kB T dx . [94.7 –0.8) An ulterior motivation for the approximation of Eq.10. . Sm.3 Free energy of a solid 251 0 –0.9) −∞ x2n = √ 1 · 3 · 5 · .5.3 –0. (10.2 –0. 10. . +∞ −kx2 /2k T B (1 + bx3 /k T) dx xe B x .

so substituting for x from Eq. so thermal expansion is positive by the argument here. For example. but it is usually only part of the reason why phonons shift their frequencies with temperature. the square root is expanded to obtain $   k 9b2 kB T ω(T) 1− . so ω decreases linearly with T owing to thermal expansion.12) where x0 is the interatomic distance used for defining the potential. the quasiharmonic model predicts no changes in phonon frequencies. This “quasiharmonic” effect on phonon frequencies is important. One way to calculate this is to consider how the curvature of the interatomic potential changes with thermal expansion.11) or x = x0  3b kB k2 x0  T = αT .14) The “quasiharmonic model” assumes that phonons remain harmonic with increasing temperature. 10. and this is usually true.2. causes the Boltzmann factor average of atom displacements to favor the positive x side of the potential.11 d2  b2 = k − 18 kB T ≡ k . dx2 (10. k2 (10. 10. 10.4 is d2  = k − 6bx . ω = k /m. 10.15) m k3 All factors in the second term of Eq. (10. 10. A pure temperature dependence of phonon frequencies often occurs.Melting 252 Discarding an odd moment in the numerator and one in the denominator of Eq. This frequency shift can be calculated with the effective force  the angular frequency is constant ) k of Eq. if volume were fixed so there is no thermal expansion.13) and we are interested in the curvature at x. approximated by adding a cubic term to a harmonic potential. Phonon frequencies change with temperature. (10. In our convention b > 0. however. and α is a linear coefficient of thermal expansion. and originates with phonon–phonon interactions and electron–phonon interactions as described in Chapter 24. but are shifted in frequency because the atom vibrations explore a different part of the interatomic potential.15 are positive.14. The skewness of the interatomic potential of Fig. Assuming that the dimensionless number 18 b2 kB T/k3 is small.8 x √ π  5/2 kB T k/2  1/2 kB T k/2 3 b √ 4 kB T π = 3b kB T . giving the fractional change in length per interval in temperature. As usual for a harmonic oscillator. 10. dx2 x k2 (10. 17:29:53 . The second derivative of Eq.

A temperature-dependent expression for the entropy of a solid includes these individual terms Ss (T) = Sh (T) + Sqh (T) + Sanh (T) + Sel (T) + Sepi (T) . The other terms are smaller corrections for physical effects described below. which dominates the entropy. 7. g(ε). elemental solid at Tm comes from its vibrational and electronic degrees of freedom. Both these dynamical contributions to entropy grow with temperature.2 Separable contributions to the entropy The entropy of a nonmagnetic.3. but Sh (T) is evaluated with the Planck distribution for elevated temperatures.16) as discussed in this section. Here Sh (T) is the harmonic contribution. characteristic of a low temperature. Near Tm . Here. an average energy for the phonons ω is used .3 Free energy of a solid 10.253 10. for illustration. the simpler high-temperature expression of Eq. The harmonic entropy is calculated with a phonon density of states (DOS).52 may be appropriate. (10.

) The sign is negative because phonon frequencies usually decrease with expansion. 10. 7. ω ∂V (10. for small α. Using Eq. as shown by Eq.18) predicting a phonon frequency at T that differs from the frequency ω0 at T = 0 ω(T) = (1 − 3γ αT) ω0 . k T  B +1 .17 gives. (10. The fractional shift of a phonon frequency with fractional change in volume defines the Grüneisen parameter γ of Eq. 10. 10.17) Sh (T) = kB ln ω The phonon DOS and ω will change as the solid undergoes thermal expansion because the interatomic potential is not parabolic. (10.  .19) (Since α is the linear coefficient of thermal expansion.15.74 γ ≡− V ∂ω . 3αT is the fractional change in volume with temperature.19 in Eq. for example.

(10.20) Sh (T) + Sqh (T) = 3kB ln ω0 (1 − 3γ αT) .  kB T +1 .

Often we 17:29:53 . [95]. k (10.2 by comparing Eqs. For the scaled potential of Fig. More details are given in the original ref.12 for the thermal expansion γ = b x0 . With b/k 1/4 in these units.2.19. and Grüneisen parameters from analyses of heat capacity data are also appropriate. γ is approximately 1. 10.22) We cannot know γ without the bond length x0 .21) + 1 + 9kB γ αT .5 to 2. The best sources are phonon frequencies. this x0 may be 6 to 8 for most metals [95]. 10. 10. 10. It may be more convenient to obtain Grüneisen parameters from other sources. Sh (T) + Sqh (T) = 3kB ln ω0 We can obtain a Grüneisen parameter from the universal equation of state of Fig. and using Eq.15 and 10. k T  B (10.

is nearly equal to the sum of the harmonic.) The details prove a bit more complicated. the phonon DOS at low temperatures is rather different from the phonon DOS at higher temperatures. Sepi . 10. such as in Eq. Contributions of a few percent of this are still important when comparing differences in free energies of phases. 101]. 24.16.254 Melting assume γ = 2. and is a contribution that is not present with the thermal excitations of either electrons or phonons alone. As shown in Fig. It does make an important contribution to the entropy of transition metals and their alloys. do interact with each other to change their frequencies when there are cubic or quartic components of the potential. 10. 10.19.1). At 900 K the harmonic contribution to the free energy −TSh is approximately –48 kJ/mol. and add contributions to correct for various nonidealities that increase with temperature.19. Sh (T) + Sqh (T) + Sel (T). which is not predicted by the quasiharmonic model. Figure 10. and it is small for semiconductors and free electron metals. the phonon frequencies are renormalized by the quasiharmonic frequency shift. 24. and anharmonicity causes additional shifts in phonon energies that may augment or detract from the quasiharmonic shift of Eq. and electronic contributions. The additional terms of Eq.3. 10. The overall shift of the phonon frequencies is approximately as predicted from the quasiharmonic model. 10. and Sepi are less well understood. The perturbation from cubic and quartic components causes a shortening of the lifetimes of these renormalized phonons. 24. Consider the harmonic part of the entropy of aluminum shown in Fig. It turns out that the phonon frequency shifts caused by the quartic part of the phonon anharmonicity 17:29:53 .6. however. which can be important in some metals at elevated temperatures. and vary for different materials. These effects are described in Sect. even at elevated temperatures.3. First. initially described as harmonic normal modes that do not interact. Many of the other terms in Eq. the adiabatic electron–phonon interaction.66). but it varies by a factor of 2 or so for different elements. so it is typical to hope they are small. 10.3 Contributions to the entropy of fcc aluminum What we have done for the entropy of the solid was to start with the entropy of a perfect harmonic crystal at low temperatures.4.5. 6. one approach is to treat them as perturbations that couple harmonic phonons.16 are of this size. This approach is appropriate because in practice Sh (T) is the largest contribution. however. When the cubic and quartic components are small. Sanh . This is very large. Sel . and usually varies for different phonons in the same element (see Sect. Finally. The electron–phonon interaction requires thermal excitations of both electrons and phonons. 102]. even though the large lifetime broadening implies a big cubic anharmonicity. 10. (The contribution Sepi is negligible for aluminum [100. quasiharmonic. It accounts for how phonons.5. makes a contribution to the entropy. The electronic entropy Sel depends on the electronic states near the Fermi level (Eq. The anharmonic entropy Sanh depends on temperature and not volume.3 seems to show that the total entropy of fcc aluminum. primarily because of a large lifetime broadening of the phonon modes [96. assessed by the Scientific Group Thermodata Europe (SGTE) [98. and aluminum is nonmagnetic. 99]. For insulators Sel is zero. described in Sect.

.314 J/(mol K).3. The harmonic curve Sh was calculated from Eq. Magnetism gives an important contribution to the entropy [103].3 Free energy of a solid 255 60 7. is shown. so there is no net effect on the entropy from the large anharmonicity of aluminum. The phonon frequencies change with temperature approximately as predicted by the quasiharmonic model. with an approximate correction for the lifetime broadening. 10.0 Sel 10 1. To this anharmonic entropy was added the electronic contribution Sel [97]. i.e.3 800 400 600 Tempertaure [K] 1000 Entropy versus temperature for fcc aluminum from the Scientific Group Thermodata Europe “SGTE database” as top dark curve.0 Sqh 0. Aluminum is a bit unusual in that it is nearly a free-electron metal with a low density of states at the Fermi level.5. so its electronic entropy is several times lower than for some transition metals like iron.0 0 200 t Fig. largely cancel the frequency shifts from the cubic.0 5. which dominates the heat capacity near the Curie temperature.20.0 SGTE database 6. At very high temperatures there is another contribution from vacancy formation that causes a discrepancy in Fig. 7. 4 Iron has a substantial contribution from magnetic entropy. but the cancellation of entropic contributions from cubic and quartic anharmonicities is not generally expected. 10. Recall that 1 kB /atom corresponds to 8. 24. the quasiharmonic contribution Sqh . interpolated at intermediate temperatures. but this effect is already included in the phonon DOS measured at elevated temperatures. For reference. giving the thin curve Sh (T) + Sqh (T) + Sanh (T) + Sel (T). 17:29:53 .10. Using the anharmonic phonon spectra measured at elevated temperatures. the phonon part plus the electron part of the entropy successfully give the entropy of fcc aluminum (Sh (T) + Sqh (T) + Sanh (T) + Sel (T)). The anharmonic vibrational entropy was obtained from experimental phonon densities of states measured on Al metal at eight different temperatures [96].4 The quasiharmonic contribution Sqh (T) in aluminum is fairly typical of other materials. and corrected for the lifetime broadening of the phonon spectrum.0 Entropy [kB/atom] Entropy [J/mol K] 50 20 2. and this is seen more clearly in the heat capacity of the solid in Fig.0 30 3. 9Bvα 2 T of Eq. and has effects on phonons [104]. 10.50 with a phonon DOS curve g(ε) measured at low temperatures. shown as the bottom curve.0 Sh + Sqh + Sanh + Sel 40 Sh 4.

The heat capacity of solid aluminum is shown as a dashed curve below Tm .5 Melting Phonon DOS curves of fcc aluminum metal.4 t Fig. 17:29:53 . versus temperature in units of Tm . 10. After [105]. Heat capacity of liquid aluminum in units of 3kB per atom.256 t Fig. 10. obtained from inelastic neutron scattering measurements at the temperatures specified in the labels (units are K) [96].

4 Entropy of a liquid The free energy of the liquid phase is the second essential piece of the thermodynamics of melting. and we use it to write a trial partition function for one atom in the liquid Z1t =  .27 differs from that of the ideal gas where the large volume V is available to each atom. Sect.. so the total volume of the liquid is V = N. This starting point requires a substantial correction. Vq (10. Atoms are mobile in the liquid phase. however.24) where λ is the thermal de Broglie wavelength. but unlike a gas. they stick together as they move around. The challenge is to identify the correlations between atom positions in the liquid that reduce the initial overestimate of the entropy. 17:29:53 .26. unlike a gas where the atoms are effectively indistinguishable at any position in its container.27) Equation 10.5 5 For both the gas and the liquid. This gives the factor of π in Eq. Here Vq is the quantum volume Vq = λ3 . For the entropy of a liquid we take the opposite approach than for a solid. Unfortunately. 2m h λ= √ . but a quantitative analysis is out of reach. 10.5.4 Entropy of a liquid 10.. but indistinguishability of atoms gives a factor of 1/N! to the right-hand side. a factor V/V can be obtained by integrating a Boltzmann factor over phase q . √ space as exp[−(p2x + p2y + p2z )/kB T]h−3 dpx dpy dpz dxdydz. because the atoms in a liquid have chemical bonding and are not so randomly dispersed as in a gas. invokes an experimental observation to obtain a useful thermodynamic criterion for melting. atom motions in a liquid are more similar to atom motions in a crystal than atom motions in a gas.257 10. obtained approximately by equating the thermal energy to the quantum kinetic energy (h/λ)2 . and start with a liquid having complete disorder.23) where  is the volume per atom. 2π mkB T kB T (10. They may be approximated as local atomic vibrations.25) (10. (10. This section explains how this can be done. The concept of a quantum volume is useful for the positional degrees of freedom of a liquid. The next section. Over short times. the atoms do not explore much of the total volume of the liquid. (10. it is more difficult to understand than the free energy of the solid. 10.26) For an idealized liquid of N noninteracting atoms  Zt =  Vq N . In many thermodynamic respects.

3.30) Scorr = −kB 2. Scorr is fairly large.7 ln Tm This is an average expression – the detailed behavior varies with chemical element. Sometimes an electronic contribution needs to be added.29) + kB + Scorr . Assessing the correlation correction can be done by analysis of the atomic structure of the liquid. the same as the vibrational heat capacity of a harmonic solid. (10. and at the high densities of a liquid this reduces considerably the fraction of their accessible volume. the liquid has a constant heat capacity of 3 kB per atom. impeding quantitative thermodynamic predictions.6 − 1. The difference for the gas is that the volume explored by each indistinguishable atom is its fraction of the entire volume occupied by the gas.6 Unfortunately. In what follows. but the volume explored in a solid is mostly from the excursions of a distinguishable atom as it vibrates. Atoms in a liquid are excluded from overlapping with other atoms.000 K. Wallace [106] suggested that these can often be neglected. 7. is expected to be largest just above Tm where the liquid is most viscous.5 shows the heat capacity of liquid aluminum from molecular dynamics simulations [105]. giving an approximate result found by Wallace [106]    T . Evidently the atomic dynamics in the liquid absorb thermal energy in ways similar to vibrations of atoms in a crystal. as discussed below. this 1nn peak decreases in amplitude.7 As a rough approximation. For this reason we need to add a correlation correction to the entropy of the liquid. we will use the corrected entropy of the liquid. and its atoms are packed most tightly. but chemical elements are as low as 8 (Li) and as high as 16 (Th) [106]. Slt    3 Slt = kB ln (10. Figure 10. Scorr . and a term 5/2 kB on the right-hand side instead of 3/2 kB .5 were obtained from simulations from the melting temperature to the boiling temperature of aluminum. Sl . 17:29:53 .28) + kB . There is a prominent peak in the heat capacity just above the melting temperature 6 There are also higher-order correlations with three.and four-body terms.9 kB /atom near Tm . The actual boiling temperature is half this value.5 kB /atom near Tm .28 looks similar to the Sackur–Tetrode equation for the translational entropy of ideal gases. which has V instead of . 7 The results of Fig. which is approximately 9. Sl = kB ln Vq 2 The correlation correction. calculated to be approximately 5. we obtain a trial entropy per atom for the liquid. so the scaling of the temperature axis needs to be taken with some caution. The pair correlation function for liquids shows a peak at first-nearest-neighbor (1nn) distances. and this Scorr is of negative sign    3 (10. and these are poorly understood. as suggested in Sect. being around 26% of the entropy of the liquid at temperatures near Tm . 10. For temperatures above Tm . A typical Slt is 12.258 Melting Following the prescriptions F = −kB T ln Z and S = −∂F/∂T. Vq 2 Equation 10.

the free electron density and the energy of Eq.” although changes in electronic energies are not surprising when there is a large change in atomic structure in going from a diamond cubic crystal to a liquid. known as “Richard’s rule. For example. 9 A more general average over elements for S is 1. and Sl is approximately 8 They are very similar for the alkali metals. which should seem surprising.. is a semiconductor as a solid. Silicon. but a metal as a liquid.8 kB /atom.2.2 gives S = 0. 10.1 Entropy difference of solid and liquid We use Eqs. which expresses the equality of free energies of solid and liquid phases.7 ln(Tm /Tm ) = 0) S = Sl − Ss .8 kB /atom upon melting.8 kB Tm . if S = 0.21 and 10. if a free electron metal does not change substantially in density upon melting. Wallace [106] says these elements undergo “anomalous melting. so the chemical bond energy is not possible to calculate with confidence. Finally. Such a consistent similarity of the electronic energy of the solid and liquid cannot be justified rigorously.259 10.30 is zero (i. but sometimes this approach is not too bad.68 will be similar for both the solid and liquid phases.8 kB Tm . the elements for which S > 0.29 to obtain the entropy change at melting.5. 6. although it is larger for many others.31) 3  . which therefore have low melting temperatures. 1.8 kB /atom are those that undergo a significant change in electronic structure upon melting.8 kB /atom.  S = −4. S 3. at Tm the second term in Eq. Conveniently.9 Interestingly. and then solve for Tm . 10. This loss of structure occurs as shear stiffness is diminished over various length scales in the liquid. 10. Our approach. and this is associated with thermal expansion against a bulk modulus – the CV does not show this increase at high temperatures. This is frustrating because the atomic structure of the liquid is not well known. we now need El − Es .5 Thermodynamic condition for Tm (just above 1 Tm in the figure). is to set El − Es = 0.e. 10. This high value of CP is associated with the loss of structure characteristic of a more ordered state. At higher temperatures the heat capacity increases again.” B 17:29:53 . and a spike in CP is expected when there is a discontinuity of entropy at Tm . 10.1 k /atom. S should be the same for all chemical elements. then Eq. For Si and Ge.8 If El − Es = 0. For this to be useful in general.32. for example. Recall that the energy difference originates with changes in electronic energies when the structure changes from solid to liquid.32) To calculate Tm with Eq.1 kB − 9kB γ αTm + kB ln  Vq  ω0 kB Tm (10. There is a minimum in CP at approximately 3 Tm . (10.5 Thermodynamic condition for Tm 10. It turns out that S = 0.8 kB /atom for many elements. 10. Germanium and tin also have an increase in electronic entropy upon melting. We can use this S for the left-hand side of Eq.

9 + ln  . we have the vibrational energy at melting. but the Grüneisen parameter γ is more of a problem.32 from quasiharmonicity. rather than sound wave velocities that sample only phonons of the longest wavelengths.9 kB /atom near Tm .Melting 260 9. Sanh (T) is not small for many elements. Frequently the term in Eq. ω0 . The most important of these parameters is ω0 . 10.36 is the energy of a harmonic oscillator 1/2 mω2 x2 . we can estimate melting temperatures in terms of the vibrational properties of the crystal itself? 10.7% of the Wigner–Seitz radius. 10. are properties of the solid phase. and we assume S = 0.038 ω . (10.8 kB . (10. and take the bigger risk that always Sl − Ss = 0. This is reasonable if the Debye temperature is determined from the heat capacity of the solid.2 Lindemann rule for Tm All of the unknown parameters in Eq.17. In short. 10. then.177 rWS .32 can be rearranged as  √ 3  2π mkB Tm ω0 0.37) π rWS 3 2π 0  2 1 kB Tm = m ω20 0. Perhaps. is neglected because it is assumed small (it is graphed on Fig. Also neglected is any true anharmonicity of the solid phase. and we can deduce the maximum atom displacement at melting. 10.8 kB /atom.32. If we know them.3 for fcc Al). rWS   4 3 2/3 m 2 kB Tm = 0.8 kB = −4. Sqh (T) = 9kB γ αTm . Sanh (T).34) h kB Tm  $ 3  3 m ω0 ln Tm = −4. (10. Ss (Tm ) 9.5. When this energy is equal to kB Tm .35) 2 2π kB m 2 ω . we can solve for Tm . we take the risk of neglecting quasiharmonicity and anharmonicity in the solid. There is a temptation to use a Debye temperature in place of this average. so Eq. an average phonon frequency defined so that ∞ ln(ω0 ) = g(ε) ln ε dε . (10. (10. The sum of the Wigner–Seitz volumes for all atoms must add up to the total volume of the 17:29:53 . γ . where x is its maximum displacement.1 kB + kB ln  . 10. (10. 10. then for the entropy of the solid near Tm .38) 2 The solid melts when its atom displacements are 17. The thermal expansion α is often known. express the atomic volume  as a sphere around each atom with a radius called the “Wigner–Seitz” radius.1 kB /atom. α.36) kB Tm = 0.33) 0 This average frequency is designed to be representative of the phonons that contribute most to the vibrational entropy in Eq.038 2/3 2π 0 Emerging from Eq. and although this is acceptable for aluminum. To complete this connection.

and there have been a number of attempts to explain it [107. although with a correction for rWS A proportionality between Tm and B is a fairly common assumption in arguments about materials properties. giving a plausible result for little effort. Aspects of this criterion have been known for perhaps 100 years. m 17:29:53 . that it is possible to describe essential behavior with a smaller number. unlike a crystal which has only a few. and for a fcc crystal. 10.38 offers good value.3 Correlation of Tm with bulk modulus Suppose we make a natural assumption for Eq. too.” The obvious problem with this explanation is that it misses the thermodynamic condition of equal free energies of solid and liquid phases at Tm . Improvements are also possible by adding corrections for how the electronic entropy Sel changes upon melting.10 Nevertheless. rWS is 55% of the 1nn distance.36 that the mω20 is a characteristic interatomic force. and this seems a promising avenue for more research. Using the fcc crystal. the Lindemann rule survives to this day because Eq. where N is the number of atoms. although some elemental trends are improved by adding the factor of rWS Nevertheless. that the melting temperature Tm is proportional to B.1 for the melting temperatures.261 10.6 shows the bulk moduli for the elements.5. The reader should be a bit annoyed that the thermodynamic properties of the liquid are not part of the Lindemann condition for melting. rWS is 62% of the interatomic spacing. We may therefore expect 2 . 108]. The agreement 2 . perhaps where N is of order hundreds or thousands. These coordinates could be the Cartesian coordinates for individual atoms. is qualitative. 10. For a simple cubic crystal. for a bcc crystal it is 57%. 109. however. An often-mentioned but tortured idea is that the “crystal shakes itself apart.1 Energy landscape A glass of a given enthalpy has a high number of equivalent atomic configurations.6 Glass transition 10. but this is an excuse and not a reason. and it is interesting to compare this figure to Fig.6. the “Lindemann rule for melting” is A crystal melts when its atom vibrations reach 10% of the interatomic spacing. and this force is proportional to the bulk modulus B.6 Glass transition material. it is evident that across most rows of the periodic table. We hope. the range in melting temperatures is not so large as the range in bulk modulus. Liquid properties were neglected primarily because they are not well understood. 10. 10. but also unlike a gas which has far more [43. but for transitions between configurations 10 The anharmonicity of the crystal at T can be large. 110]. according to Lawson [93]. Figure 10. 10. The number of spatial coordinates for all atoms in a glass is 3N. This neglect is partly responsible for the inaccuracies of the rule – it is reliable only to 20–25%.

During annealing. as envisioned in Fig. Experimental evidence suggests that these local wells have higher curvature after the glass is given a thermal treatment. thermal activation causes regions in the glass to make transitions between states as in Fig. and there is an average movement down the energy landscape.7a. Temperatures of measurement were from 1 to 300 K. and the macroscopic 17:29:53 . and the distribution of these energies and entropies set the thermodynamics of the glass. The minima of the small features in Fig. Four local minima. more regions leave states of higher energy than enter them. along which changes proceed.” that allows the atoms to settle into more favorable local configurations. A consequence is that most atoms tend to be in stiffer potential wells.262 t Fig. These changes in configurations cause changes in the internal energy of the material. and the density of minima along the reaction coordinate is depicted at the right of Fig.7.7a. 10.” are indicated in Fig. termed the “energy landscape” of a glass. or “basins. 10.7a. an “annealing. After [46. Relatively few basins with the deepest minima or highest maxima are expected. 10. with numbers in units of 1011 Pa. Because the annealing is at a low temperature. 111].6 Melting Bulk moduli of the elements.” {χi }. ({χi }). and graphical depiction with diameters proportional to bulk modulus.7a represent atomic sites about which an atom vibrates at finite temperature. it is more useful to transform the coordinates to a set of “reaction coordinates. It is therefore interesting to speculate on the energy surface. 10. 10. 10.

7a and b can be understood with activated state rate theory [42. On a scale of a few hundred atoms. By increasing the energy in Fig. for three f = 3. From Fig. For a more complicated function ({χi }) having more coordinates. 113]. which represents a “strong” glass.11 These points are saddle points of the energy landscape ({χi }). A more detailed picture is a goal of ongoing research. 17:29:53 . however. of the number of saddle points to i i the number of extrema in ({χi }) could be obtained through binomial combinatorics of positive and negative  (3N)! signs.7a. For large N. or minima with {∂ 2 /∂χi2 > 0 ∀ χi }. Figure 10. (i. 10.2 Fragile and strong The transitions across Figs. but it is known that the enthalpy of crystallization is reduced after such annealings.7a).7a above the lowest-energy states. Qualitatively.. The analysis gives f = 12 3N j=0 (3N−j)!j! − 1. (b) Energy landscape of a “strong” glass.e. Saddle points represent instabilities in ({χi }) in at least one reaction coordinate. for two coordinates f = 1. and the density of saddle points in a multidimensional ({χi }).7a. 3N = 1) we obtain f = 0. giving f 23N . in metallic glasses it is believed that the internal stresses are reduced after annealing. 10. 10.7b.10. A guess at this thermal energy is indicated in Fig. where some ∂ 2 /∂χi2 < 0 and other ∂ 2 /∂χi2 > 0. so we know that an annealed glass has found a state in the lower range of energies in the energy landscape of Fig. we expect by statistics even more possibilities for mixed second derivatives. f can be approximated with the maximum term in the sum. For local minima and local maxima. 10. this general behavior may help account for why atom vibrational frequencies are reduced at elevated temperatures. The functions ρχ  (E) and ρ{χi } (E) are schematic densities of minima in (χ  ).6.7a it is easy to understand the zero slope at maxima where {∂ 2 /∂χi2 < 0 ∀ χi }. we expect more saddle points.. χi . so for one coordinate (i. f . 10. with label for the characteristic energy of a glass transition. the ratio. the complicated function ({χi }) should have many points where a coarse average of its derivatives is zero with respect to atomic coordinates. 10.e. 10. they can cooperate to give shear flow. and therefore give liquid-like behavior. is the simplest to 11 If the signs of the ∂ 2 /∂χ 2 were arbitrary when {∂/∂χ = 0}. elastic modulus goes up a bit. and saddle points dominate rapidly with higher numbers of coordinates. {∂/∂χi = 0 ∀ χi } for the four basins in Fig.6 Glass transition 263 F E E kB Tg a χi ρ χ (E ) i ρ χi (E ) F t Fig. 112. 10.7 b χi (a) Schematic diagram of energy landscape of a “fragile” glass. More significantly. it may help explain the origin of the glass transition itself – when more unstable modes are available by thermal activation.

as are some entangled polymers. and the glass transition appears less abrupt with temperature. As a general rule. 10. In many ways we have better intuition about what “melting” means in three dimensions compared with two. • Two-dimensional Potts model. measurements show that its viscosity decreases rapidly. In practice. The “fragile” glass of Fig. On the other hand. Glassy states are more difficult to achieve in fragile glass formers.7. • Two-dimensional vector interactions where the spin is confined to a plane. This includes the x–y model. Although there are many other coordinates that allow diffusional motions of atoms than shown in Fig.7a is more complicated because with elevated temperature.264 Melting understand because temperature gives access to more states in a systematic way. where there is a finite set of orientations of spins. Metallic glasses. the theory of melting in two dimensions is better developed than melting in three dimensions. the glassy state in a strong glass-forming material is much more robust. the idea is that the activation energy for diffusional processes remains approximately constant with temperature. metallic alloys and small molecules tend to be fragile glass formers. It is unfortunate that the results from two-dimensional melting do not transfer well to three dimensions. While less convenient for some analytical approaches than the Ising model. 10. are strong glass formers. however.1 Two-dimensional models We need to be careful about formulating a two-dimensional model because we have a choice of: • Three-dimensional vector interactions between atoms or spins arranged on a plane (such as a crystal surface). 10.7 Two dimensions There are major differences between melting in two dimensions and in three. It therefore has a finite set of pairwise interaction energies that do not require a computation of s1 · s2 for every pair. Curiously. materials such as SiO2 .7b. This is relatively late on the timescale for finding a new structure for metals. for example. where the pairwise interaction strength goes as s1 · s2 . were not reported until 1960 when rapid cooling techniques were available to Pol Duwez’s research group [34]. which is a Heisenberg model with sz = 0. and motion along the coordinates {χi } is much more rapid. This includes a Heisenberg model between spins. so spins are confined to the x–y plane. q of them (where q is perhaps 3 or 4). 10. The fragile glass has an activation energy for diffusion that decreases with temperature – above the glass transition (in the liquid phase). the Potts model can be handy for computer simulations. having dealt with liquid water and ice in daily life. which form random networks of strongly bonded atoms. 17:29:53 . more basins are overcome.

11. The dislocation energy increases with the volume of the crystal. however.2. In the simplest approach to the Kosterlitz–Thouless transition.3. the Potts model is identical to the Ising model. The configurational entropy of centering the dislocation on any crystal site of the 2D crystal is  S = kB ln R2 a2  . depends on the density of dislocations. When q = 2. 4π (1 − ν) rc (10. the crystal can deform easily by shear.41) . (10. similar to the case for ledges discussed in Sect. such as ordering or unmixing. With cooling. This argument does not hold in two dimensions. which is in units of dislocations per unit area. G is the shear modulus. but it does not have a transition to a phase with true long-range order. but it is useful to know that the Heisenberg model in two dimensions does not have a transition to an ordered phase at finite temperature. The proofs are beyond the present scope. 10.7 Two dimensions 265 • Two-dimensional Ising model. Being line defects. we seek the critical temperature for having a dislocation in a two-dimensional crystal. once a dislocation is known to pass through a plane.2 Kosterlitz–Thouless melting transition In three dimensions. and it exhibits a continuous phase transition upon cooling. ν is the Poisson ratio. 4π (1 − ν) rc a2 17:29:53 (10. Dislocation arrangements do not provide enough configurational entropy for dislocations to be thermodynamic defects in three-dimensional crystals. and rc is the “core radius” of the dislocation. but it becomes first order. assumed to be the same as the same as the lattice parameter and interatomic spacing.10.40) and the free energy of adding a dislocation is F=  2   R R Ga2 ln − T kB ln . the x–y model has a transition to a phase with order over a mesoscopic distance. The convenience is that the site variables are +1 or −1.7. When q > 4. We know that the Ising model has a phase transition in two dimensions. corresponding to some models of an A–B alloy. dislocations in a crystal have a high energy and low entropy. a phase transition still occurs in the Potts model. The vanishing of the macroscopic shear modulus is recognized as the onset of melting. with the spins parallel or antiparallel to a z-axis. The configurational entropy. perhaps 5a. it is also known that the line extends continuously above and below the plane. The idea is that once a dislocation is expected in equilibrium.39) where a is the Burgers’ vector. however. but the story is more complicated for the models with vectorial interactions. The energy per unit length of an edge dislocation is E=   R Ga2 ln .

for example by having the compression field from one dislocation overlap the tension field from the other. the weight of the cubic term in Eq. γ = 2. and the growth rate is slower? 10.3 For an anharmonic oscillator with a potential of the form U(x) = kx2 + cx3 + dx4 (10.2 was neglected for R  rc . however. assuming k = 50 N/m. 10. and the constant ln(25) 3. and they interact over large distances. The cancellation of the logarithm factors for energy and entropy was the key to this analysis.1 Use the quasiharmonic approximation of Sect. and an interatomic distance is 2 × 10−10 m.2 (a) Estimate a thermal velocity for an iron atom at its melting temperature.3 to calculate α. Assuming random signs for the partial derivatives with respect 17:29:53 . kB 8π (1 − ν) (10.266 Melting At the temperature Tc there is no cost in free energy to add one dislocation     25 R2 R Ga2 ln = Tc kB ln . This reduction in elastic energy lowers the melting temperature of the Kosterlitz–Thouless transition.43) (10. because dislocations interact strongly. Consider a system of twice the size that contains two dislocations.42) (10. 10. If the growth rate of the solid into the liquid is 1 cm/min. what fraction of iron atoms attach to the surface sites? (b) Why is this fraction so low? (Hint: You may challenge the assumption of a 1 Å travel. 10. 4π (1 − ν) rc rc Ga2 = Tc .4.44 gives an overestimate of the melting temperature.) 10.4 Assume you have found a stationary point in a function of all 3N coordinates of a system of 100 atoms. 10. Suppose an iron atom in the liquid must travel a distance of 1 Å to reach a suitable site on a growing crystal surface. Two edge dislocations can configure themselves to reduce the total elastic energy. The present approach would give the same critical temperature if the dislocations were independent. the linear coefficient of thermal expansion of a crystal.45) use harmonic expressions in the classical limit to calculate the vibrational entropy for small c and d.44) where rc = 5a was assumed.) Why is this fraction even lower if the undercooling is smaller. 4π (1 − ν) rc rc2       Ga2 R R ln = Tc kB 2 ln + ln(25) . How does α change with the Grüneisen parameter γ ? (Hint: Note the common dependence of γ and α on b. A more detailed study of the Kosterlitz–Thouless transition [114] has shown that Eq. Problems 10. This is not the case.

5 Compare the melting temperatures of the 4d transition metals with the cohesive energy predicted by the following simple model by J. graph the cohesive energy for the elements from Sr to Ag. show that the cohesive energy for an element with n d-electrons is Ew n(10 − n) . Make an estimate of Ew . 10. not 0. which we take as E = 0. with a width Ew and density of states 10/Ew (for 10 d-electrons). but the integral starts at −Ew /2.46) Ec (n) = − 20 (Hint: See Eq. Plot a suitably scaled set of melting temperatures from Fig. (10. The state at the middle of this band has the same energy as a collection of isolated atoms. Friedel. (a) Assuming the band is filled from bottom (most bonding at −Ew /2) to top (most antibonding). 17:29:53 .1 on the same graph. compared with the isolated atoms. what is the relative probability of the stationary point being a minimum.) (b) Assuming one nonparticipating 5s-electron. compared with the probability of it being a saddle point? 10. 6.267 Problems to these individual coordinates. In this model the d-electrons form a rectangular band. The antibonding and bonding states are symmetrically disposed above and below this state.67.

Most solid–solid phase transformations require some consideration of elastic energy. G–P zones form at a surprisingly rapid rate. The eutectoid transformation is described. A nucleus of a distinct phase must have an interface to the surrounding parent phase. Some examples are given. elastic energy. and its role in steel metallurgy is discussed in brief. but this is a large topic that may be of interest for further study. surface thermodynamics. suggesting an important role for elastic energy. Chapter 4 presented concepts of nucleation in a homogeneous medium or at a generic surface. Figure 11. and then grows as atoms diffuse to its surface. where a small particle of the new phase nucleates in the parent phase. and kinetic processes at surfaces. Finally. there remains a tendency to grow the microstructure to a larger spatial scale to eliminate the energy of the interfaces between phases.1 Guinier–Preston zones Guinier–Preston (G–P) zones are extremely small.11 Transformations involving precipitates and interfaces Most phase transformations in materials occur by nucleation and growth. and their shapes are often plates or disks. the first new particles can be quite small. but additional microstructural aspects of nucleation are presented here. this chapter explains important aspects of surface energy. Some ideas about this coarsening process are presented. even after the phase fractions are equilibrated. This interface between the precipitate and matrix has an atomic structure and chemical composition that is not simply a termination of bulk crystal of the precipitate and the matrix. They are precipitates of a sort. and the balance between surface energy. and volume free energy is altered as a precipitate grows and changes its shape. Diffusional transport to the new phase often controls the rate of growth. nonequilibrium precipitates that form at low temperatures in unmixing alloy systems. but numerous nucleation and growth transformations in different materials have received significant study. 11. For free surfaces and for interfaces between crystals. but have received special status owing to their nonequilibrium character and extremely small size. Especially at low temperatures. and the rate of the phase transformation.1 is a high-resolution transmission electron microscopy image of a Cu-rich G–P zone in a crystal of fcc Al. The details of the black and white contrast are not simple to interpret as atoms 17:32:08 . and their crystal structures may differ from the equilibrium phases that form with increased temperature or longer times. considering the low temperature. The presence of a plate of Cu atoms gives a dark row in the center of the image.

In a quenched solid solution.1 shows another important structural feature of G–P zones. 11. the quench preserves a high concentration of vacancies.1 evolve into G–P(2) zones that have two parallel plates of Cu atoms spaced by three (200) planes. With time the vacancies are annihilated by migrating out of the crystal to grain boundaries.1 Guinier–Preston zones or channels. Copper is also an fcc metal. The alloys often have rather dilute solute concentrations. With time. The Cu atoms are located in registry with Al atoms. an alloy is quenched rapidly from a high temperature where the solid solution phase is stable. This structure is somewhat closer to the equilibrium θ phase of Al2 Cu. although quantitative information is challenging to obtain. and diffuses in the matrix until it finds another solute atom. but with a considerably smaller lattice parameter (0. and the zones are coherent with the matrix. however. Nevertheless. such as Cu atoms to the platelike G–P zones in fcc Al.1 G–P(1) zone in Al–Cu crystal aligned along a 100. the G–P(1) zones of Fig. and has a diameter of approximately 8 nm. since the local structure around the Cu atoms is not optimal (and cannot be so owing to the constraint of the surrounding Al crystal).361 nm instead of 0. but a significant rearrangement of atoms is still necessary for the transformation to the equilibrium phase. G–P zones do not form in alloys cooled slowly from high temperature. At the zone. so the Cu atoms are not positioned comfortably with respect to nearby Cu atoms. for a “vacancy pump” mechanism that makes the vacancies effective for longer times than expected by their random walk. t Fig. is about one atom in thickness. whose equilibrium concentration increases with temperature as exp(−Ef /kB T). or at a slightly higher temperature. To form G–P zones. The growth of G–P zones usually involves an anomalously fast transport of atoms. so the diffusion distances must be longer than expected from estimates of √ x D(T) t. however. The idea is that the vacancies are partly bound to the solute atoms. 11. since this results in larger precipitates of the equilibrium phase. This is a source of positive electronic energy. Besides preserving the high-temperature solid solution. the G–P zone.405 nm for Al). and the process repeats. There is evidence. or to vacancy sinks like dislocations. where D(T) is the diffusion coefficient at the temperature T. the excess vacancies provide a burst of diffusivity that allows rapid G–P zone formation. which is seen edge-on. Figure 11.269 11. Diffusion processes are enhanced by a few orders of magnitude by a rapid quench. and help the solute atoms move to a growing G–P zone. the vacancy becomes free.

Chang and J. direction.M. 17:32:08 . half the lattice parameter of fcc Al. Image courtesy of Y.202 nm. Howe. The separation between neighboring white (or black) dots is 0.C.

a.270 Transformations involving precipitates and interfaces There is also some strain in the surrounding matrix of fcc Al. 17:32:08 . so this is not the full explanation.2. δ is typically several percent of the interatomic distance. however. Over the surfaces of metals. 11. the electron density tends to smooth out. The figure depicts both “relaxations” vertically.1 Surface reconstruction Figure 11. so the relaxations depend on the crystallographic structure and are generally 3a Vacuum a–δ t Fig. so the nuclei in the top layer move towards the atoms beneath. These “coherency strains. relaxation usually causes the top layer of atoms to be a bit closer to the layer below by the amount δ. It is not so large in Al–Cu as in other alloys. This altered electron density tends to reduce the attractive electrostatic potential for the positive ion cores. although it depends strongly on the crystallographic orientation of the surface.2 shows how arrangements of atoms at free surfaces differ from atom arrangements in the bulk. 11. and “reconstructions” horizontally. Compared with the atomic spacing in the bulk. A higher electron density does not necessarily imply a closer interatomic distance. This coherency strain energy is another positive (unfavorable) contribution to the thermodynamic free energy of zone formation. This is a subtle effect. but this effect diminishes rapidly below the first layer.2 a Bulk Generic surface relaxation (vertical shrinkage) and surface reconstruction (new arrangements of atoms in the plane of the surface). The explanation of this contraction during relaxation is sometimes based on the transfer of electrons from the missing bonds above the surface to bonds between the first and second atomic layers.2 Surface structure and thermodynamics 11. For the first layer.” would vanish if there were a loss of registry between the fcc Al matrix and the Cu atoms in the zone. helping to reduce the electronic kinetic energy. but much less than would occur if the Cu atoms had their natural lattice parameter. and there is a general expectation that it will increase with the square of the difference in atomic radii of the solute and solvent atoms. Owing to the shorter length of the Cu–Al bond than the Al–Al. coherency strains distort the Al matrix by pulling the (200) matrix planes closer to the plane of the G–P zone.

not large. Structure taken from the NIST Surface Structure Database (SSD. Hermann). ledges. and adatom. An issue. it may require a high-temperature annealing before it forms. and roughness Figure 11. Consider the growth dynamics of this surface if atoms are slowly added by physical vapor deposition. and an adatom moves over the surface until it finds an 17:32:08 . such as pairs of ions having opposite charges.2 shows what would be described as a [Po(100)-(3×3)] reconstruction (assuming a similar rearrangement out of the plane of the paper. For semiconductors. Again.4a shows the surface of a simple cubic crystal. and assuming that Po has a simple cubic structure). M. Surface reconstructions involve horizontal (and vertical) repositioning of atoms in the plane of the surface. 11. kinks.2 Surface structure and thermodynamics Adatoms Dimers Rest atoms (1×1) bulk t Fig. kink. as a rule the surfaces are composed of structural units with charge neutrality. vacancy. Atom arrangements in the unreconstructed surface are shown in the lower right. At intermediate temperatures. Watson. Such is the case for the Si (111) reconstruction. but the same concepts apply to an interface between a precipitate and a matrix crystal. Van Hove.2 Terraces. so the energy contribution to the free energy is large. 11. Figure 11. and K.271 11. for example. Today it seems that many complicated reconstructions like these can be predicted by electronic energy calculations that assume zero temperature. which occurs after annealing under vacuum.3.2. 11.A. ledge. Some surface reconstructions are quite intricate. surface diffusion of the adatoms is expected to be relatively rapid. For ionic solids. Visualization with Balsac (by K. however. Although it is stable at low temperatures. sometimes the geometries of the surface reconstructions can be understood by a tendency to eliminate dangling bonds by displacing atoms across atomic-scale valleys to make new bonds that bridge across the valley. is that many surface structures are metastable. and the entropic contributions seem to be small. Such a surface may be expected at a vacuum interface. These bonds break the translational periodicity of the bulk. and defines pictorially a terrace. and may require an activation to achieve equilibrium. such as the [Si(111)-(7×7)] reconstruction shown in Fig. Fritz–Haber Institute Berlin). by P.3 (7×7) The (7×7) reconstruction of a (111) surface of Si. Hermann. the surface reconstruction is driven by the altered electronic bonding characteristics of the atoms at a free surface.R.

but two. If adatoms move rapidly. shows a surface where terraces. this would be zs = 4. 11.” an adatom as “A. energetically favorable site. 11. There will be many such unsatisfied bonds on a rough surface. A kink is even more favorable because the adatom has three neighbors.” (b) Surface with more defects and more roughness. however. Figure 11.5. a thermodynamic analysis can make a prediction about surface roughness. Two vacancies are labeled as “V.4 b a (a) A surface of a simple cubic crystal with two terraces. The process of surface growth need not be so orderly. Especially at lower temperatures there may be kinetic limitations for adatom migration. Each upwards step produces an atom with unsatisfied bonds on its left. Although the surface structure may be dominated by kinetics as suggested here. and implies a down step without bonds to its right.4a.4b. for example. 11. and kinks are so numerous that their individual dynamics may not control the growth process.5 X X X Three possibilities for a surface segment with ledges. Such sites are at ledges. If X X t Fig. ledges. the atoms deposited on the surface will attach to the kinks. and the process begins anew with another terrace. and a ledge as labeled. there may be much more disorder in the surface. 11.” and an atom in the vapor as “G. 11. Unsatisfied bonds of surface atoms are shown in a one-dimensional model in Fig.3 Thermodynamics of surface defects (low temperatures) We first treat surface steps as defects with an energy cost. one kink. where the adatom does not have just one neighbor as it does on a terrace. An unsatisfied bond between an atom and a vacant site has the energy +ε. For vapor deposition at high rates. so smooth surfaces are expected at low temperatures where energy dominates the thermodynamics. When the kink reaches the edge of the crystal.” Atom sites without steps (and those on downwards steps) have atoms to their left. and few on a smooth surface.Transformations involving precipitates and interfaces 272 Terrace Kink Ledge V G V A t Fig. Note that each up step causes a loss of bonds to the left. Eventually the upper terrace moves from left to right across the surface in Fig. Suppose each surface atom has zs possible bonds – for lateral bonds on a (001) surface of a simple cubic crystal. 17:32:08 .2. and calculate their thermodynamic probability in equilibrium. a new ledge forms and it again moves horizontally. and the ledge will grow horizontally one atom at a time. marked with an “X.

The thermodynamics of ledges requires a more interesting adaptation from one to two dimensions.7) where the last expression assumes p  1.5) (11. and there are two possibilities at each step position (up or down). and the ledge runs in and out of the plane of the paper. they also have a higher entropy because all atoms in a ledge are constrained to be neighbors in a line. the equilibrium value of p is found at the minimum of Fs = E − TS. which must be multiplied by N for each upwards step (where N is also the number of atoms along a line in and out of the plane of the paper. but with increasing temperature there is an exponentially increasing number of steps or ledges on the surface.23 was used for the first term of the entropy.6) (11.4) (11. 2 1−p   p zs ε = 2 exp − . here assuming a square surface containing N×N atom sites). Setting the derivative dFs /dp = 0 dFs . The present approach applies better to adatoms. 2 (11.273 11. but this distinction is important for two-dimensional surfaces. The energy cost of unsatisfied bonds is Es = zs ε pN .3 remains the same. The number of ways of arranging pN steps over N surface sites times the number of ways of orienting each step gives   N! pN Ss = kB ln 2 (11. for every up step there must be a down step without bonds to its right. with N denoting the horizontal sites in Fig. 11. 2. 11. the number of up steps. The entropy of the rough surface is determined by p.5. 11.1) where zs = 4 is the in-plane coordination number. The big change is to the energy. p(T) = 2 exp − 2kB T 0= (11. where the familiar result of Eq. At low temperatures the surface is smooth. For a given temperature. dp     zs ε p 0= N + N kB T ln − ln 2 . 1−p 2kB T   zs ε . The up steps and down steps are assumed to be distributed at random over the surface. whereas adatoms are independent and can assume any position on the surface. The problem can therefore be formulated in terms of pN/2.2 Surface structure and thermodynamics the average surface height is zero for the N sites across the surface. which ranges from 0 < p < 1. The 17:32:08 .2) (N − pN)! (pN)!   (11.5 (perhaps with a small correction to account for different angles of the ledge in and out of the paper).3) Ss = −kB N p ln p + (1 − p) ln(1 − p) + kB pN ln 2 . Although adatoms have a higher energy than an atom at a ledge. This one-dimensional treatment did not distinguish between adatoms and ledges. N denotes the number of atoms in a line from left to right in Fig. The expression for configurational entropy Ss in Eq. In two dimensions.

At T = 0 we expect a completely flat surface. There are a number NA = N − NB vacancies on the surface. These may not be the same strength because between any two adjacent positions on the surface there may be a vertical change of more than one atom in height. Now suppose we place some atoms on 1 Incidentally. 2. with the probability c = NB /N of a site being occupied. (11. 10. the interactions between atoms on a rough surface lead to a cooperative transition.6.7.10 can be used to follow the roughness versus temperature. 2 There are also vertical interactions to consider. The “solidon-solid” model assumes that the energy of a step between adjacent surface sites scales as their difference in height.2. because the surface structure is not confined to a plane. 2. squared. where reductions in roughness lead to increasingly favorable bonding. The atoms begin to come together into terraces. High-temperature approach to Tc We start at high temperatures with NB atoms on the N sites of a two-dimensional surface. 17:32:08 . The calculation of the free energy is now identical to that of a two-dimensional solid solution as performed in Sect. we can explain the onset of a thermodynamic “roughening transition” at an intermediate temperature.274 Transformations involving precipitates and interfaces consequence is that a ledge is highly improbable at finite temperature because it has high energy and low configurational entropy.8) where 4V = eAA + eBB − 2eAB .36 Tc = −zs V/kB . denoted here as an A-atom for consistency with notation of Sect. this same argument applies to dislocations. although its configurational entropy is zero. a smoothing of the surface will set in by lateral interactions between the adatoms. and a layer of different chemical species is deposited on top of it at modest temperature. At sufficiently high temperature we expect a random occupancy of these atoms on the surface sites. Low-temperature approach to Tc Approaching the roughening transition temperature from below is a more reliable way of finding Tc . 2.2. the more favorable it is to add an atom to the terrace. the result is the critical temperature of Eq.1 11. For a surface coverage with c = 1/2.2. since this has the minimum energy. This approach is with risk. which are therefore considered to be non- thermodynamic defects in three-dimensional crystals: see Sect. Taking a view of a crystal surface as a solid solution of adatoms and vacancies.4 Roughening transition (intermediate temperature) The previous section showed a thermodynamic tendency for a surface to be rough at high temperatures. 2. At lower temperatures. This approach works best when there is a substrate of immobile atoms. With decreasing temperature. especially at higher temperatures.2. and smooth at low temperatures.2 The other surface species is the vacancy.6. The phase diagram of Fig. and the more ordered is the terrace.

so eAB = eBB = 0 and 4V = eAA .9 is therefore F= L − 4V − T kB ln(zs − 2) . We make the simple assumption that the number of possible directions for the next atom on the edge of the mesa is zs − 2. a (11. being close to zs V per unit length. This gives Tc = − 4V . If this transition occurs below the melting temperature. The A–A bonding is favorable. where a is the interatomic spacing.11) The free energy. so the critical temperature occurs when F of Eq.12) Our comparison is to a free energy for a flat surface with no mesas. 11. so 4V < 0. a (11. kB ln(zs − 2) (11. The energy and entropy depend on L. The total number of ways of arranging the edge of length L around a mesa is  = (zs − 2)L/a . assume zero energy for interaction with the vacant B-atoms. For a simple cubic lattice with zs = 4. 17:32:08 . 11. giving a kink in the edge. The number of such locations is L/a.10) so the configurational entropy is Sconf = kB ln  = kB L ln(zs − 2) .9) but this will be different for other lattices. Assume the energy scales with L. Of the zs candidate sites to place the next atom on the edge. the energy is E=− L 4V . we disallow that site that goes backwards to our previously placed atom and another site is eliminated if we forbid overhangs. we follow the atoms around the perimeter of the mesa and allow a change in direction at each unit length step. 3 To understand the sign.13 shows quantitative difficulties in predicting actual temperatures for the roughening transition.12 is zero.3 For the configurational entropy. but an A-atom along the perimeter has lost an A-neighbor. these atoms will cluster together to form a mesa of one atom in height. To minimize the energy cost of vertical steps.13) A comparison of Eqs. with a perimeter length of L.2 Surface structure and thermodynamics 275 top of a flat surface. The surface roughness will also affect surface properties such as gas adsorption. the roughening transition is expected to have important effects on the surfaces of crystals. with Eq.8 and 11.11. a (11. (11. F = E − TSconf . and L is assumed large because there are few mesas on the surface. 11. but this cooperative transition from a smooth surface at low temperatures to a rough surface at high temperatures is generally expected over a narrow range of temperatures.

Transformations involving precipitates and interfaces 276 11. Atoms can be added in front of the ledge. Because these form starting points at different radial distances from the dislocation core. 2π r (11.2.6. showing spiral pattern.6a marks the ledge in a view from above the surface. the ledge evolves into a spiral. Figure 11.14) Ledge 1 2 3 4 a C A B 1 2 4 3 D b C A B D t Fig. 11. and four locations for atom addition are labeled 1–4. 11.6 c Adding atoms to sites along a surface ledge created by a screw dislocation. 17:32:08 . Dislocation core intersects the surface at the small circle. intersecting a surface as shown at the top of Fig. offers a mechanism for continuous surface growth.6b shows why a spiral pattern is expected. 15. Treating the problem as a continuum without crystallography.6a. to be specific) have been attached to each point 1–4 in Fig.1 Screw dislocation mechanism A screw dislocation. An equal number of atoms (13 atoms. The screw dislocation creates a ledge on the surface that extends from the dislocation core to the edge of the crystal surface.3 Surface structure and kinetics 11. 11. (b) Pattern of adding 13 atoms to the 4 sites at different radius shown in part a. Figure 11. Arrows point perpendicular to the radial line to the dislocation core. the inner radii make more wraps around the dislocation core. This important growth mechanism is depicted in Fig. What is interesting is that if atoms are added at the same rate to all locations along the ledge. the rotation angle decreases with the initial radius θ= Na . (a) Initial position of the ledge. (c) Set of the four endpoints (A–D) of the patterns in part b.3. and four sites in front of the ledge.

Specific surface energies. and Fig. the high curvature causes atoms at the ledge to have a smaller average number of neighboring atoms than the atoms along the ledge at larger radii.2 Layer or island growth The fabrication of thin film devices by methods of atom vapor deposition often requires attention to the roughness of interfaces between layers of different materials. 4 We assume there is no large change of radius for the atoms added to the spiral around the angle θ .6b suggests a physical picture of adding rows of atoms as strings of equal length from the initial position. 11. (b) island growth (VW). The thermal issues described in Sect. 11.6c shows how the terminations of these strings are consistent with a spiral.2. (Equivalently.11. 5 If you like the string analogy. The figure shows the three common types of thin-film growth: σBV FV σAB a VW σBV σBV σAB σAV b SK t Fig. and the material was cooled to low temperatures where atomic mobility was suppressed. The problem is more complicated when B-atoms are deposited onto a substrate of A-atoms.4 Figure 11.7 σBV σAB c Three common growth modes for thin films of B-atoms on an A-atom substrate. consider the structure to be formed with shorter strings at smaller radii. 11.3. and a is the interatomic separation. A feature of a spiral structure is that at small radii.5 Nevertheless. 11. B-atoms and vacuum. (a) Layer-by-layer growth (FV). 11. with initial surface indicated by a dashed line. are indicated between A-atoms. σXY . spiral structures are often found at surfaces.7 for a flat substrate surface indicated by the dashed line.4 can be relevant. 11. there is a higher density of corner atoms for the structure evolving from point 1 on the left of Fig. (c) Stranski–Krastanov growth (SK). At small radii. They are usually not in thermodynamic equilibrium. These structures were often formed by growth at high temperatures. the ledges are expected to have larger separations. but these considerations were for free surfaces of one chemical element. so this energy penalty should slow the growth rate of the inner parts of the growth spiral.3 Surface structure and kinetics 277 where N is the number of atoms added to each atom position at the initial radius r.6b than from point 4 on the right.) This is more costly energetically. sometimes with ledges at regular radial separations of tens of microns. 17:32:08 . Such is the case shown in Fig.

11. The difference in surface energy for FV and VW growth comes from the fraction 1 − f of surface that is not covered by islands in Fig. Some “quantum dot” structures have been fabricated with the help of elastic interactions through the substrate.16) On the other hand. • “island” or Volmer–Weber (VW) growth in Fig. As the deposited layer becomes thicker and more robust. there is necessarily an interface between the precipitate and the parent phase with unlike-atom pairs. 11. When σAB + σBV σAV .8. • Stranski–Krastanov (SK) growth in Fig. Another consequence of elastic strains may be to cause the deposited atoms to arrange as islands. Island growth is promoted by a high specific surface energy of the A/B interface. 11. (11. When a precipitate of different composition nucleates in a parent phase. Here a chemical contribution to the interface energy is estimated with a simple picture of two solid solutions in contact as shown in Fig. since the lattice parameters and crystal structures of the precipitate and parent phase are usually different. which differs from the average of the bond energy across 17:32:08 . however. and the substrate may warp. 11.278 Transformations involving precipitates and interfaces • “layer-by-layer” or Frank–Van der Merwe (FV) growth in Fig. The real situation is more complicated than this. 11. The plane of the interface has the average bond energy Eαβ . the substrate may accommodate an increasing fraction of the strain. It should be added that elastic strains usually accompany thin film growth. The critical condition for island growth is EFV > EVW . owing to lattice mismatch between the substrate surface and the B-atom layers. 11. The details of the bonding across the interface are complicated. and peeling of the film is common for film thicknesses of 1 micron or more. σAB . We ignored the vertical sides of the islands.7a and 11.7b.15) (11.4 Chemical energy of a precipitate interface Chemical unmixing occurs to maximize the number of like-atom neighbors. the sides may have a different surface energy owing to anisotropy.7c.7a. layer-by-layer FV growth is expected. often in a semiregular array on the surface. relative to the specific surface energies of the vacuum interfaces. since we considered only a two-dimensional structure. The strains can be large.7b shows that those regions of surface covered by islands in Fig. which add energy cost.7b. the strains of mismatch are taken up by the thin layer of B-atoms. with the assumption of a surface coordination number zs for atoms at the interface.7a. Furthermore. σAB + σBV > σAV . In the early stages growth. 11. as does the FV growth in Fig. 11. A comparison of Figs.7b have equal projected areas of A–B and B–V interfaces. when σAB + σBV < σAV . A bond-counting argument is used. which starts with layers and later develops islands on top of them. These elastic strains can cause a thin film to break free of the substrate. Stranski–Krastanov growth often occurs. 11.

equivalent planes in the α-phase and β-phase by γch = Eαβ − Eαα + Eββ . 2 (11.17) where γch is the chemical energy per atom at the interface. Counting bonds across equivalent planes in the α-phase and β-phase amounts to figuring out the number of like and unlike pairs of atoms in solid solutions having the concentrations of B-atoms cα and cβ zs . 11.4 Chemical energy of a precipitate interface 279 A-rich t Fig.11.8 α -phase α = (l – c ) pA A A pα B = cA B β = (l – c ) pA B B-rich pβ B = cB β -phase Interface (dashed line) between two phases of different chemical composition.

18) eAA (1 − cα )2 + eBB c2α + 2eAB cα (1 − cα ) . 2 . Eαα = (11.

2. zs Eββ = (11. each atom sees atoms in the other solid solution.26. On the other hand. which has a different composition zs . across the α/β interface shown as the dashed line in Fig. The bond counting is in fact the same as for solid solutions in Eq. 11.19) eAA (1 − cβ )2 + eBB c2β + 2eAB cβ (1 − cβ ) . 2 where all pairs of atoms in the bonds are in the same phase α or β.8.

so γch > 0.17 for counting between two sublattices of an ordered alloy. 11. and cα cβ 2 zs cα − cβ 4V . 2. this interface energy γch is expected to be an overestimate because the atom arrangements at the interface are not so regular as to allow the full counting as on an Ising lattice.20 into Eq. as expected. cβ . (All bonds across the interface begin and end in different phases. Eαβ = eAA (1 − cα )(1 − cβ ) + eBB cα cβ 2 +eAB cα (1 − cβ ) + eAB cβ (1 − cα ) . and grouping terms in powers of cα .17.22) γch = − where again For an unmixing alloy 4V < 0.21) 4V = eAA + eBB − 2eAB . The chemical energy per atom at the interface makes a positive contribution to the interface energy.) Substituting Eqs. (11. (11. 17:32:08 .20) where the counting is similar to that shown in Fig. In practice.18–11. 11. 2 (11.

” which discourages an abrupt change in chemical composition. as would be expected if they were to minimize their surface energy in a cubic matrix. 16. with permission from Elsevier. 6. while maintaining a crystallographic orientation relationship with the Al matrix.5 Elastic energy and shape of growing precipitates Figure 11. Smoothing of the composition profile at the interface reduces the interfacial energy from the sharp composition gradient. the energy of the interface structure (e.5. and wide in two (one wide dimension extends upwards out of these 2D images). plus structural effects on the interfacial energy. Sect.5. Section 12. A more subtle issue is that the chemical composition may not change abruptly at the interface. and Eel tends to promote widening it. 17:32:08 ..Transformations involving precipitates and interfaces 280 This counting of bonds implies that the energy of chemical bonding is independent of. and additive to. cause significant departures of the interface energy from the γch obtained from a simple bond-counting model. the surface energy Es promotes a thickening of the precipitate. Reprinted from [115]. 11. 11.9 shows images of plate-like precipitates seen edge-on.1). but the situation a b 100 nm d c t Fig. Approximately.2 discusses the “square gradient energy.g.9 100 nm 100 nm 100 nm Transmission electron microscopy images showing plate-like precipitates in a Al–Cu–Mg alloy annealed for increasing times at the same temperature. Some issues of preferred shapes for precipitates were mentioned in Sect. The plates widen and thicken as they grow. These additional chemical effects. In practice this is not so simple because different atom configurations prefer different local structures. Another physical effect is a chemical segregation to the interface – the different local structure of the interface can favor a different chemical composition. It is obvious that these precipitates are not shaped at all as spheres. They are thin in one dimension. Elastic energy plays a bigger role in determining their shape than surface energy.

Etot = Cπ l d2 32 (11.5 Elastic energy and shape of growing precipitates 281 evolves during a precipitation phase transformation. (b) Redrawn precipitate from panel a with features labeled.11. without the parent material around it to provide elastic constraint. 11. their relative importance changes with the volume of the precipitate particle. We seek the thickness of the precipitate versus the volume of the precipitate. 17:32:08 . Etot = Eel + Es . 2 (11. 11. a situation like Fig. where C includes elastic moduli and geometrical factors. (11. Eq.) For simplicity. and is multiplied by C (d3 )2 . so we √ replace l with (4v)/(πd) $ 4v 3/2 2 4v Etot = Cπ d 3 + 2σs .23) ij0 = ⎣0 1 0⎦ . v = d π (l/2)2 . if 3 = 1 . so the only significant transformation strain is the fractional change in length 3 along the z-axis.24) + 2π σs (l/2) . thereby minimizing the elastic energy that goes as (d 3 )2 (times geometrical factors). 0 0 3 so the strain 1 is equal along xˆ and yˆ . Without surface energy. where the size of the precipitate d is minimized in the zˆ-direction of the large mismatch. assume that 1 ∼ 0. circular surfaces.26) π πd d Elastic field t Fig. 11. which increases rapidly with the distortion caused by the precipitate. An important point evident from the geometry of Fig.10 is that the elastic distortion is largest around the edge of the precipitate.10 is expected. Although both Eel and Es are positive. 11.10 a l b (a) Preferred shape of a tetragonal transformation product if good lattice matching is along horizontal and out-of-plane directions as described in text. πl. where l is the long direction in the plane of the plate. can first promote widening. The stress-free transformation strain6 for the z-variant of a tetragonal transformation is ⎡ ⎤ 1 0 0 (11. 6 The transformation strain describes the change in shape of a volume of parent material when transformed. (Incidentally.25) where the surface energy is assumed to come from two flat. Consider a relatively simple case where the precipitate is tetragonal and the parent phase is cubic. The elastic energy also scales with the length of circumference. A “stress-free” transformation occurs in free space.23 would describe a cubic-to-cubic transformation. This region has a volume that scales as the circumference. but the strain along zˆ is different. This section shows how Eel . then promote thickening during a precipitation transformation.

the amount of solute passing through any surface scales with the surface area. and the aspect ratio decreases as the precipitate grows. One mechanism for the diffusional growth of precipitates at grain boundaries is the “collector plate mechanism.” It can be understood by considering fluxes of atoms through surfaces of different areas. elastic energy. Consequently the length l must increase as v2/5 because v ∝ d l2 . Notice that Eq. for example. Assuming the composition is uniform. It is appropriate for computer simulations of precipitation transformations. A grain boundary allotriomorph. d √ dEtot 8σs v . To grow a precipitate that has nucleated homogeneously in the middle of a crystal. Of course. v. = 0 = 3C π vd 32 − dd π d2   8σs 2 v d5 = . 4.1 Collector plate mechanism In alloys.5 (an allotriomorph is a crystal of abnormal shape). 3C π 3 34 (11.9 grow more in width than in thickness. 11.6. d/l. A closely related concept is that the presence of a second precipitate in the matrix will alter the strain field around the first precipitate. There are many hundreds of times more Pb atoms in the allotriomorph than in the spherical precipitate. the relatively high elastic energy causes the precipitate to change its shape as it grows larger. indeed. It seems that the precipitates in Fig. 17:32:08 .282 Transformations involving precipitates and interfaces For a given volume. The full competition between surface energy. For platelike precipitates. therefore changes as v−1/5 . was shown in Fig. a solute flux must pass through a surface around the precipitate. which appears with the elastic energy in the ratio of σs /(C 32 ). The aspect ratio of the precipitate.28) so the thickness d increases as v1/5 .28 predicts more thickening for larger surface energy. the total energy will be a minimum for the optimal thickness. thickening somewhat more slowly than it widens.6 Precipitation at grain boundaries and defects 11. rich in Pb. and the material is in steady state. 11. This figure also shows a small spherical Pb precipitate within one of the Al grains. second-phase precipitates often form at “grain boundaries” (interfaces between crystals in polycrystalline materials). the diffusion process also sets a distance between precipitates because they need to collect solute from the matrix.27) (11. Often the mutual arrangements of precipitates are determined in part by how they can configure themselves to minimize the total elastic energy. and a second precipitate may not form in a location that has been depleted of solute during the formation of the first precipitate. and diffusion is complicated. This surface area is small when the precipitate is small. This size difference occurs because the growth rate of the grain boundary allotriomorph is hundreds of times faster. 11.

There are a number of processes that could influence these kinetics. and sometimes estimates of the ratio of grain boundary diffusivity to bulk diffusivity are possible from experimental measurements of the growth rate of grain boundary allotriomorphs. and is transported instantaneously to a precipitate situated on a grain boundary.6. 3.11 11. and the figure is two-dimensional.11a.7 Perhaps curvature alters the growth rate of small particles as discussed in Sect. (b) The rapid diffusion of atoms along grain boundaries gives a much greater growth rate of the grain boundary allotriomorph than the isolated precipitate. In practice. this factor can be hundreds or thousands. however.2 Precipitate-free zone Figure 11.9. the growth rate of grain boundary allotriomorphs seems to be more typically t1/4 . The enhancement factor is much larger in three-dimensional processes. 11. diffusion-controlled processes predict a growth rate of precipitate volume in proportion to t1/2 (t is time). 11. versus only six into the homogeneous precipitate. For example. Quantitative predictions probably require more than the simple theory that is described here. The growth rate of a precipitate located at the grain boundary is therefore enhanced by a factor equal to the ratio of the net surface area of the two grains around it to area of the precipitate. 11.11. Near the grain boundary. moving the time dependence towards t1/3 . To understand the collector plate mechanism. as is typical of diffusion over a fixed distance.12 shows a microstructure that could result from chemical unmixing with the collector plate mechanism. there are about 150 flux arrows into the grain boundary of Fig. and the sizes and separations of the precipitates are not affected by the grain boundary. It is possible to improve on the optimistic assumption of an infinite diffusivity for atoms inside a grain boundary. Nevertheless.6 Precipitation at grain boundaries and defects a b (a) Small arrows indicate fluxes into areas of grain boundary and into areas of precipitates. 11. The idea is that the total flux of solute atoms through a surface near a grain boundary is large because the grain boundary is large. homogeneous nucleation occurred efficiently.41) is used for the diffusional transport into the grain boundary when the grain boundary width is large compared with the diffusion length. such as a nonequilibrium concentration of vacancies or other transient phenomena. 11.283 t Fig. but the basic principle of the collector plate mechanism is well accepted. the problem seems to be a bit more complex than such a simple adaptation allows. However. the nucleation and growth of the allotriomorphs consumes 7 Typically a one-dimensional error function solution (Eq. 17:32:08 . In particular. assume that after solute reaches the grain boundary it has infinite diffusivity. The collector plate mechanism is illustrated with Fig. Far away from the grain boundary. Notice the “precipitate-free zone” around the grain boundary.

) 11.Transformations involving precipitates and interfaces 284 t Fig.12 a 200 μm 10 μm b Precipitate-free zone near grain boundaries. or both. and undergoing a coarsening process as described in Sect. From [116]. The chemical potential for solute atoms in the large allotriomorphs may be lower than for the small precipitates. other internal interfaces such as antiphase domain boundaries or stacking faults can also serve as sites for heterogeneous nucleation. If the atoms are not well bonded across the surface between the allotriomorphs and the crystal interiors. This region is depleted in solute. Note the precipitate-free zone next to the grain boundary. To some extent.” (A process of “tearing along the perforated line. considerable solute that is transported by the collector plate mechanism. thermodynamics. this can be understood as a rescaling of distances. larger distances between the precipitates. and the solute atoms in these large precipitates come from the region near the grain boundaries. Incidentally. where the efficiency of the collector plate mechanism allows for more rapid growth. However. (a) Overall microstructure showing grains of β-Ti. with grain boundary allotriomorphs at the boundaries. 11. a process that can be favored by kinetics. coarse precipitates on grain boundaries can have important consequences for the fracture toughness of polycrystalline materials. showing some similarities to grain boundary precipitation. 17:32:08 . Images are scanning electron microscopy backscatter images of an annealed TiMoZrFe orthopaedic alloy. and homogeneously nucleated precipitates within the grains.9. and a large zone between the allotriomorphs and the precipitates in the interior of the crystals. grain boundary allotriomorphs can be favored by the kinetics of nucleation (heterogeneous nucleation is usually faster than homogeneous). and the elimination of grain boundary energy is favorable.) Not surprisingly. too. there is more complexity to this problem because the precipitates in the interior of the crystals are interchanging atoms by diffusion. the kinetics of diffusion (the grain boundary “collector plate” speeds solute atoms to the precipitate). so the allotriomorphs may grow at the expense of the homogeneous precipitates.” roughly speaking.3 Other sites for heterogeneous nucleation The previous sections discussed the formation of precipitates at grain boundaries. a fracture crack may propagate along grain boundaries in a process called “intergranular fracture. (b) higher magnification image showing the intragranular primary α-precipitates and fine-scale secondary α forming as a result of the ageing treatment. so precipitates are not found in this region. 11.6. For example. and thermodynamics (the chemical potential is more favorable for atoms added to a large allotriomorph.

8 and 15. The elastic strain field around a precipitate can alter the tendency to form similar precipitates in its vicinity. shown in Fig. θ-carbide).1 The iron–carbon phase diagram Much of ferrous metallurgy is centered around the eutectoid point in the Fe–C phase diagram at 732◦ C and 0. which is several atoms in diameter. 15. The effect can be positive or negative. δ (bcc). On the other hand. and may move to where the field is strongest near the dislocation line. and cementite (Fe3 C. 11.77 wt. the dislocation. Both the interface energies and the strain fields around precipitates can play a role in the kinetics or thermodynamics of forming secondary phases. The second important feature of a dislocation is its strain field. but martensitic transformations are not precipitation processes. but the cause and effect may not be obvious. α. for example. the new phases often form at interfaces between pre-existing precipitates and the matrix.7 The eutectoid reaction and ferrous metallurgy 11. γ . This ambient pressure T–c phase diagram shows five phases: liquid.7 The eutectoid reaction and ferrous metallurgy The one-dimensional defect in crystals. are essential for understanding the heat treatments of steels.7. 17:32:08 .% C. A misfitting precipitate can make it less likely for another precipitate to form nearby because the elastic energy penalty will increase as the square of the distortion. a small substitutional solute atom may be drawn into the compressive field of the dislocation. α (bcc).) 11. Precipitates can themselves serve as sites for the heterogeneous nucleation of new phases.” The thermodynamics of precipitation can also be favored at a dislocation because a precipitate crystal has a more favorable atom configuration than the dislocation core it replaces.9 is an example of this. or perhaps the dislocation was the nucleation site for the precipitate. The strain fields can attract or repel solute atoms having specific volumes different from matrix atoms. dislocations are often found around precipitates. For example. has two salient features. The size mismatch of a precipitate in the matrix may be responsible for generating the dislocation.” in that the formation of one precipitate makes it energetically more favorable to form another.13. Not surprisingly. like that of a grain boundary. a precipitate that creates shear stresses around it may make the local environment favorable for forming another precipitate with a counteracting shear stress. The three of these at the eutectoid point.285 11. “Edge dislocations” generate both compressive and tensile stresses in the matrix around them. (The burst formation of martensite plates in Figs. can lead to rapid solute diffusion down the dislocation line in a process called “pipe diffusion. The lower density of the core. and θ . γ (fcc). If there is a sequence of phases that form during a process of interdiffusion or temperature variations. One is the poor atomic coordination in its core. The formation of precipitates can therefore be “autocatalytic. There is also a metastable tetragonal phase α  (bct).

lower bainite. hcp -Fe phase is formed from α-Fe β-Fe was once defined as nonmagnetic bcc iron at a temperature above the Curie temperature. configured in different shapes or “morphologies.7.% C or 6. 11.Transformations involving precipitates and interfaces 286 δ + liquid Temperature [°C] 1600 δ Liquid 1400 α+γ γ + liquid δ+γ 1200 Austenite 1000 γ Fe3C + Liquid θ cementite Fe3C γ + Fe3C 800 Ferrite α t Fig.13 600 α + Fe3C 400 (α + θ) 0 2 4 Weight percent carbon 6 Iron-rich side of the iron–carbon phase diagram.” resulting from different processes of the phase transformations.9 We consider first the pearlite transformation.14 is a micrograph of pearlite. with α-Fe and graphite being the stable phases.7 wt. which contain mineral phases in different arrangements depending on their conditions of formation. Pearlite is prettier than upper bainite. cementite is also metastable. there has been an accumulation of names for these phases: ferrite = α = bcc austenite = γ = fcc martensite = α  = bct cementite = θ = θ-carbide = Fe3 C at pressures above 12 GPa. Since cementite is so robust. which is a supersaturation of up to 1 percent of C in α-like Fe. They are combinations of α plus θ phases.% C. pearlite. such as ledeburite. 17:32:08 . The spatial scale of the wave patterning is approximately the wavelength of optical light. and 8 Incidentally. without consideration of graphite. it is appropriate to place cementite on the phase diagram at 25 at.8 Over the years.2 Pearlite – a eutectoid reaction This section describes pearlite. 9 There is an apt correspondence to different types of rocks. upper bainite. A more compelling reason to describe the pearlite reaction is that it is a typical eutectoid reaction. and graphite formation suppressed. where a parent phase (a liquid for a eutectic reaction. More baffling are the names for steel microstructures. 11. named for its iridescent pearly appearance when it forms with fine structures. for example. spherodite. Figure 11. Weight (or mass) percent is traditional for the composition axis. These microstructures are not phases.

15a depicts the heterogeneity of the carbon in the austenite in front of the growing pearlite colony.30) where L is the latent heat of the phase transformation (and all factors are positive). Working with a cubic centimeter of material.29 and setting G = 0 gives the minimum value of λmin . now including interfaces between the ferrite and cementite. 17:32:08 . the bulk free energy difference GV is expected to scale linearly with undercooling T. and λ in cm. and collided during growth. 11. austenite in the case of steel) transforms into two different phases that grow together in a diffusion-controlled process.2.13 shows that austenite can dissolve up to 2 wt. 4. surrounded by dark regions of α-Fe. the favorable GV overcomes the surface energy.7 The eutectoid reaction and ferrous metallurgy 5μm Pearlite colonies.% carbon. so there is an optimal λ for a given temperature of the pearlite reaction.31) Figure 11. 11.” were formed separately. We first find a minimum value of λ. For λ > λmin . σαθ as energy per cm2 . and shown schematically in Fig. Tc (11.6 or Eq. 11.15a).15a. G = GV + 2 σαθ . GV = −L T .3. 4. 11. Substituting into Eq.29) where the factor of 2 accounts for two interfaces in the distance λ (see Fig. Separate groups of wave patterns in Fig. however.287 t Fig.14. Image by Michel Shock (public domain). A smaller λ creates more surface energy σαθ .14. called “colonies” or “cells. with GV = Gγ − Gα+θ as energy per cm3 . 11. as for Eq. 11. L T (11. The ferrite takes iron from the austenite but takes little carbon. seen as thin strips in cross-section in Fig.14 11. whereas the ferrite contains very little carbon. λ (11. Figure 11. Within the colonies are thin lamellae of ferrite and cementite phases.11. As for Eq. For this first-order phase transition. balancing the surface and volume energies as in Sect. 4. and the diffusion of carbon encourages the lamellae to maintain a close spacing periodicity λ along the growth front shown in Fig. we seek the free energy G of transforming a volume of austenite into ferrite plus cementite. favoring the pearlite reaction λmin = 2σαθ Tc . Image is a scanning electron micrograph of an etched specimen of steel. The carbon needs to redistribute spatially during the pearlite reaction γ → α + θ . 1. 11.15a.

(b) Free energy versus composition curves for steel undercooled below the eutectoid temperature. Note that the opposite is true for the β-phase in Fig. Sect. 11. as is the case for the cementite phase in pearlite.15b are for a generic eutectoid transformation below the eutectoid temperature. the common tangent line drawn in Fig.6 0.33) (11. in this case the γ -phase (austenite).15b. 11. The free energy versus composition curves (cf. For local equilibrium at the interface between the α. As the pearlite colony grows to the right in Fig.8) of Fig. showing depletion of carbon in front of the cementite and excess carbon in front of the ferrite phase.0 a 0. whereas the cementite needs more carbon that the average composition of the austenite. λ λ   2σαθ c −LT + . The larger the undercooling.34) (11. 11.8 1. 11. We estimate the concentration gradient with λ/2 for distance.0 (a) Growth front of pearlite in steel.15b). so 2 c G λ  c 2σαθ v = −κD GV + ..2 0.15a J = −D 2 c .15 0. and a c from a condition of local equilibrium at the interfaces of the austentite and the growing ferrite and cementite. is proportional to the carbon flux J (but in the opposite direction). carbon rejected in front of the ferrite diffuses down a concentration gradient to the carbon deficit in front of the cementite. It is also assumed that the velocity of growth is proportional to the free energy of the reaction G.15a – the austenite in front of the growing pearlite has an excess of carbon.15a. 11.4 Concentration (c) b 0.15b shows a feature that is consistent with the picture in Fig. v.Transformations involving precipitates and interfaces 288 ferrite ferrite Free energy cementite carbon λ cementite Δc Δc Fγ Fα ferrite F⎯ austentite cementite t Fig. has a high free energy and is unstable. λ (11. the larger the value of c for the difference in carbon concentration in front of the ferrite and cementite (see the construction for c in Fig. 11.35) . v = −κD λ Tc λ v = −κ  D 17:32:08 (11. 11. 2.32) The velocity of growth. 11. This difference gives the concentration gradient c 2/λ that drives diffusion of carbon in Fig.and γ -phases. The parent phase.

To obtain the lamellar periodicity with the highest growth velocity.40 is appropriate for a specific c. and shorter diffusion distances. Temperatures well above 900 ◦ C are usually required for good diffusional homogenization.7 The eutectoid reaction and ferrous metallurgy making use of Eqs. The pearlite microstructure grows faster and finer with increased undercooling. The larger undercooling gives a larger GV . at least compared with room temperature.39) vm = κD 4σαθ Tc Tc 4σαθ Tc   1 c LT 2 . a smaller lamellar spacing λ causes the free energy curves Fα and Fβ in Fig. 11. The concentration gradient is decreased proportionately.16). The optimal lamellar spacing turns out to be fairly close to λmin . indicating the dominant role of carbon diffusion in the pearlite transformation. The kinetics of diffusion favor a small value of λmax to minimize diffusion distances. At these temperatures. We also neglected details of the concentration field. 11. set the first derivative equal to zero   4σαθ LT dv − 3 = 0 = κDc . λmax . Nevertheless.35 gives a maximum growth velocity   c LT −LT 2σαθ LT + .3 Heat treating of steel Before forming a pearlitic microstructure. and blocks of steel can be forged into useful shapes when in the austenite phase (see Fig. 11. 11. and to help erase compositional inhomogeneities of other solutes.40. the present analysis does show the decrease in lamellar spacing with undercooling.36) dλ Tc λ2max λmax 4σαθ Tc .15b to be raised with respect to Fγ . T.37 into Eq. which allows for more surface energy. 17:32:08 . the specific c depends on λ. (11. 11.289 11. at least until the diffusion coefficient D for carbon in austenite becomes small. and absorbing the 2 and other geometrical factors in the constant κ. closer lamellar spacing. so c is decreased. 11. and the diffusion dynamics. Recall that λmin is set by the thermodynamic condition of equality of surface energy and bulk free energy.37) λmax = LT (11. carbon steels are usually given an “austenitizing” heat treatment to transform them into a single-phase solid solution of C in fcc Fe. curvatures of the interfaces between austenite and ferrite and between austenite and cementite. as is the growth velocity of Eq. At a fixed temperature. There are a number of other issues that were not treated carefully in the present analysis. 11. Substituting λmax from Eq. the steel is relatively easy to deform.40) vm = κD 8 σαθ Tc The growth velocity depends strongly on the undercooling.29 and 11.7.30. 11. (11.38) λmax = 2λmin . (11. (11. Although Eq. and the increase in growth velocity with undercooling.

This “proeutectoid” ferrite usually nucleates and grows along the grain boundaries of the austenite. Below 732 ◦ C. Image reproduced under license from Aloysius Patrimonio. but some have more (hypereutectoid steels). contains nonequilibrium carbide phases such as “-carbide” inside the ferrite crystals.com. if a steel is cooled rapidly to a low 17:32:08 . but is still a mixture of ferrite and cementite. which forms from approximately 350 to 550 ◦ C. a key microstructural feature is the grain size (size of austenite crystals). In both cases the steel is austenite. slowing the pearlite reaction. Finally.16 a b (a) Iconic image of a blacksmith forging a steel bar. For hypoeutectoid steels. these cementite crystals can serve as nucleation sites for the pearlite microstructure. cooling from the austenite single-phase region passes through a two-phase region of ferrite plus austenite. these ferrite crystals can serve as nuclei for the pearlite reaction. so if the ferrite finds a preferential crystallographic orientation with one austenite crystal.Transformations involving precipitates and interfaces 290 t Fig. and a consequence is a change in reaction mechanism.77 wt. Lower bainite. it is probably less favorably oriented with respect to the grain on the other side of the boundary. glowing with yellow blackbody radiation. reaching a maximum rate at a temperature of about 550 ◦ C. http://retroclipartz.% C (hypoeutectoid steel). At lower temperatures. so some ferrite forms during cooling. When the steel is cooled below 732 ◦ C. the free energy for the decomposition of austenite into ferrite plus carbide continues to increase with undercooling.40 should be considered D(T)). Lower bainite is sometimes distinguished from upper bainite by its feathery appearance in an optical microscope. Most steels have carbon concentrations less than 0. Muroran Japan (human in lower right). 11. 11. the diffusion of carbon becomes slow (the D in Eq. which forms at lower temperatures. Upper bainite. During austenitizing. which may be kept small by the addition of a percent of Mn to the alloy. In this case it is the cementite that nucleates on the austenite grain boundaries. (b) Large forging press at Japan Steel Works. The pearlite reaction speeds up with undercooling below 732 ◦ C. Different microstructures are formed in steels cooled to lower temperatures. Typically a ferrite crystal grows more quickly into the austenite crystal with which it has a poor orientation compatibility. has a more angular morphology than pearlite. Cooling a hypereutectoid steel from the austenite region passes through a two-phase region of austentite plus cementite. and tends to grow into the austenite grain with which the cementite does not have a good surface orientation. for example. Few austenite grain boundaries have special orientations. Nevertheless.

the extended volume concept allows the β-phase to grow and nucleate in previously transformed material as shown in the bottom diagram. At time t3 .49. and a steady-state nucleation rate was obtained as Eq. Johnson and Mehl [121]. The interplay of solute depletion. 11. 4. especially in multicomponent alloys. and Avrami [122]–[124]. Reprinted from [125]. 11. Figure 11. β-particles begin to impinge on one another.17 t2 t3 Evolution of the microstructure of a single-phase α transforming to the single-phase β at three times t1 < t2 < t3 . the matrix α-phase is depleted in solute. nucleation. as more and more β-nuclei form and grow. where t1 < t2 < t3 . 4. and is one reason why steel has been so important in the industrial age. and remains a topic of research [117]–[119]. For the formation of β-phase after the quench of Fig. the austentite transforms to martensite in a diffusionless process involving shear.1. but the carbon imparts a tetragonality to the bcc unit cell (so it is bct and designated α  ). altering the timedependence of the fraction of new β-phase. this rate is no longer accurate for two reasons. At longer times. and is explained here.3.291 11. The activation free energy of nucleation and the nucleation rate depend strongly on supersaturation. Martensitic transformations are described in Sect. Martensite is approximately described as a supersaturated solid solution of carbon in α-iron. 17:32:08 . This overlap problem is handled by a clever method devised by Kolmogorov [120]. Martensite is extremely hard mechanically.8 The Kolmogorov–Johnson–Mehl–Avrami growth equation temperature to suppress carbide formation. and growth is complicated. At t1 the density of nuclei is t1 t Fig. so the nucleation rate J decreases with time.17 depicts the microstructure of a nucleation and growth phase transformation at three times. The particles stop growing where they touch. 15. Even the nucleation of a new phase in a single-component system (which does not require long-range diffusion) is challenging at intermediate times when the new particles grow to impingement.8 The Kolmogorov–Johnson–Mehl–Avrami growth equation Nucleation was considered in detail in Chapter 4.

a particle that nucleated at the earlier time t has the radius v(t − t ) and the volume 4π/3 [v(t − t )]3 . β β only a part of the change in Vex (t) can contribute to the real V β (t). it proves valuable for calculating the fraction of β-phase at later times.) This remaining fraction of untransformed volume is   V β (t) fun (t) = 1 − .41) 0 where V is the total volume of the system. We assume that all spherical particles grow with the radial velocity v.45) This is the Kolmogorov–Johnson–Mehl–Avrami (KJMA) equation. = 1 − exp − V 3 0 (11. and each β-particle is approximately the critical size at this early time. The essential trick to handle impingement is to assume that in an increment of time. that this Vex (t) grows to arbitrary size at large times. so JV = Nδ(t = 0). The integration over t accounts for the nucleation times of all particles. further nucleation gives an increased density of β-phase particles. which is constant at all times. 11. (11. The fraction of dVex (t) β that contributes to dV (t) is the fraction of untransformed phase that remains at time t. but has a sigmoidal shape with increasing slope that later gives way to an asymptote at long time with V β /V = 1. The KJMA Eq. too.41. Without impingement of particles.42) V and the fraction of new extended volume that contributes to the new real volume of β-phase is   Vβ β dVex (t) . Eq.44) Using Eq. V (11. The actual shape of V β /V is typically calculated for two cases. (All β the rest of Vex (t) are in regions that have already transformed. 11. For determining the volume of β-phase. At time t.45 predicts that the volume fraction at t = 0 is zero.41 is valid only at early times before particle impingement. and the nuclei that formed at earlier times have grown larger. Note. however. dV β (t) = 1 − V (11. the actual volume fraction of β-phase is    4π t 3 V β (t)  3  J v (t − t ) dt . At t2 . Nevertheless. the total volume of β-phase is β Vex (t) = V 4π 3  t J v3 (t − t )3 dt . The first case assumes that all nucleation occurs simultaneously. (11.43) By integration β Vex (t)   V β (t) = −V ln 1 − . where N is 17:32:08 . and for this it is called the “extended volume” β (and hence the subscript “ex”). whereas the actual V β (t) cannot exceed V. 11. the volume of the alloy.Transformations involving precipitates and interfaces 292 predicted by the nucleation rate J.

11. Integration of Eq. The Dirac delta function sets t = 0 and selects the integrand as   V β (t) 4π 3 3 = 1 − exp − v Nt . Jss .45 gives a more rapid dependence on time .293 11.9 Coarsening the number of nucleation sites per unit volume.46) V 3 The second case assumes the nucleation rate. is a constant for all times. (11.

and separated by the distance R. The smallest precipitates have the highest surface energy per atom and these will disappear. where At = 4π (r12 + r22 ) . 3 1 (11. This process is called “coarsening” of the precipitate microstructure. Avrami suggested that a plot of ln[ln(1 − V β /V)] vs. with radii r1 and r2 . Conservation of solute requires that if some β-phase precipitates grow. The precipitates are assumed to be spherical.and B-atoms are approximately unmixed into their expected concentrations. r1 < r2 ) to precipitate 2 will decrease continuously the total surface area of the precipitates. and the exponent of t will indicate the dominance of homogeneous or heterogeneous nucleation. Because the unmixing is complete. This section estimates the time dependence of the average precipitate radius. Equation 11. so all A. π V β (t) = 1 − exp − v3 Jss t4 . We formulate a simple problem with two β-phase precipitates in an α-phase matrix.46.46 and 11.49) . Avrami assumed the heterogeneous sites were distributed randomly throughout the bulk material. 11. others must shrink. the total surface area At is not conserved. 17:32:08 (11. 11. Since the above two examples are limiting cases of nucleation behavior. Consistent with Eq. t will result in a straight line with a slope between 3 and 4. the total volume of precipitates Vt is a constant with time Vt = 4π 3 (r + r23 ) . The essential picture and variables are shown in Fig. 11.18.47) Notice the change in exponent from t3 to t4 between Eqs. It is nearly in equilibrium. 11.9 Coarsening Suppose a diffusional phase transformation is complete.48) On the other hand. This assumption is not good for polycrystalline solids when nucleation tends to occur at grain boundaries or perhaps dislocation lines. Suppose the microstructure has B-rich β-phase precipitates distributed within an A-rich α-phase matrix.46 can be modified to accommodate these processes. Diffusion of B-atoms from precipitate 1 (with smaller radius.47. however [126]. V 3 (11. There remains only surface energy that can be reduced if the β-phase precipitates grow larger.

4. ∇μ  R   = 3σ v 1 − 1 1 .294 Transformations involving precipitates and interfaces m2 < m1 r2 r1 J R t Fig. but the treatment of ζ is of greater concern. the gradient in chemical potential is approximately  = v G1 − G2 . r G (11. shows that a more accurate expression for the chemical potential should be 2σ v/r. Since unmixing is complete.52) (11. surface energy adds to the free energy a G [energy per unit volume] 4π r2 σ Aσ . specifically Eq. It may be appropriate to change the constants in what follows. Nature will find a way to reduce the surface area by growing precipitate 2 at the expense of precipitate 1. 11. it is straightforward to show that two equal-sized spheres.51) and is smaller.18 Two precipitates of β-phase coarsening in an α-phase matrix. there is no reduction in the chemical free energy per unit volume. and this drives a diffusive flux to precipitate 2. have a total surface area of At2 = (3Vt )2/3 (8π )1/3 . each of volume Vt /2. For a precipitate of radius r. but how do we calculate the rate? The coarsening problem is essentially a diffusion problem. = 4π 3 V 3 r 3σ G + . 17:32:08 .54) (11.10 10 Section 16.3. but there is an addition to the free energy in the microstructure from the surface area. and as one precipitate becomes larger at the expense of the other.55) where v is the volume per atom. with a flux of atoms driven by the difference in chemical potential of B-atoms in the two precipitates. (11. an equilibrium quantity. The chemical potential for atoms in precipitate 1 is higher because of its larger surface-to-volume ratio. (11. For example.50) whereas a single sphere with volume Vt has a total surface area of At1 = (3Vt )2/3 (4π )1/3 . ∇μ r1 r2 R (11. For our two precipitates separated by the distance R. It can also be shown that At2 is the maximum surface area for the two precipitates. At decreases continuously until only one precipitate remains with area At1 . 16.53) where σ is the specific surface energy.

J = −cM ∇μ   3cMσ v 1 1 J = − − . 9.61) (11. R r1 r2 (11. Returning to Eq..56) (11. 11.9 Coarsening  drives a diffusion flux from the smaller precipitate 1 to the larger precipitate 2. dt vA   r 3cMσ Av2 ζ t  r22 dr2 = dt . R r1 r2 It is tempting to make an approximation    1 1 1 − = ζ = constant . 11.19) at different times. 4π 0 0  1/3 9cMσ Av2 ζ t1/3 .58 enables the elementary integration in Eq.60) where A is an effective area for the flux. These geometrical issues are challenging. 11.58 will evolve in opposite directions (which is true). and Eq. An important analytical treatment of coarsening was published in 1961 by Lifshitz and Slyozov [127] and independently by Wagner [128]. This ∇μ Using a variant of Fick’s first law for diffusion in a potential gradient (e. dt dt dr2 4π r22 = −JvA .59) (11. and solute concentration c  .62) (11.64 (shown in Fig.57 with Eq.64 is valid for 0 < ρ < 1. 11. What is interesting is that the precipitate size distribution is “self-similar. 11.58 dr2 1 4πr22 = 3cMσ vζ . and the full 17:32:08 . and precipitate 1 loses volume at the same rate dV1 dV2 = −JvA = − .57) (11. and will compensate each other (which is not quite right). The growth of precipitate 2 is proportional to the magnitude of the flux J. 11. but assuming a constant ζ of Eq. They confirmed the t1/3 coarsening law.58) The idea is that as one precipitate grows at the expense of the other. With increasing time.63) For coarsening. the radius of precipitate 2 grows as t1/3 . the peak of the distribution grows outwards to larger radius as t1/3 . the two factors in parentheses () in Eq.5 − ρ ρ − 1.295 11.5.5 4 2 exp .69) with a “mobility.64) f (ρ) = ρ 9 3+ρ 1. 11.g. This “t1/3 coarsening law” has been shown to hold true for the average radius when there is a distribution of precipitate sizes. There is also a question of the effective area over which the diffusion flux occurs.” M having units [velocity per force].5 where ρ = r/r0 is a dimensionless radius. dt (11. but also found this nonintuitive functional form for the precipitate size distribution  7/3  11/3   ρ 3 1. Eq. r2 = 4π (11.” The distribution of radii has the form of Eq. (11. 11.62 below.

distribution follows by scaling the radial dimension (and decreasing the height of the distribution for normalization).3 (a) Starting with Eq.1 Calculate the vacancy concentration in local equilibrium with a spherical void of radius r.0 f (ρ) 1. and the fraction 1 − f is not. (b) How does the aspect ratio scale with surface energy σs . assume that the fraction of surface f is covered by islands. many experimental tests seem to confirm self-similar behavior of the particle size distribution during coarsening. ignoring the vertical edges. and with atomic volume v. Today.2 (a) Calculate and compare the total surface energies for Frank–Van der Merwe and Volmer–Weber growth shown in Figs. 11.7a and 11. In practice. 11. The energy of formation of the vacancy in bulk material is Ef . this self-similar behavior is also challenged by transient effects from a different initial radius distribution function. computer simulation is a more general method for calculating coarsening behavior [129]. (b) Explain how elastic energy from the crystallographic mismatch of the film on the substrate could lead to Stranski–Krastanov growth. 11.16.0 0.8 1. 11. 11. v.4 0.5.2 1.2 0.5 0. For Volmer–Weber growth.Transformations involving precipitates and interfaces 296 2. Problems 11.5 1. specific surface energy γ .28.19 0. Nevertheless.5. presented with dimensionless radius.6 0. This “LSW” distribution is strictly valid only for low volume fractions of precipitate. Use your expression for total energy to derive Eq. elastic stiffness C. The function is negligible for ρ > 1. and transformation strain 3 ? Why? 17:32:08 .4 1. and the temperature is T.0 ρ [r/r0] 1. but it can be successful for volume fractions approaching 0.4 and strictly zero for ρ ≥ 1. and Xray tomographic methods promise more detailed experimental measurements in the near future. 11.0 0.6 Grain size distribution function from LSW theory. calculate the aspect ratio d/l of a platelike precipitate as a function of precipitate volume.0 t Fig.7b.

have sizes of 50 nm. Also explain how this could slow the growth rate of a pearlite colony.6 Consider a lamellar eutectic microstructure where the β-phase is the minority phase. show that local equilibrium between the α-phase and the liquid gives more solute in the liquid for greater undercooling. compared with a precipitate in the bulk if the grain boundary diffusivity is infinite? (Assume one precipitate of each type per grain.4 Suppose the grain size of an alloy is 1 micron. Assuming the rods grow in the arrangement of a square of edge length λ. What is the correction to your answer from part a? 11. Why does this facilitate a smaller lamellar spacing with increased undercooling? (b) With smaller lamellar spacing. both homogeneous and heterogeneous. (a) How much faster will a precipitate at the grain boundary grow.15b. You can also assume all shapes are spheres.15b. 11. Explain how this alters the relative positions of the free energy curves in Fig.297 Problems 11. It may become favorable for the β-phase to form as rods when its fraction of volume is small. and vice versa for the β-phase.) (b) Suppose the grain boundary diffusivity is 100 times that of the bulk. 11. the total surface area between β and α remains fixed even as the amount of β decreases and its lamellae are thinner. Use the collector plate mechanism to answer the following. If the center-to-center separation between sheets of the β-phase is fixed at λ. there is an increase in the surface energy term of Eq. and precipitates.29. 11. 11. what is the critical volume fraction of β in α at which the transition from lamellae to rods is favored by surface energy? 17:32:08 .5 (a) Using Fig.

The “square gradient energy” is an important new concept presented in this chapter. An unstable free energy function may prove useful for short times. it was pointed out that there are conceptual problems with a free energy that is concave downwards because the alloy is unstable. from 0 to +ζ . and the total length L. but the free energy pertains to equilibrium states. x. we use the thermodynamic tendencies near equilibrium to obtain a chemical potential to drive a diffusion flux that causes unmixing. an elastic energy is often important. we add a term to the free energy that includes the square of the composition gradient. 12. the composition of the sample is slightly greater than the average. Taking a kinetic approach. This is easy for a homogeneous sample of concentration c0 .1 shows a simple one-dimensional concentration fluctuation in the material. At the end of Sect. where f is the Helmholtz free energy per unit volume. there is an energy cost for gradients in composition. In addition to the surface energy. This term affects the time and length scales of the unmixing transformation. too. the total free energy is simply Lf (c0 ). since a region with a large composition gradient begins to look like an interface. but it is also essential to phase field theory and to the Ginzburg–Landau theory of superconductivity. 2. essentially as we have done since Chapter 2. Nucleation creates a distinct surface between the new phase and the parent phase. Nevertheless. Similarly.12 Spinodal decomposition Figures 1.d and 1. Spinodal decomposition does not involve a surface in the usual sense because it begins with infinitesimally small changes in composition.e. without concern for the spatial size of the region. Following the classic approach of John Cahn [130]–[134]. and the two phases differ significantly in their chemical composition or structure. Over the spatial range.6a. Taking the cross-sectional area of the sample to be unity. however.7 on unmixing phase diagrams. There is also an elastic energy that increases with the extent of unmixing and alters the “coherent spinodal” on the unmixing phase diagram. c(x) = c0 + δc for 0 < x < +ζ .5c. Consider the free energy to depend only on the local composition.b illustrate the difference between chemical unmixing that occurs by nucleation and growth (the topic of the previous Chapter 11) and spinodal decomposition (the topic of Chapter 12). in the range from −ζ to 0 the 17:31:12 .1 Concentration fluctuations and the free energy of solution First consider the free energy without a gradient energy. Figure 12. specifically the square of the gradient. i.

change is governed by the curvature ∂ 2 f /∂c2 . however. 17:31:12 .2.1.3) F = ζ 2 ∂c c=c0 where the second derivative ∂ 2 f /∂c2 is evaluated at c0 . from an initial state of uniform composition to the state depicted in Fig. We seek the change in the total free energy.9b). are the limits of stability of the solution.1 and 12. Fig. (12. If the amplitude of the fluctuation is small. Figure 12. δc + ζ ∂c c=c0 2 ∂c2 c=c0 where.2. c=c0 so the sign of the total free energy In Eq. Over the ranges of composition labeled “stable” and “metastable. 12. the free energy of the material can be found by Taylor-expanding about c0 . and the alloy will return to a homogeneous state. inflection points where ∂ 2 f /∂c2 = 0.3 both the terms ζ and (δc)2 are positive. 12. one-dimensional concentration fluctuation in a material of average composition c0 .” and with temperature the locus of these points defines 1 In the metastable region. F for the region 0 to +ζ becomes ∂f 1 ∂ 2 f (12.1 Concentration fluctuations and free energy of solution 299 t Fig. δc + ζ ∂c c=c0 2 ∂c2 c=c0 The change in total free energy after introducing the fluctuation is the sum of Eqs.” the second derivative of f is positive. the free energy change over the region where x < 0 is ∂f 1 ∂ 2 f F − = ζ f (c0 ) − ζ (12. In these ranges. any energy contribution from the “surface” at x = 0 is neglected.2) (δc)2 + · · · − ζ f (c0 ) .12. c(x) = c0 − δc in this region.1 The two concentrations cs in Fig.1) F + = ζ f (c0 ) + ζ (δc)2 + · · · − ζ f (c0 ) . composition fluctuations that are large enough to create a critical nucleus resembling the stable α2 -phase will grow.2 shows the curvature of c=c0 f (c) for a binary system exhibiting a miscibility gap (cf. 12. small fluctuations in concentration cause an increase in the free energy of solution. Small fluctuations in composition will decay. 12. They are called the “spinodal compositions. The positive and negative parts of the fluctuation account for the same amount of solute. minus f (c0 ) for the homogeneous alloy ∂ 2 f (δc)2 . for the time being. concentration is less than average.1 Idealized. and by conservation of solute. 12. Similarly. so the fluctuations are unfavorable. F. Therefore. 2.

2. Now suppose we broaden out the interface between regions of different composition. since it has steeper composition gradients that are more like surfaces.3a and 12.Spinodal decomposition d2F dc2 dc2 Metastable Stable Fig. 12. It lies inside the “miscibility line” where unmixing occurs across the common tangent of Fig. in fact.3b. 2.6 cs 0. 12.1 Justification for the square of the composition gradient The composition fluctuation of Fig.0 Free energy F as a function of composition for a system exhibiting a miscibility gap. Both are composition profiles with the same volumes for each composition (although there are two triangles in Fig.2 cs <0 Metastable Free energy d2F =0 Unstable 0.3a. the free energies of the profiles of Figs.3b versus one in 12. that there will be a greater free energy for the composition profile of Fig. metastable and unstable.2 where ∂ 2 f /∂c2 < 0. as shown in Fig.2 Adding a square gradient term to the free energy F(c) 12. or does it? 12. the “chemical spinodal. Compare Figs. the net positive or negative areas are equal).8 Concentration (c) Stable 300 1. 12. A curious feature of an infinitesimal change in composition is that there is no surface energy.” The chemical spinodal is shown in Fig. 12.11a. with its sharp interface between regions of different composition at x = 0.1 is quite abrupt.3a and 12. and the two phases become better defined? When does surface energy appear. In this highly unstable range of compositions of Fig. Also shown are the ranges of concentration for which the solution is stable. however. however. even infinitesimally small fluctuations will grow in amplitude. 17:31:12 . It also has more such interfaces per unit length.0 t >0 0.3b would be equal. 12.3b. 12.5a.11. We expect. What happens between the regions of different compositions as the unmixing transformation evolves.3. a surface between regions of different compositions. 12. 2.2 d2F dc2 0. 12. We expect to pay an energy penalty for such an abrupt modulation in composition since this is. so this is not a proper comparison.4 0. and therefore no nucleation barrier to start the unmixing transformation. 12. Compare with Fig. If the free energy were a function of the composition only.

3c over Fig.4) 2 This implies a continuum theory.3d. 12.g. A natural next step in the complexity is to allow the free energy to depend in some way on the gradient of composition. there will be gradients and curvatures of composition that can affect f (c).) When there is a small local fluctuation of c in an average composition of c0 . 12. 12. If we were to add a term to the free energy that is proportional to the magnitude of the composition gradient. when we compare the energies of the gradient regions of Figs.. ∇c. The free energy therefore does not depend on the magnitude of the gradient. but are beyond the present scope. which is fine for the present arguments. we notice that although the gradients are steeper in Fig.2 With inhomogeneities of composition. they occur over a proportionately smaller range in x.3 Four cases of composition fluctuations with composition gradients. Including all terms to second order ∂f ∂f − ∂f − → → 2 f (c0 +c.2 Adding a square gradient term to the free energy F(c) Fig. are the same for panels a and c.3c and 12. some atoms in a solid solution may have five solute neighbors. a simple expression for the free energy density like f (c) will suffice.2. 2 ∂c2 ∂(∇c)2 ∂c ∂ ∇c (12. for example. 12.3d. however. and another may have seven.301 12.3d. These considerations are important.2 Derivation for the square of the composition gradient For a simple solid solution that has uniform composition everywhere. The gradients in panels b and d extend over only half the range of x for panels a and c. but are twice as large for panels b and d. The square of the gradient favors the profile of Fig. ∇ 2 c) = f (c0 ) + c + − → ∇c + ∂∇ 2 c ∇ c ∂c ∂ ∇c   ∂ 2f − ∂ 2f 1 ∂ 2f − → →2 2 (c) + − + → (∇c) + 2 − → c ∇c . Of course. It turns out that we need to add a positive term with the square of the gradient to the free energy to account for “surface-like” effects. The gradients. e. when integrated over volume it would not matter if the gradient were steeper because a steeper gradient occurs over a proportionally smaller volume. dc/dx. t More subtly. (This additional dependence can be neglected when the − → composition is uniform since ∇c = 0. 17:31:12 . 12. 12. the free energy density is more complicated than a simple function of average composition. at the atomic level the material will show statistical fluctuations in local chemical environments.

∇ 2 c). and contributions to f (c) could be assigned to the different features in the composition profile. Such a detailed c(x. ∂x ⎡ ⎛ ⎞ ⎤ ∂f ∂ dc dU = ⎣ ⎝  2  ⎠ ⎦ dx . If there were a linear term. (c)2 + (∇c)2 + f (c0 +c. the magnitude |∇c| exists over a range of distances that is inversely − → proportional to the gradient itself. which came from the gradient.7) Identifying dV so that V is a gradient dV = U= ∂ ∂ 2c dx.9) . 12. z) is usually not practical. and it is usually separate c(x. It can ∂x2 ∂c 2 ∂x ). Also. The sign of the gradient (or the sign of dc/dx in 1D) switches with direction. (12.Spinodal decomposition 302 Equation 12. must be zero. be recast To do so. z) contains this information. y. f . ∇ 2 c) = f (c0 ) + − → 2 2 ∂c 2 ∂(∇c)2 ∂∇ 2 c (12. Although not fully justified. (12. y. The total free energy.1.6) ∂ c ∂x2 ∂ ∂x 2 The fourth term in brackets (with the ∂2c ) came from the Laplacian. this has become a standard part of the explanation of spinodal decomposition.5) At this point we reduce the problem to one dimension. linear in − → c and ∇c. ∇c. FT .2. is an integral of the free energy per unit volume. This simplifies the result 1 ∂ 2f 1 ∂ 2f − ∂f − → → ∇ 2c . so ∂ c ∂x2 ∂c . The same argument pertains to the last term of Eq. so the fluctuation would not be infinitesimal and a Taylor series approach would not make sense. ∇c.4. such as the derivative with respect to the composition gradient. z) impractical to− → into its different features. into the form of the third term (with the integrate by parts following the prescription   UdV = UV − VdU . ∂x2 so ∂f  2 .8) (12. To discount the gradient − → term. It is necessary to go to second order in composition derivatives to consider an instability of free energy against an infinitesimally small fluctuation. z) with f (c0 +c. and we allow some independence between the contributions to f (c) from the local composition and from the spatial derivatives of composition. The first-order terms.4 may look peculiar because of its derivatives with respect to composition derivatives. A detailed concentration profile including all features of c(x. y. the free energy could be lowered more and more with an increasingly large c for the fluctuation. Instead we replace f c(x. y. over the volume of the material (now the length in x times the area. which is not realistic for a solid solution. 12. ∇c. as explained at the − → end of Sect. so F cannot depend on |∇c| either. A)  2   1 ∂ 2f 1 ∂ 2f ∂c 2 (c) + FT = A f (c0 ) + 2 ∂c2 2 ∂ ∂c 2 ∂x ∂x  2  ∂f ∂ c +  2  dx . a practical necessity to avoid numerical computation. ∂ c ∂c ∂ dx V= ∂x2 17:31:12 (12. first note that it is a vector.

7 and 9.10 (without the first term on the right-hand side) into 12.2 Adding a square gradient term to the free energy F(c) 303 Substituting into Eq.18) Equation 12. Notice that it is proportional to (∂c/∂x)2 .” 17:31:12 .16) (12.11 is familiar from Sections 2. FT = A f (c0 ) + 2 ∂c2 ∂x This is sometimes written. so dc/dx = ∂c/∂x) gives ⎤ ⎞ ⎡ ⎛  2  2    ∂f ∂c ∂f ∂ ∂f ∂c ∂ c ⎦ dx .4.15) (12. or the square of the gradient in 3D. as  2  ∂c dx .14) (12. FT = A f (c) + κ ∂x and in three dimensions it is FT =  − →2 f (c) + κ ∇c dV .12.17) (12. 12.13) ∂ c ∂x2 ∂ 2f ∂c 2 . Substituting Eq. is called the “Cahn–Hilliard equation. (12. κ = κ2 − 2 ∂c Equation 12.  2  − ⎣ ⎝  2 ⎠ dx =  2  ∂c ∂ ∂ c ∂x ∂x2 ∂ ∂ c ∂ ∂ c ∂x ∂x2 ∂x2 (12.7 (and noting that c = c(x).18. 12.11) 2 ∂ ∂c 2 ∂c ∂ ∂ 2 c ∂x ∂x ∂x2 The first line in Eq. An important point is that this term does not scale with the volume of the material.12) κ FT = A f (c0 ) + 2 2 2 ∂c 2 ∂c ∂x with the new parameters evaluated at the alloy composition c0 κ1 = κ2 = ∂ ∂f  2 . ∂ ∂x ∂κ1 1 .12 simplifies to  2  1 ∂ 2f ∂c 2 (c) + κ dx . This is seen more clearly with definitions    2  1 ∂ 2f 1 ∂κ1 ∂c 2 (c) + − 2 dx . 12.10) ∂x2 The first term on the right-hand side of Eq. (12.6  1 ∂ 2f FT = A f (c0 ) + (c)2 2 ∂c2 ⎡ ⎛ ⎞⎤   1 ⎣ ∂ 2f ∂ ⎝ ∂f ⎠⎦ ∂c 2   + −2 dx . (12. The second line is new. The first term can be neglected for bulk materials. 12. (12. which contains the square gradient energy. whereas the last term does increase with the volume (range of x).10 is to be evaluated at the ends of the material. with an obvious definition.

This is discussed in more detail after a mathematical excursion into the calculus of variations.1. A more gradual variation in composition gives a smaller gradient. The problem becomes one of kinetics. as is the case at high temperatures. The net effect of the favorable unmixing term and the unfavorable gradient energy term is to set an optimal spatial scale for the unmixing. ∂ 2 f /∂c2 . where the free energy increases quadratically with the composition gradient.2. 12. This is not so simple as shown in Fig. 12. the gradient would be large.3 Suppose the solid solution is stable. so sharp gradients are disfavored at all temperatures. At the critical temperature where ∂ 2 f /∂c2 changes sign. 12. 17:31:12 . The gradient energy promotes a longer spatial scale. it is beneficial energetically to spread the composition variation over a wider range of distance. 12. On the other hand. The difficulty is that we need to conserve atoms. and an arbitrary increase in composition in one location must be balanced by a decrease in composition elsewhere. since we could have more regions with small −c.2. The constant κ1 is the composition dependence of how the free energy depends on the lumpiness of composition. which drives diffusion. and seek the condition when FT becomes unstable against the formation of a small concentration fluctuation. also acts like a restoring force at high temperature. The constant κ2 is more obviously like a spring constant. this ordinary chemical restoring force is lost. With an increasingly larger volume free energy below the critical temperature. Since it is the square of the gradient that is important. but diffusion occurs more efficiently over smaller spatial scales. and fewer regions with very large +c. an energy proportional to the magnitude of the gradient would have 4 no such preference.4 Note that the second term from the ordinary chemical free energy. the characteristic modulation wavelength for the composition fluctuations becomes shorter at lower temperatures. The quantity in the square braces in Eq.12 acts like a spring constant.3 Effects of the square of the composition gradient The composition gradient describes something like a fuzzy surface. If the surface were sharp in composition. A penalty is paid when the fuzzy surface gets sharper. the term in square brackets and the gradient energy remain positive.3.Spinodal decomposition 304 12. 12. and some of the steps may be both unfamiliar and subtle.4 Preview of the Cahn approach to spinodal decomposition There is a lot in what follows.3 Constrained minimization of the free energy We seek to minimize the FT of Eq. We seek to minimize FT subject to the constraint of constant alloy composition 3 As we saw in the discussion of Fig. 12.1 as you go forward. It may be helpful to follow the steps summarized in Box 12.12.

1 Calculus of variations Suppose we have a function f that depends on x. • Identify the wavevectors.W. c(x). the “Euler equation. When f is integrated over a range of x  L/2  dc f x. • Take the divergence of Fick’s first law to get an equation for the kinetics of composition change. 2π/λ. dc/dx) that depends on average composition. This brings a constraint equation and a constant μ.3. although largely in one dimension. dc/dt. Other references include [131]–[134]. L/2 A c(x) dx = A L c0 . (12. and add elastic energy to the free energy.1 Summary of the Cahn approach to spinodal decomposition The following approach to spinodal decomposition is that of J. of the compositional fluctuations that can grow. and dc/dx. and identify the fastestgrowing compositional wavelength. which is the chemical potential. c(x). The Cahn approach. (dc/dx)2 • Use calculus of variations (developed below as a mathematical excursion) to minimize f (x.3 Constrained minimization of the free energy Box 12. d2 f /dc2 – the square of the composition gradient.” • Use the method of Lagrange multipliers to conserve chemical composition.20) −L/2 Minimizing an integral with a constraint equation over the same limits requires a pair of results from a topic of mathematics called “calculus of variations. c(x). dc/dx) over the volume of material.” We next make a brief excursion through this topic. is a classic. 12. (12. and on composition fluctuations in the material through two terms – the compositional effect on the free energy. • Recast the composition fluctuation δc(x) as a Fourier transform.21) . c(x). λ. with added explanation. • Recognize that the lattice parameter of the crystal changes with chemical unmixing. Cahn [130]. It includes these steps • Develop a free energy density f (x. where flux is −→ proportional to ∇μ. dx . obtaining a condition for the optimal compositional profile. FT = dx −L/2 17:31:12 (12. • Use the gradient of the chemical potential to obtain a diffusive flux with Fick’s first law.19) −L/2 L/2   c(x) − c0 dx = 0 .305 12. This suppresses the temperature of unmixing.

and define a small “variation. ∂c ∂ dx dx δf = (12. dx dx where δx = 0 because the same range of integration is always used.26 in this integral L/2  0= −L/2  ∂f d ∂f δc + dc δc dx . For small variations δc(x).Spinodal decomposition 306 the integral has different values. c(x).23) δf = f x. we call f a “functional. depending on the specific profile of c(x).26) The condition that FT is an extremum (minimum. we hope) is simple L/2 0= −L/2   dc δf x. C(x). (12. Consider a composition profile c(x) that is quite close to this desired C(x). . since c depends only on x. ∂c ∂ dx dx (12.28 is integrated by parts to give ∂f dc ∂ dx d δc = dx  ∂f dc ∂ dx L/2 δc L/2  − −L/2 −L/2 d ∂f dc dx ∂ dx  δc dx . and Eq. dx . 12. we expect ∂f ∂f dc δc + dc δ . 17:31:12 .29) 5 Because f is a function of the functions c(x) and dc/dx.27) This is an analog to the familiar case that df = 0 when seeking an extremum of a function such as f (x).” δc(x) as δc(x) ≡ C(x) − c(x) . dx dx dx dx dx (12.24 becomes δf = ∂f ∂f d δc + dc δc .” instead of a function. c(x). This profile sets the derivative dc/dx. ∂c ∂ dx dx (12.28) There is one more simplification needed before the main result.24) We make an important observation about the last variation in Eq.5 Using Eq. We seek the concentration profile C(x) that gives an extremum of FT (which may be a maximum or minimum of FT ). (12. 12. (12. this nearby profile c(x) causes a change in the integrand at any point x     dc dC − f x.25) The observation is that the operator δc commutes with the operator d/dx. The second term in the square braces of Eq. dx (12. 12.22) Compared with the optimal profile C(x) that minimizes FT .24 δ dC dc d d dc = − = (C − c) = δc . 12.

32) −L/2 by adding Nμ δc to the integrand of Eq. which is the condition of conservation of atoms. (12.3 Constrained minimization of the free energy The simplification is that the first term on the right-hand side of Eq. Here Nμ is a constant called a “Lagrange multiplier.6 which becomes  L/2  d ∂f ∂f 0= − Nμ δc dx . of course.29 vanishes at both limits. 12. This can be argued to be true because at the limits. the composition is fixed. 12. which is over the same limits of integration. 12.31) This Euler equation is the condition for FT to have a minimum value if we are free to select the composition profile. 12. 12.30) The final trick is an observation. If the term in square braces in Eq. It turns out that μ is the chemical potential (free energy per atom). 12.27 remain the same up to Eq.28 becomes.27. 12. 17:31:12 . 12. the “Euler equation” of variational calculus d ∂f ∂f =0.30. as usual.32 equals 0.27. The arguments that follow Eq. (12. we cannot use the previous argument in passing from Eq.30 for the same result. it remains zero.30 to Eq. 12. With the definition K = f − Nμ .307 12.20.34) 6 We could have directly added Eq. We therefore try to bring in the constraint of constant alloy composition L/2 δc dx = 0 (12. Equation 12. we could pick a δc(x) that would make the total integral nonzero.32 to 12. to conserve atoms.33) − dc ∂c dx ∂ dx −L/2 Unfortunately. − dc ∂c dx ∂ dx (12. It is selected. L/2  0= −L/2 d ∂f ∂f − dc ∂c dx ∂ dx  δc dx . when Nμ δc is subtracted from the integrand of Eq.” and is yet to be determined. This condition is over the same range of x as the extremum problem that gave the Euler equation.3. 12. 12. Because Eq.2 Constraint of constant composition We return to the constraint of constant alloy composition. Our δc = δc(x) is arbitrary.31 because δc is no longer arbitrary owing to Eq. This gives the main result. Eq.32. It is also physically reasonable that this constant term becomes negligible as the other term grows with L. Although we lose this degree of freedom in selecting a δc. 12. (12. finally. the new Nμ is an adjustable parameter that can be tuned so the integrand vanishes. 12.30 were not zero everywhere.

The real situation for unmixing becomes one of kinetics. 12. 2. and this infinitesimally small contribution to the free energy can be neglected. It is in the kinetics that the square gradient energy becomes important in a continuous phase transformation.5a. − dc ∂c dx ∂ dx (12. so it is expected that the two-phase region might be diminished in size.37) after substitution into Eq. Both the miscibility line and the chemical spinodal remain unchanged.16 1 ∂ 2f (c)2 + κ f = f (c0 ) + 2 ∂c2  ∂c ∂x 2 .11b or 12. by setting a very long spatial scale for the unmixing. which is the usual form of the chemical potential. In equilibrium. ∂c ∂x (12.36) and using this definition of f (c) (as in Eq. Composition gradients are infinitesimally small at the upper critical temperature for the composition c = 0.11b. On the other hand. we have μ = (1/N) ∂f /∂c. the composition gradient and the square gradient energy can be suppressed. In essence. 2.36 is an extra cost of unmixing. 2 ∂c2 (12. More generally.17) f (c) ≡ f (c0 ) + 1 ∂ 2f (c)2 .3 Minimizing a constrained free energy functional Using the free energy per unit volume from Eq. Allowing for equilibrium allows for infinitely long spatial scales.3.39) (12.5. the unmixing process would require an infinite time if the unmixing were over infinite distances. Eq. 17:31:12 .38) f (c) + κ f (c) + κ dc ∂c ∂x dx ∂ dx ∂x Recognizing that ∂/∂c δc = ∂/∂c (c − c0 ) = 1 ∂f d ∂c − Nμ − 2κ =0. however. and we approach it with the diffusion equation. 12. We recognize μ as a chemical potential that now includes the effects of the square gradient energy. (12. (12. 12.35      2  2 d ∂ ∂ ∂c ∂c − Nμδc − − Nμδc = 0 .40) Note that when κ = 0. The problem is more subtle for two reasons.31 becomes ∂K d ∂K =0. 12. ∂c dx ∂x ∂f ∂ 2c − 2κ 2 = Nμ .308 Spinodal decomposition it is straightforward to show by substitution for f that the Euler equation of 12.35) 12. These free energy arguments suggest how the square gradient energy may alter the unmixing phase diagram of Fig. the last term in Eq. 12. so the square gradient energy has no effect on the equilibrium phase diagram of Fig.40 is also a constant.

1 involves random mass transport between adjacent slices of material. A gradient in composition drives the flux of atoms. the three-dimensional form of Eq. Equations 9. We therefore use the continuity equation ∂c ∂ =− J. as suggested by Fig. the jump frequencies of atoms are biased by their chemical potential.4. J = −M ∂x ∂c ∂x   2 ∂ 3c ∂ f dc − 2κ . J −→ J = −M ∇μ .46) This is analogous to the three-dimensional form obtained by applying the threedimensional continuity equation ∂c  · J .44) − 2κ∇ 2 c . 12. In general.40   ∂ 2c ∂ ∂f − 2κ 2 .4 The diffusion equation 12. 3.43  2 2  ∂c ∂ 4c ∂ f∂ c − 2κ =M .43) where κ is the gradient energy coefficient. 17:31:12 . Using the chemical potential of Eq. ∂t ∂x to make the transformation of Eq. 12. 12.87 to 9.47) 7 When lattice sites are conserved and populated fully. so if one slice has a higher concentration than a second slice next to it.41) where M is a mobility coefficient.42 is   ∂f (12. 3. using the gradient of the chemical potential7 (including the squared gradient energy) to drive the diffusive flux. though. there will be a net transport of atoms from the first slice to the second. ∂t ∂c2 ∂x2 ∂x4 (12. We take a similar approach here. (12. now assumed independent of concentration. The gradient in chemical potential follows the gradient in composition when there are no chemical interactions between atoms.309 12. the gradients of chemical potential can be replaced by a single gradient in the difference of chemical potentials between the two species A and B in solution.6. J = −M grad ∂c Solving for fluxes is typically less useful than solving for the time evolution of the concentration profile. The idea is that the atoms jump in random directions.45) (12.42) (12.90 gave a modification of the diffusion flux owing to chemical preferences.1 Diffusion driven by a gradient in chemical potential The standard derivation of the diffusion equation in Sect. J = −M ∂c2 dx ∂x3 (12. for example.4 The diffusion equation 12. = −∇ ∂t (12. Without derivation.

The approach is to work with a Fourier transform expression for the concentration  ∞ A(Q.48) with the definition δc(r.46 has a notoriously inconvenient form. inversely related to the wavelength λ as Q ≡ 2π/λ. t) is the only function of t. we would be back to the case of unmixing used for obtaining the unmixing phase diagram in Fig. Although composition fluctuations with other wavelengths also grow. but it is possible to make progress if the concentration fluctuation δc is small. The alloy cannot unmix with infinitesimal fluctuations if ∂ 2 f /∂c2 > 0.Spinodal decomposition 310 to Eq. After dividing by c(x) − c0 in the form of Eq. (12. and at any time it is the inverse Fourier transform of the concentration fluctuation  1 +R(Q)t ∞ δc e−iQx dx . and the system would be stable. t) eiQx dQ . 2κQ4 in Eq. 12. all small-amplitude concentration fluctuations would decay exponen- tially with time.52) When Q < Qc . so we need a larger free energy driving force for unmixing. so the gradient energy were unimportant. 12. and it must be larger than the gradient energy term. (12. 2κ ∂c2 (12. This requires that ∂ 2 f /∂c2 be negative. t) ≡ c(r) − c0 .49  2  ∂ f 2 4 R(Q) = M (−1)Q − 2κQ . 12. the gradient energy tends to suppress unmixing. if ∂ 2 f /∂c2 were positive. we note that exp(iQx) is the only function of x. 12.51) ∂c2 To grow a concentration modulation. This critical condition defines a critical wavevector Qc (or critical wavenumber)  1 1 ∂ 2f 2 Qc = − . is assumed to be an exponential function of time. 17:31:12 . and A(Q.51. 12. A.50) e A(Q. If κ were zero. t) = 2π −∞ where R(Q) is a growth rate for the wavevector Q. (12.46.49 into our modified diffusion equation Eq.8 The critical condition is when the two terms are equal in Eq.11b because a concentration modulation would grow whenever ∂ 2 f /∂c2 is negative. 12.2 Fourier transform solution Equation 12.4. When we substitute Eq. 12. 2. Another important step is to find the wavevector Qmax that grows most rapidly. The amplitude for the concentration modulation. For finite κ (which is always positive).49) δc = c(x) − c0 = −∞ where Q is a wavevector.51. a composition modulation can grow because the gradient energy is not large enough to suppress the modulation. = ∇ · M grad − 2κ∇ δc ∂t ∂ δc (12. the most rapidly growing 8 Of course. R(Q) should be positive – when R(Q) is negative the concentration modulations will decay with time.44    ∂δc ∂f 2  .

For the case of 17:31:12 .54.. Spinodal decomposition is known to occur in some alloys such as Co–Cu. 12. but this obviously depends on the atomic mobility or diffusion coefficient. but it is not so easily identified because its characteristically short time and spatial scales make it a challenge to demonstrate by experiment. R(Q). Q. It suggests a real-space concentration profile characterized by an approximately periodic. and Qmax refers to the maximum in R. wavelength should set the spatial scale for the early stages of spinodal decomposition. Qmax = − 4κ ∂c2 (12. 3D interconnected network of regions of high and low solute concentrations. Estimating Qmax or R(Qmax ) is challenging. the Fourier spectrum will be dominated by a single component. Qmax = Qc / 2.53) (12. shortly after a quench into the unstable region of the phase diagram. R(Q) is positive and the amplitude A(Q. Differentiating Eq. and new methods of rapid quenching may better prepare an alloy for unmixing by spinodal decomposition.54) √ From Eqs. The value Qc corresponds to the point R = 0. The characteristic rate is something like 10−3 s.4 shows the behavior of the amplification factor. a concentration fluctuation creates an elastic strain field in the material. For wavenumbers less than Qc .5 Effects of elastic energy Amplification factor R(Q) vs.311 t Fig. vs.4 12. As a result. there is a difference in size of the different atoms in a alloy. dQ ∂c  1 1 ∂ 2f 2 . However.5 Effects of elastic energy In general. Today we believe that unmixing by spinodal decomposition is much less common than unmixing by nucleation and growth. This selective amplification implies that.51   ∂ 2f dR(Q) = 0 = M −2 2 Qmax − 8κQ3max . 12. t) grows exponentially with time. Figure 12. The most rapid growth is for modulations with Qmax .e. Q. 12. The conventional thinking is that the wavelength λ for maximum growth is of order 10 nm.52 and 12. i. longer wavelengths. only recently have methods become routine for characterizing composition at dimensions less than 10 nm. 12.

To include the elastic energy in the continuum theory of spinodal decomposition. and utilizing the fact that the Fourier components do not interact. the Cahn–Hilliard free energy equation (Eq.Spinodal decomposition 312 nucleation. we saw that the positive elastic energy from a misfitting spherical nucleus must be added to the work of formation. sometimes markedly changing the kinetics.18) must be modified as  . The strain energy generated by growing concentration fluctuations also plays an important role in the spinodal decomposition of crystalline solids. The elastic strain energy for an arbitrary composition fluctuation can be found in much the same way as was done in solving the diffusion equation by Fourier transform methods. computing the elastic energy of each Fourier component. By Fourier transforming the composition.55) vol where E is Young’s modulus. ν is Poisson’s ratio and η is the fractional change of lattice parameter with composition. 12. one obtains a general elastic energy contribution e= E η2 1−ν  (δc)2 d3 r . (12.

” is the limit of stability with elastic contributions included. 2. and the alloy “s” (having a composition within the assessed spinodal) showing characteristics of continuous unmixing [135]. It is found by displacing the chemical spinodal downward by an amount equal to the additional term in Eq. This curve is called the “chemical spinodal. Their chemical compositions were on the two sides of the spinodal line.6). 12. with the alloy “n” (having a composition between the miscibility and spinodal lines) showing characteristics of nucleation and growth.56) η2 (δc)2 + κ (∇c)2 d3 r . 2 1−ν ∂c (12.5b shows the distribution of Cr atoms in two Fe–Cr alloys that were prepared as solid solutions. Cahn showed that the 2η2 E/(1 − ν) term can be large in many solid solutions. 12.2). 17:31:12 . Figure 12. the “coherent spinodal. E − → (12. Between the solid curve for the miscibility line and the dashed line for the coherent spinodal in Fig. f (c) + F= 1−ν vol It is a straightforward exercise to show that the limit of stability of a solid solution is no longer ∂ 2 f /∂c2 = 0 but 2η2 E ∂ 2f + =0.57.5a. 12. as marked approximately by the “n” and “s” in the phase diagram of Fig. the solid solution is unstable against unmixing by nucleation and growth (the region labeled “metastable” in Fig. 12.5a. The heavy solid line shows the equilibrium miscibility gap (cf. Below the equilibrium phase boundary is a dashed curve corresponding to the locus of points such that ∂ 2 f /∂c2 = 0.” The lower curve.57) A more complete thermodynamic picture of spinodal decomposition can now be illustrated with Fig. then annealed to induce unmixing. Fig. 12. Notice the different morphologies of these partly unmixed alloys.5.

Problems 313 1.e. Not only does the additional elastic term change the criterion for the limit of stability. t = 0) exp −MQ 1−ν ∂c2 The new term with the elastic constants implies a decrease in the rate at which concentration waves are amplified.1 Explain why the chemical free energy cannot depend linearly on the magnitude of − → the composition gradient.2 0. it alters the kinetics of decomposition.56.4 0. i. (12. 12. whereas “s” shows characteristics of spinodal decomposition. Spinodal decomposition in anisotropic systems is characterized by preferential rapid growth of concentration waves along elastically soft directions [133]. the linearized form of the diffusion equation (12. |∇c|.18? (b) Calculate an approximate value of κ for an equiatomic alloy that undergoes spinodal unmixing. Problems 12.8 α" Chemical α' n 0. too.4 α Temperature (zV/k) 1. t) = A(Q.. 12.0 b n s (a) Phase diagram of a binary system with a miscibility gap (solid curve). for Qc = 2π/λ is 10 nm. Suppose the wavelength.2 (a) What are the units of κ in Eq.58) A(Q. Also shown are the chemical spinodal where ∂ 2 f /∂c2 = 0 and the coherent spinodal for which ∂ 2 f /∂c2 + 2Eη2 /(1 − v) = 0. With Eq.48) has a solution in Fourier space given by  2    2 ∂ f 2 E 2 + 2η + 2κ Q t . the positive elastic energy contribution slows the reaction. approximately as indicated in the phase diagram of part a. λ.2 0.5 a 0.4 0. As for the nucleation of a misfitting sphere.6 Concentration (c) 0.0 t Fig.2 Miscibility line 1.0 0.%Cr and Fe–38at.0 0. 12.%Cr after annealing at a temperature of 85% of the peak temperature of the coherent spinodal. The assumption of an elastically isotropic solid is often inaccurate. 12.8 1. Horizontal width is 8 nm. however. (b) Atomic probe tomography maps of Cr atoms in alloys of Fe–27at. Image “n” shows characteristics of nucleation and growth. Reprinted from [135].6 s Coherent 0. 17:31:12 . with permission from Elsevier.

(Hint: Consider the chemical free energy expressions such as Eqs.314 Spinodal decomposition Suppose the critical temperature for unmixing is 1000 K.20–2. and the spinodal decomposition occurs at 500 K. 2.23 and their critical temperature.3 Suppose a concentration fluctuation has the form δ0 .) 12.

59) δc(x) = 2 Using this expression in Eq. derive the growth rate R(Q) of Eq.51. 12. iQx e − e−iQx . (12. 17:31:12 .50. 12.

Chapter 12 − → explained the concept of a square gradient energy κ(∇c)2 . and consistent with our experience that we can clearly distinguish a solid from a 17:31:09 .1. The phase field method can be used for problems other than chemical unmixing and hence its name – the different phases or structures in a material are represented as fields in space. 13. and how a set of order parameters {ηj } vary along the trace. − →2 κ(∇c) served much like a surface energy. and identifies the emergence of modulations or sharp interfaces in time [136]–[138].1.g. When integrated over volume. is analyzed for the typical case of an antiphase domain boundary. The idea is illustrated in Fig. composition) and a nonconserved order parameter (e. and with spatially uniform atom arrangements. This is a useful definition. compact. and the different equations for the evolution of a conserved order parameter (e. The structure of an interface.13 Phase field theory Chapters 10 and 11 discussed phase transformations where melting and precipitation occurred as distinct regions with distinct interfaces. “Phase field theory. This second approach is conceptually more general than the first because it is possible to obtain a sharp interface and a specific surface energy σ in the limit as the concentration gradient becomes very steep. and the dynamics of domain growth are described. spin orientation) are discussed. where it has had considerable success.” discussed in this chapter. The specific surface energy σ .1 Spatial distribution of phases and interfaces 13.. especially its width. which shows a one-dimensional trace through a polycrystalline microstructure. This chapter first describes the phase field concept in more detail.g.1 Phases as distinct volumes with abrupt interfaces To this point we have considered a phase to be a region of material that is distinct. Much of the interest in phase field methods stems from advances in computing. A phase is therefore a distinguishable region with uniform properties. and mentions its use for solidification problems. multiplied by the area of interface. works with the composition profile as a field in space. was essential to the thermodynamics and kinetics. Numerical methods for integrating partial differential equations in a phase field model can be more convenient than handling the boundary conditions at moving interfaces between phases.. An example of a free energy function is presented. located in a well-defined volume. 13.

Sometimes this diffuse interface is physically real. liquid. Chapter 12 gave physical examples of such interfaces during spinodal decomposition. 13. such as the velocity of the solidification front during freezing. the solid and liquid phases had separate free energy functions. The composition was allowed to be a continuous variable. used to describe chemical long-range order in 17:31:09 . The chemical composition varied continuously between different regions.2 Phases as field quantities with diffuse interface In contrast. there was a specific composition for the solid and a specific composition for the liquid given by the common tangent rule (Fig. their interfaces are abrupt and their trajectories are easy to identify.5). 2. (b) Dependence of order parameters on position. T) and FL (c. for example. We now generalize this approach. 2.. If the two phases are distinct. After image by Cenna. 13.1 b x Position Illustrative treatment of grain boundary interfaces with the phase field method.10). T).g. One way to follow the growth of a microstructure in a material is by following the motions of interfaces. where interfaces between phases have state variables that are intermediate between those of two adjacent regions. To the free energy we add a term that includes the square of the gradient of the state variable (and complex materials can require terms having more than one gradient). or the motions of other interfaces in solids (e. and sometimes a diffuse interface is a computational convenience. sometimes it is an averaged quantity. Wikipedia (2012). (a) Grain structure of polycrystalline material. This number is much like the order parameter.1. this chapter assumes diffuse interfaces between phases. and at low temperatures the unmixing was understood with a single F(c.10 and 11. This distinctness of the two phases was relaxed somewhat when we considered chemical unmixing on an Ising lattice (Fig. Sections 5. L. As the temperature increased to Tc . the distinctness of the two unmixed phases became infinitesimally small. FS (c. however. At a fixed temperature.9). with each crystallite described by an order parameter ηj .Phase field theory 316 a y +1 x η1 η2 η3 η4 η5 η 0 t Fig. and the free energy depended on the square of the composition gradient. For a simple binary alloy. We first parameterize the phases throughout a material with a continuous number. T) curve having minima at two different compositions.

) Free energy versus composition curves for three different values of φ are shown in Fig. The dashed curve. We also allow the phase variable φ(x.4 φ 0.e.0 0.0 0. Figure 13. There is no reason why L cannot also vary with position.2. and 1.1 Spatial distribution of phases and interfaces 317 F(c) F(φ.T′ ) φ=1 φ = 0.0 0. this free energy functional must be compatible with the materials physics by having two minima at φ = 0 and φ = 1. This is not physically accurate. 13. and a minimum for the solid (φ = 1) at small c. FS (c. 13.0 b 0.0 for regions of space that are solid. The free energy is no longer two functions.. for example. the functional F has a minimum for the liquid (φ = 0) at large c. In an alloy with domains as shown in Fig.0 for regions of space that are liquid. say.5. y.8 1.2 0. Sect. For example. a transition between two minima of F occurs across a region of space with intermediate values of φ and c. c. In “phase field theory” we do the same for phases.2b. so to be compatible with the Gibbs phase rule of Sect.4 0. and no minimum at. T  ) that is acceptable for solidification.8 1.10.6 0. Our interest at the time was finding the thermodynamic value of L(T).c.0 c (a) Typical function F(φ. so we can make a spatial field of the LRO parameter in the material.13.5 φ=0 0.6 0. An additional minimum could produce an additional phase.2a is continuous in φ and c. 7. T) and FL (c. L could switch from. T) are not allowed. T) that are confined to regions of solid or liquid. L(x.2 a 0. (Evidently the B-rich solid melts at a lower temperature than the A-rich solid. What is new is that the order parameter is allowed to take intermediate values in the transition region between the two domains. The interface between the solid and liquid phases is therefore diffuse. of 17:31:09 . T  ) at fixed T  for both solid and liquid phases. Because F(φ. c. such features in F(φ. y. z).8 0. c. and no additional phase is predicted with this free energy functional. 2. intersects the curves for φ = 0 and φ = 1. the common tangent. the common tangent intersects only the curves for φ = 0 and φ = 1.2 t Fig. i. F(φ.6 0.4 c 0.5. and (b) three curves F(c. φ = 0. T  ) of Fig. giving the order as a function of temperature. Equilibrium is predicted between one solid phase and one liquid phase. and misses the curve for φ = 0.2 0. z) to take intermediate values at the transition between liquid and solid. Fortunately. For solidification.2a shows a F(φ. we assign a number 0. 13. T) that applies over all space.8. 2. c. c. T  ) for different φ. +1 to –1 across the antiphase domain boundary in the middle of the figure. At this temperature. but is now a functional of the function φ.

the surface energy σ .1 In other cases. the crystals are less elongated. 5.” so it shows the microstructure of the cast metal. the thermal diffusivity Dth .” Much attention has been given to solidification microstructures. interfaces really are physically diffuse. which is controlled by the rate of heat extraction from the liquid–solid interface. In other situations when a diffuse interface is not physically realistic. so heat flow was from the hot liquid center to the cold walls during solidification.7e. The cracking followed approximately the crystal boundaries in a process termed “intergranular fracture. Instead this section identifies some features of solidification. Mold walls were at left and right. however.3). since they affect the mechanical properties of cast metals (causing sometimes a tendency for intergranular fracture. but as pointed out earlier with Fig. 13. and the phase field method gives physical information about how a diffuse interface varies in chemical order or in chemical composition. The mold walls were cold. being more “equiaxed. the phase field method can still be useful for computational modeling. The spacing between the growing columns or dendrites is an important quantity. especially if it predicts an interface that is narrow in width. In the center of the casting. undercooling T.2 mm. 17:31:09 . the growth rate of dendrites is caused by an instability.3 Cold wall Cold wall An important success of phase field theory has been the prediction of solidification phenomena such as dendrite growth [137–140]. 13. 13. 1 mm Image of a fracture surface of a metal casting. and the free energy difference of the solid and liquid G. the figure shows how the metal crystals grew with a columnar structure with spacings of approximately 0.3 shows the fracture surface of a metal that was cast into a mold. however. Most growth models are organized to predict trends as functions of v.Phase field theory 318 course. and is therefore challenging to calculate. 1 Sometimes it can be accurate in an average sense. as in Fig.2 Solidification t Fig. and the next sections explain some general features of phase field theory. the solute diffusivity Dm . Near the walls. Some physical parameters that control the problem are the velocity of the solid–liquid interface v. Figure 13. It is not practical to develop here the analytical tools and methods of these calculations.

such as the adaptive mesh refinement techniques that allow accurate treatment of small structures. Some advances in this field are well beyond the scope of this text. For example. vc = Tc (13. blocking their growth. the required undercooling for freezing is strongly time-dependent. 17:31:09 (13. where xc is the distance over which a characteristic −→ undercooling Tc develops in the temperature gradient ∇T. a steady-state growth morphology is expected after this coarsening process is complete. The third and fourth frames show instances where secondary arms from the longer primary dendrites deplete the solute in front of shorter dendrites. Nevertheless.3. 13. but the darker gray nuclei were misoriented at a 25◦ angle. 13. This correctly corresponds to a smaller dendrite spacing for faster rates of freezing. The crystallographic direction for growth of the light gray nuclei (at top and bottom) is horizontal. the growth velocity allows approximately enough time for solute diffusion over the tip of the growing dendrite. Under steady-state growth conditions. The chemical segregation coefficient.3) . An important characteristic velocity is −→ ∇T Dm .1) where r is the radius of the growing dendrite tip. Furthermore. which is a strong effect.4 are informative. and tertiary arms take their place. however. a high density of dendrites nucleates initially. many of the misoriented dendrites become blocked. giving rise to an increasing volume fraction of secondary arms. After time. k = cs /cl is also important – if it is much different from 1. simulation results such as those in Fig. Phase field calculations are well suited for numerical simulations of solidification dynamics. along the direction of heat extraction. the governing equations are often specific to particular problems.3 Ginzburg–Landau equation and order parameter 13.2) This vc can be considered as Dm /xc . a longer time is required for the liquid and solid phases to develop their proper chemical compositions. In these simulations.3 Ginzburg–Landau equation and order parameter The Péclet number is defined as P= vr . but it misses the dependence of freezing on the temperature. and are an active field of research.1 Ginzburg–Landau equation The Ginzburg–Landau equation is often a starting point for phase field models − → F(η) = Aη2 + Bη4 + C(∇η)2 . The power-law dependence is not simple. The larger is vc . the smaller is xc . This may set a spatial length scale. These interactions through solute diffusion cause the primary spacing of dendrite arms to become larger with time. 2 Dm (13.319 13. and are worth careful examination. When this number is approximately 1.

The contours in the liquid phase denote solute concentration.1 t Fig. in the direction of heat extraction. G = 10 K/cm. (Courtesy of A.4 wt.4 0.4wt% acetone.) 17:31:09 . Tourret.3 0. V = 25 µm/s c/cl0 0. Karma and D. 13.2 0. simulated by a phase field model. The two solid phases form as dendrites that grow to the right.4 0 Time snapshots of dendritic growth during the solidification of a bicrystalline alloy.% acetone alloy in a fixed temperature gradient of 10 K/cm.1 mm SCN–0. moving at 25 micron/s. Parameters were for a succinonitrile–0. as indicated by the scale at far right.

2 Conserved and nonconserved order parameters The order parameter need not be a conserved quantity like composition. 12. This figure shows how the bulk part of the free energy varies with φ and c. T) dV . c. since one is conservative and the other is not. T) with the Euler equation.e. In brief: • The evolution of a conservative parameter. 13. both the Cahn–Hilliard equation and the Ginzburg–Landau equation include contributions from the square of the gradient in order parameter. (13. Equations governing the time evolution of conservative fields tend to be more complicated than equations for nonconservative ones.3 is of the form − → − → F(c. For nonconserved order parameters. i.6) ∂t ∂c 2 Equation 13. c. they have higher-order spatial derivatives. usually the kinetics are proportional to the steepest gradient in total free energy.18) that was used to discuss spinodal decomposition in Chapter 12. giving a total free energy of the form  − −→ 2 → 2 κc ∇c + κφ ∇φ + f (φ. T) of Fig. diffusion of atoms down a chemical potential gradient). c. 12. where ∂c/∂t equals the (negative) divergence of flux. and the field variables follow this path directly. Nonconserved order parameters have different governing equations than conserved order parameters [141]: • Nonconserved order parameters avoid the need for Lagrange multipliers that provide constraints such as conservation of solute (Sect.. suppose η is a magnetic order parameter. (13. ∂FT /∂c. the square gradients of c and φ are not treated in the same way.3 Ginzburg–Landau equation and order parameter expressing the free energy density in terms of an order parameter η and its gradient. c for example. The flux. • Nonconserved order parameters do not use a diffusion equation to follow the time evolution of the field quantity (e.48. (13. The order parameter η could be a conserved quantity like composition. For example. c. as usual.3.2). this is the Cahn–Hilliard equation (12.g. allowing a phase field model to include interfaces and other heterogeneities. 13..5) FT (φ.2 depends on three fields that can have some independence.321 13. Nevertheless. The magnetic spin alignment is not a conserved quantity – the magnetic alignment can change in one region without compensation in another region. 17:31:09 . There are also energy costs for sharp variations in these quantities – gradients in both composition and phase fields will have such terms. is Mc times the (negative) gradient of the chemical potential. for which Eq.4) When integrated over volume of the material. 13. ∇c) = f (c) + κ(∇c)2 .3.6 is a diffusion equation. In general form it is2    ∂c ∂FT  = ∇ · Mc grad . The function F(φ. was developed in differential form in Eq. T) = vol For minimizing FT (φ.

2. 7.1 to calculate the width of a flat interface between ordered domains. the energetic relationships between the different field quantities. c. c. Two ordered domains are shown with an antiphase domain boundary (APDB) between them in Fig. but beyond the scope of the present text.3 Extensions to multiple fields Other field quantities can be included in a more complete model of a material. and in Sect. 13. and may alter these other fields because strain energy is often large. 13. Specifically. The governing equations for equilibrium structures are therefore not practical for analytical solutions. Furthermore.4 Interfaces between domains We first use the Ginzburg–Landau equation to calculate an equilibrium structure.5 shows this situation. 13. Domain growth is a counterpoint to the problem of spinodal decomposition in Chapter 12.4. at least at intermediate temperatures. as does the problem of chemical unmixing. is simply proportional to a mobility times the effect of φ on the free energy ∂FT ∂φ = −Mφ . Magnetic fields.2 is abrupt.5.322 Phase field theory • The evolution of a nonconservative parameter.1 Interface structure The boundary in Fig. Strain fields. which are tensor fields. 13. also follow variations in the other field quantities. so the areas of the two domains are not conserved. it is used in Sect. which are vector fields. and they are coupled through f (φ. Ordered domains are a good example of a field with a nonconserved order parameter. 13. one domain can grow at the expense of the other.and B-atoms near the boundary. but in general there will be some disorder around the boundary. Figure 13. these spatial variations in the different types of fields generally do not track each other in straightforward ways. vary with φ.3. T) in Eq. φ for example. Loca− → tions with the highest square gradient energies (∇ξ )2 of different fields {ξ } may not be co-located. These dynamics of boundary motion do not require long-range diffusion. which used a diffusion equation because composition is a conserved order parameter. 7. By interchanging pairs of A.4. where a 17:31:09 . such as a magnetoelastic energy. In a complex microstructure.7) Both ∂c/∂t and ∂φ/∂t must be integrated simultaneously to follow the time evolution of the system. 13. ∂t ∂φ (13. and T. and the time-dependence of the evolution of a complex microstructure invariably requires a computer simulation.2 the time evolution of a three-dimensional interface is calculated. and then to discuss the dynamics of a field of a nonconserved order parameter. are not necessarily linear in the field quantities.4. These are topics of current research.

12. domain on the left has an order parameter of –0. ∇η). 12.8 would be approximately T = 0. but it does not alter the free energy.4.8.0 –10 t Fig. The Euler equation for our antiphase domain boundary problem. The Ginzburg– Landau equation (13.31.4 Interfaces between domains 323 LRO parameter (η) 1.0 –0.   − →  f (η) + κ (∇η)2 − f (η0 ) dV . based on Fig.78 T . The order parameter undergoes a transition across the APDB. is ∂fi d ∂fi =0. FAPDB = FAPDB = vol  ∞ −∞  f (η) + κ dη dx (13.e.3 The sign of the order parameter is useful for identifying the domain. Away from the boundary. (13.5 –5 0 Position (x) 5 Variation of order parameter across antiphase domain boundary at zero position.12) dx 3 The temperature for L = ±0.. a chessboard with its black square on the upper left has the same number of black–white pairs as a rotated chessboard with its white square at the upper left. i.9) 2 dx . the order parameters in the two domains have the equilibrium values |η0 | for the temperature of interest. (13. 7. This is given by the difference FAPDB = FT (η) − FT (η0 ) .11) The Euler equation for minimizing a free energy that depends on a field quantity and its − → gradient.8) (13. f (η. What is of interest is the cost in free energy of the interface.5 0. The goal of this section is to calculate the equilibrium width and profile of the transition region.1.10) where the last equation was a transition to one dimension. − ∂η dx ∂ dη (13.13. 17:31:09 . and f (η) ≡ f (η) − f (η0 ) .4 Therefore the free energy of the bulk material can include only even powers of the order parameter. was discussed in Sect. c 4 For example.3) has the right form.5 –1. the equivalent of Eq. not the bulk free energy FT (η0 ) at the equilibrium value of η0 . 13.3.0 0.8 and the domain on the right has an order parameter of +0.

23) FAPDB = dη −ηeq 17:31:09 .19) (13. (13.15) (13.10.17) The following relationship is useful for rewriting the integrand of Eq. 13. a minimum of FAPDB is found by applying Eq. Following the prescribed steps from the calculus of variations.10)  ∞ FAPDB = 2 f (η) dx .12 to the integrand of Eq.Phase field theory 324 where fi is the integrand of Eq. dx dx  2 dη . giving a condition on f (η) d dη ∂f (η) − 2κ =0 . ∂η dx d2 η ∂f (η) = 2κ 2 . f (η) = κ dx and dη = dx $ f (η) . (13. f (η) = 2κ dx2 It is more convenient to change to integration over x  2 d η dη dx . f (η) = 2κ dx2 dx (13.14) (13.17 d dx  dη dx 2 =2 dη d2 η .20) (13.22) −∞ It is more convenient to integrate over the range of the order parameter η. 13.16) (13. 13. 13.17 becomes      d dη 2 f (η) = κ dx .20 is used to rewrite our expression for the minimum change in free energy caused by the presence of an antiphase domain boundary (Eq.21) Equation 13.18) so Eq. ∂η dx dx d2 η ∂f (η) − 2κ 2 = 0 . ∂η dx An expression for f (η) is found by integrating over η  2 d η dη . 13. κ (13. instead of the infinite range of x  ηeq dx f (η) dη . 13.10. dx dx2 (13.13) (13.

11. (13.4 Interfaces between domains 325 3.21  FAPDB = FAPDB = ηeq −ηeq  ηeq $ f (η) ) −ηeq κ dη . f0 dη = dx 17:31:09 (13. we need an expression for f (η) of Eq. 13. (13. and a local maximum of +f0 at η = 0. 13. f (η) κ f (η) dη .25) Not surprisingly. Eq. The following form for this bulk part of the Ginzburg–Landau potential has the right properties       η 4 η 2 + . has two minima at zero for η = ±ηeq .26.24) (13.6 –1 0 η / ηeq 1 2 Bulk part of the Ginzburg–Landau potential for antiphase domain boundary. 13.0 –2 t Fig.26) f (η) = f0 1 − 2 ηeq ηeq This potential.0 0.0 Δf (η) 2.5 2.28) (13. (13.5 0. The square-root dependence on κ reflects the possibility of altering the width of the boundary for different κ. 13. the free energy cost of an APDB increases with κ.29) (13.30) .21 allows us to get an expression for the characteristic width of the APDB  $     1/2 f0 η 4 η 2 dη + . 13. dx = f0 $ κ x= ηeq . and using Eq. 13.5 Δ f0 1. as shown next. which gives the strength of the square gradient term.0 1. shown in Fig. For further progress. Using it in Eq.27) = 1−2 dx κ ηeq ηeq In the limit of small η $ f0 .13. κ $ κ dη .6.

η(x)   $ f0 1 x .27 $   1 f0 dx . 12.5 (see also [142]). obtained by integrating Eq. one domain grows at the expense of the other.326 Phase field theory where x is a characteristic width. the “conservative dynamics” of spinodal decomposition is the more difficult problem – the constraint of constant composition required a Lagrange multiplier in Sect.3. 13. depends on the relative magnitude of the square gradient energy through κ. 13. (13.2. the interface width increases in proportion to the value of ηeq . the interchange of neighboring A. without the square gradient term) favors as much volume as possible with the full equilibrium value of ηeq . another case of a nonconserved order parameter. The width increases with the square gradient energy (larger κ).4. which brings its own issues. 13. reducing the area of domain boundaries to minimize the interface energy. whereas ordered domains are not. The kinetics were driven by atom diffusion down the gradient of the chemical potential. 5 Atom movement is likely through a vacancy mechanism. The characteristic width of the APDB.e. (Finally. A complete solution for the shape of the APDB can be obtained by rearranging Eq. The important difference is that chemical composition is a conserved quantity. η(x) = ηeq tanh ηeq κ (13.32) = ηeq κ which gives immediately the profile of the antiphase domain boundary. In both cases. and hence favors a narrow interfacial region.2 applies the same reasoning to the width of an interface between magnetic domains. the mechanism involves A.) Section 17. so a diffusion equation had to be adapted for the kinetics of spinodal decomposition. Actually. but the scary integral evaluates to a clean analytic result   $ f0 η −1 ηeq tanh x.2 Interface kinetics Chapter 12 developed the kinetic theory of spinodal decomposition with a formulation analogous to what we will use here for the kinetics of domain interface motion with phase field theory. This profile is physically reasonable.and β-sublattices does not conserve the sizes of the ordered domains. and the magnitude of the bulk free energy through f0 . On the other hand.8.5 and the numbers of atoms are unchanged. because a wider interface minimizes the gradient energy cost.31) dη =   2  4 1/2 κ η η 1 − 2 ηeq + ηeq The integral on the right is trivial. (13. x.and B-atoms exchanging positions. 13.33) This is the profile shown in Fig. the difference in bulk free energy (the f0 . for the same gradient dη/dx. which became a chemical potential (i. Over time. 17:31:09 . the free energy per atom)..29 over the range of −ηeq < η < +ηeq . On the other hand.and B-atoms between the α.

To study dynamics.36) (13.4 Interfaces between domains 327 For the “nonconservative dynamics” of antiphase domain boundaries. but it is nevertheless useful to confirm that a flat APDB does not move. 13. of course. This allows us to use some of our prior results from one-dimensional analysis of 6 A curved interface is described by a functional dependence of one coordinate on another. 13. (13. This is correct because in one dimension the free energy does not depend on the position of the APDB. for developing a kinetic theory the essential idea is that the free energy decreases with time in proportion to the mobility. M ∂η = −M ∂t  δFAPDB δη  .10. we use results from the previous section to minimize the variation of FAPDB with respect to the variation in η   ∂fi d ∂fi − =0 ∂η dx ∂ dη dx  ∞  2 dη ∂η dx = 0 = −M2κ ∂t −∞ dx ∂η = −M ∂t (13.14 is ∂f δFAPDB = − 2κ∇ 2 η . As discussed previously in Sect.34 using a curvilinear coordinate χ that is perpendicular to the APDB.34) The previous section used this equation in one dimension. it is natural to use a spherical coordinate system with r normal to the surface of the domain to analyze the effect of curvature on growth velocity. for example.13. 7 For a spherical domain. For kinetics.6 All interfaces in one dimension are flat. we need to align a variable like x along the direction normal to the APDB.22.37 was obtained from results of Eqs. which has no thermodynamic tendency to move forward or backward. 13. minimizing Eq. 13. This cannot be done with one coordinate. In this case we set to zero the time-dependence for the relaxation of FAPDB .37) where Eq. this corresponds to a flat interface.6. The result is zero. however.12) directly. δη ∂η (13. In three dimensions.1 provides a mathematical result for the Laplacian of Eq. The natural coordinate system follows the shape of the three-dimensional APDB. In both cases. We work with the Euler equation (13.35) For our one-dimensional problem.7 The text in Box 13.20 and 13. 17:31:09 . we impose no constraint on ordered domains that requires a Lagrange multiplier. 13. we need to work in two or three dimensions. The three-dimensional analog of Eq. 3. which also pertains to a flat interface in three dimensions. setting it equal to zero to identify the minimum of FAPDB for a one-dimensional APDB. a one-dimensional model has limited value because there is no concept of curvature in one dimension. and determine the velocity of the interface along this direction.

In curvilinear coordinates we can likewise adapt expressions for the gradient and divergence that are familiar from Cartesian coordinates 1 ∂η ˆ 1 ∂η 1 ∂η − → ∇η = m ˆ + nˆ . so the Laplacian depends only on the coordinate l   1 ∂ hm hn ∂η . The Laplacian can be written as →  ·− ∇η .42) ∇ 2η = hl hm hn ∂l hl ∂l ∂m hm ∂m ∂n hn ∂n As a check. we arrange the distance dχ as normal to the surface of the APDB. 13.1 Some results in curvilinear coordinates In any curvilinear coordinate system with the coordinates l. ∇ hl hm hn ∂l ∂m ∂n (13.46) We obtain the result ∇ 2η = ∂ 2η ∂η  ∇ · χ . for a Cartesian system hl = hm = hn = 1. (13.38) ∂l ∂l ∂l so for a small change of dl in l. For our APDB geometry we need a curvilinear coordinate system with one coordinate pointing normal to the surface of the boundary.39–13. which we choose as l. n. hl hm hn ∂l  . are unit vectors along the coordinates l. (13. m. For example. + 2 ∂χ ∂χ 17:31:09 (13.42 assume their familiar forms. m. There are similar expressions for hm and hn .39) (13. l+ hl ∂l hm ∂m hn ∂n   ∂ ∂ ∂  · n = 1 (nl hm hn ) + (nm hl hn ) + (nn hl hm ) . along the l-direction we define %  2  2  2 ∂x ∂y ∂z hl ≡ + + .45) (13.328 Phase field theory Box 13. (13. we can calculate incremental lengths along the coordinate directions with the Pythagorean theorem and incremental lengths in a Cartesian system. (13. and Eqs. the incremental change of distance is ds = hl dl. dχ = hl dl and the Laplacian is   ∂η 1 ∂ hm hn .40) where ˆl. n.47) . The other two coordinates are in the plane of the boundary. That is. ˆ nˆ .41) ∇ 2η = ∇        ∂ hl hn ∂η ∂ hl hm ∂η ∂ hm hn ∂η 1 + + . m.43) ∇ 2η = hl hm hn ∂l hl ∂l For consistency of notation with our one-dimensional problem of the APDB. the divergence of a vector along the l direction is   1 ∂  ∇ · χ = (hm hn ) .44) ∇ 2η = hm hn ∂χ ∂χ The differentiation is worked as usual ∇ 2η = ∂ 2η ∂η 1 + ∂χ 2 ∂χ hl hm hn  ∂ hm hn ∂l Likewise. (13. (13.

on the other hand.4 Interfaces between domains an APDB along x to develop the Allen–Cahn theory of APDB dynamics [143]. and dη/dt is for a fixed APDB orientation. δη ∂χ (13. v.51) A velocity of APDB motion. 13. For a sphere. 13.52) dχ giving the simple expression for curvature-driven interface motion. as shown in Fig.50) where K1 + K2 is the mean curvature from the two orthogonal directions on the surface of the APDB. 13.48) δη ∂η ∂χ ∂χ The first two terms in square brackets in Eq. we substitute into our kinetic equation (13. 13. For a saddle surface. the two curvatures along different coordinates have opposite signs. however. 13.8. 17:31:09 . stating that a small domain of high curvature will shrink most rapidly. Monte Carlo simulations of ordered domains in the form of Fig. using the result from Eq. for example. and zero velocity of the domain boundary.14 this bracketed quantity is zero. as discussed next.48 pertain to a flat interface with a normal along χ. Domain evolution in three dimensions is not simple because the interface velocity depends on the mean curvature in Eq.53. so δFAPDB ∂η  = −2κ ∇ · χ .53) This seems at first an intuitive equation. (13. What is interesting in three dimensions (and not in two) are topological structures such as shown in Fig. It is therefore possible to have K1 = −K2 . Less intuitive is the definition of a mean curvature in three dimensions.47   δFAPDB ∂f ∂ 2η ∂η  = − 2κ 2 − 2κ ∇ · χ . is defined as dη dt v ≡ − dη . All points on this surface are saddle points having zero mean curvature. Such a microstructure may have a high free energy. Returning to Eq. 13. but it is reasonably stable against growth.35) to obtain ∂η dη = −M  (K1 + K2 ) . Since both are subject to change during microstructural evolution. K1 + K2 = −∇ (13. 13. the velocity v is an instantaneous one.  By Eq.7.329 13.7 show that small thermal fluctuations of the surface are restored. and the structure is metastable. (13. (13. A large fluctuation that causes two nodes to merge will lead to collapse of the entire structure. 13. ∂t dχ (13. 8 It should be noted that the dη/dχ is for a fixed time.34.49) Finally. so K1 = K2 .8 v = M  (K1 + K2 ) . in all directions on the surface the curvatures are equal. It is related to the divergence as  · χ . and perhaps our intuition is based on growth and shrinkage of spherical particles as in nucleation.

13. At higher concentrations. c. caused by coalescence of units in the upper center. a periodic minimal surface with zero mean curvature everywhere.2 to vary with temperature? If there in the liquid above the melting temperature. discuss  is no chemical unmixing  2 2 how ∂ F(φ. 1. Problems 13.7 t Fig. and some of the arguments for the energy of APDBs should apply to this unmixing problem even though the composition is a conserved quantity. Section 11.9 described the coarsening of precipitate microstructures. especially around equiatomic compositions. or have at least partial cancellation of their principal curvatures.8 Schwarz P-surface. T)/∂c may vary with φ. Many of the points on the isoconcentration surface appear to be at saddle points of the interface.198. In alloys of higher concentration. For example. 13.Phase field theory 330 t Fig. but considered only a small fraction of precipitate phase. T 17:31:09 .” the solute atoms in a random solid solution make a connected network all the way across an infinite crystal.245 [144]. These topological features of the boundary structure likely play a role in setting the kinetics of coarsening. and therefore prominent features of it.1 How would you expect the free energy curves in Fig. 13. for an fcc crystal the percolation threshold is c = 0. the interfaces between the unmixed phases will have more complicated curvatures. it is therefore unrealistic to model unmixed regions as spheres. Notice again the structures of Fig. Initiation of collapse of a periodic minimal surface in a Monte Carlo simulation. The lower driving force for moving saddle surfaces may cause these parts of the interface to be relatively persistent in an evolving microstructure.6b for chemical unmixing. and for a bcc crystal it is c = 0. At higher concentrations above a “percolation threshold.

(b) What is the functional form for how the radius of the shrinking domain changes with time? 17:31:09 .47 to 13. 13.331 Problems √ 13.2 Explain why the free energy cost of an APDB (cf. (a) Using expressions for the Laplacian and divergence in spherical coordinates. 13.25) varies as κ. 13.49 to explain the shrinkage of the antiphase domain with time.3 Consider a large ordered domain that contains a small sphere of an antiphase domain. develop the equations analogous to Eqs.

such as sound waves in a solid.14 Method of concentration waves and chemical ordering This chapter analyzes the thermodynamic stability of “static concentration waves. so it is called a “static” concentration wave. Waves of very small k correspond to macroscopic wavelengths. which is an important example of how Fourier transform methods can treat long-range interactions in materials thermodynamics. For k = 0 there are no relative displacements of adjacent atoms.1 First Brillouin zone Here is a brief review of reciprocal space.e.. so the amplitude of the concentration wave.1 shows how a row of atoms responds to waves of different wavevector. the demand for translational symmetry of the free energy sets a condition for the vectors in the star of the ordered structure known as the “Landau–Lifshitz criterion.1 Structure in real space and reciprocal space 14. A convenient feature of this approach is that an ordered structure can be described by a single wavevector. and this variation can be written as a wave. so we do not consider the intermediate phases of the concentration wave. at least not in our main examples. These methods of static concentration waves were developed by Armen G. η. serves as a long-range order parameter. k. 17:32:20 . with crests denoting B-atoms and troughs the A-atoms. If a second-order phase transition is possible (i. which gives a wave of infinite wavelength (k = 2π/λ. the ordered structure evolves from the disordered with infinitesimal amplitude at the critical temperature). Perhaps the simplest case is k = 0 near the middle of the figure. 148]. where λ is wavelength). or a small set of wavevectors. for example. A key step for phase transitions is writing the free energy in terms of the amplitudes of static concentration waves. Khachaturyan [145]–[147] and are explained in his book and review article [134.” Chapter 14 ends with a more general formulation of the free energy in terms of static concentration waves. 14.” The idea is that an ordered structure can be described as a variation of chemical composition from site to site on a crystal lattice. Another important difference from conventional waves is that the atom sites are exactly on the tops of crests or at the bottoms of troughs. The disordered solid solution has no such periodicity. Figure 14. and then explains the “star” of the wavevector of an ordered structure.1. This chapter begins with a review of how periodic structures in real space are described by wavevectors in k-space. and no wave. The wave does not propagate.

1 Structure in real space and reciprocal space k = 2π/a k = π/a First Brillouin zone k=0 k = –π/a k = –2π/a t Fig. Note carefully how the atom displacement pattern in the second wave from the top in Fig. so the result is no mutual displacement between adjacent atoms.” For all analyses 1 Sometimes these large k-vectors are used to account for momentum transfers with the center-of-mass of the crystal.1. When k becomes larger. i. the wave patterns repeat in reverse order from the sequence of k from 0 to π/a. a region called the “first Brillouin zone.333 14.1 Atom displacements for different wavevectors. As shown in Fig. the two waves that give the identical wave of intermediate atom displacements. i. however. Please identify the three waves that give zero relative displacements of atoms. The range of the first Brillouin zone.1 A more subtle point is that for k from π/a to 2π/a.. 14. and not the wavevector parameter itself. Of course only the physical atom displacements are important. and check the effect of changing the sign of k. λ = 2a. k = 2π /a. and the picture is equivalent to that for k = 0. There is no wave. 14. as labeled at left.1 is identical to the wave two below it. What happens at twice this value. but are otherwise similar.. the condition k = ±π/a gives the atom displacement pattern with the shortest wavelength that can be sustained on a onedimensional chain.e. so they are mutually out-of-phase. They have k of opposite sign. All wavelike atom displacements can be accounted for with the wavevectors between −π/a < k < +π/a. Here the wavelength is matched exactly to the atom periodicity.1. k. A wide range of k therefore gives a redundant description of the atom displacements. 14. 14. so the atom displacement pattern has a smaller λ.1 show a typical sinusoidal pattern of atom displacements.e. 17:32:20 . The two waves adjacent to the null wave of k = 0 in Fig. 14. is shown at the top and bottom of Fig. −π/a < k < π/a. when k = ± 2π /2a = ±π/a. is identified at right. but these issues do not arise in what follows. an issue arises when λ exceeds twice the interatomic spacing.

Importantly. aˆy. but for higher dimensions the higher Brillouin zones do not have the same shape. g. l} are integers.1) where {m. for which {a1 . as demonstrated with Fig.5) The problem is particularly easy for simple cubic crystals. we will consider the wavevectors k = +π/a and k = −π/a to be identical because they give the same physical effects. a3∗ } are obtained from the real primitive translation vectors {a1 . We can tile an entire crystal in r-space by covering it with translated unit cells.3 to 14.4) (14. The numerators in Eqs. o} are integers.8) 2 This analogy works very well in one dimension.2 A characteristic of the k-space structure of a crystal is that the origin at k = 0 is repeated at intervals of k = 2π/a in k-space. a2 . The reciprocal primitive translation vectors {a1∗ . but a special one that gives a periodicity of the crystal. a reciprocal lattice vector. a 2π zˆ .3) (14. k. a3 }  = ma1 + na2 + oa3 . These points separated by multiples of 2π/a comprise the “reciprocal lattice” of the crystal.2) where {h. but the structure is set by the content of the first Brillouin zone. a3 } by these definitions a2 × a3 . A reciprocal lattice is generated from a set of primitive translation vectors in k-space. R is generated with the primitive lattice translation vectors {a1 . R (14. a2 . and the denominators are a3 (without a direction) 2π xˆ . which have units of length−1 . a2∗ = 2π a1 · [a2 × a3 ] a1 × a3 . a3∗ = 2π a1 · [a2 × a3 ] a1∗ = 2π (14. is generated with its own primitive translation vectors {a1∗ .1. 14. n. a 2π a2∗ = yˆ . 14. 17:32:20 . a3 } = {aˆx. 14. the Brillouin zone plays a role in k-space that parallels the role played by the unit cell in r-space. we will work within the first Brillouin zone. a2 .1. (14. a1 · [a2 × a3 ] a3 × a1 . but the content is set by that of one unit cell.7) (14. a2∗ . a3∗ = a a1∗ = (14. although their volumes are the same. a3∗ } g = ha1∗ + ka2∗ + la3∗ . In the same way that a crystal lattice vector in real space. aˆz}.2 Reciprocal lattice For a crystal.334 Method of concentration waves and chemical ordering with concentration waves of atoms. . Each point of the reciprocal lattice is a wavevector.6) (14. Likewise we can tile k-space with Brillouin zones.5 are a2 (with a direction). a2∗ .

14. is an integer. 3 However. (This is the “Laue condition” for diffraction. however.3 To obtain the allowed diffractions of the bcc crystal.2 for a simple cubic crystal  ∗ ∗ ∗ e−ig·R = e−i(ha1 +ka2 +la3 )·(ma1 +na2 +oa3 ) . 14. o}. The Cartesian axis reference frame is more convenient.  −ig·R e  −ig·R e −i2π(mh+nk+ol) =e −i2π integer =e (14. There is a perfect correspondence to X-ray diffraction. 17:32:20 . 14.2 as the reciprocal of the bcc. such as (100) in both fcc and bcc.9) where the Kronecker delta δij equals 1 when i = j.11) = +1 .2. n. are consequences of using a simple cubic lattice for reference.1 Structure in real space and reciprocal space 335 For a simple cubic lattice. 14. so we are accustomed to using “structure factor rules.” which are derived for a cubic lattice with basis vectors. (14. look at its reciprocal lattice. The correspondence is illustrated in Fig.10) . the edge of the bcc unit cell 4 should be a.1 and 14.12 means that all wavevectors of Eq. are equivalent. to use the indexed fcc unit cell in Fig. Exponentials involving lattice translation vectors and reciprocal lattice translation vectors become quite simple. inner products of real and reciprocal lattice translation vectors benefit from the relationship ai · aj∗ = 2π δij . 14.2. k. or (110) in fcc. 14. rather than 2a. for example4 113 313 022 022 222 002 202 002 111 t Fig. Perhaps the most important is that the reciprocal lattice of a body-centered cubic (bcc) crystal is a face-centered cubic (fcc) crystal. evaluated by summing exponential phase factors over atoms at all  is equal to g. instead of the nonorthogonal primitive lattice vectors of the real space fcc and bcc structures. Eq. A few correspondences between reciprocal lattices and real lattices are worth knowing. and vice versa. will be strong whenever the diffraction vector k reciprocal lattice. (14. (14.2 000 202 121 311 020 220 200 011 020 000 222 112 101 110 211 220 200 Indexed bcc and fcc Cartesian unit cells. and is zero otherwise.) Not all reciprocal lattices are so obvious as the simple cubic reciprocal lattice of the simple cubic lattice. l). Using Eqs.14. The diffracted wave. For the periodicities of atom concentrations.12) where the sum of products of integers. which is fcc.2. The missing diffraction peaks. a vector of the {m. 14. denoted g = (h. mh + nk + ol. Allowed diffractions can be read from the indices in Fig. indexed with Cartesian axes in units of cube edge 2a to avoid half-integral indices.

the ordered structure will require information on atom occupancy for at least this number of sites.3 Ordered structures described by concentration waves Numerous ordered structures are found in nature5 . B32. Analogously. (bcc structure factor rule: the three indices sum to an even integer) fcc diffractions: (111). in the unit cell can be as few as two for a binary alloy (and in fact there are two rk for the B2 structure). . These {rk } are basis vectors of the ordered unit cell and give real space (r-space) information about which atom is placed where. (220). 17:32:20 . The number of atom position vectors. t Fig. 14. (211). 5 All simple structures that can be constructed with stoichiometric ratios such as 1:1 or 1:3 on fcc or bcc lattices are in fact found in nature.1.3 shows five ordered structures on fcc or bcc parent lattices. Perhaps not surprisingly. These are the wavevectors for the static concentration waves of the ordered structure. The chemical formulae are for “prototypes” of compounds that take these structures. Figure 14. L12 . Names of the structures in “strukturbericht” designation are D03 .1. Wavevectors such as (3. (200). which reveals additional wavevectors as “superlattice” diffractions in the diffraction pattern. (200). (311). A good way to measure the wavevector for an ordered structure is by X-ray diffractometry. but if the standard cubic unit cell requires more atoms (such as four for the fcc cell). L10 . low-order wavevectors such as (100) and (111) generate simple ordered structures such as L10 or B2. for example. (222) . It is convenient to describe an ordered structure by specifying a wavevector in k-space because a wavevector can account for the periodic alternation of the different atoms in the structure. In what follows we go the other way. and pick a wavevector first. and B2. 14.Method of concentration waves and chemical ordering 336 bcc diffractions: (110). rk . (220). (fcc structure factor rule: the three indices are all even or all odd integers).5) generate ordered structures with longer-range periodicities that are less common in nature.4.3 Five ordered structures on bcc and fcc parent lattices. . . We then find the ordered structure that it generates. The atom positions for four of them are listed in Table 14. (222) . . the wavevector can be obtained by performing a Fourier transform of the chemical modulation in the ordered structure.

0.x + krk.6 For all rk in Table 14.x xˆ + rk. this is easier than it may sound at first. provided it always evaluates to ±1 at all sites of the crystal. 1/2.x xˆ + rk. because it uses tricks like the one at Eq. 1. k.14) = +1 if integer is even . 1/2} Fe: a{1/2. a a Equation 14. the exponential in Eq.1 Structure in real space and reciprocal space 337 Table 14. 3/2.19) 6 It is sometimes also appropriate to add the complex conjugate to a wave to ensure that the concentration is real. 1/2} Fe: a{0. 0} Cu: a{1/2. rk. It is therefore often acceptable to use a complex exponential for a concentration wave. 1} Fe: a{0. 0} Fe: a{1. 1/2. 1/2} Fe: {0. 1/2. 0.z zˆ) = (hrk. 0} Al: a{1/2.14  2π a  rk −ik· = e−iπ integer .14.17) (14. (14. 0 } Cl: a{1/2. 0} Cu: a{1/2. 1/2} Cu: a{0.y yˆ + rk. 0} Cu: a{1/2.1. 1. 0} Au: a{1/2. k= rk. 1/2} L10 L12 D03 Au: {0.y yˆ + rk. (14.16) e−ik·rk = e−i e (hˆx + kˆy + lˆz)·(rk. 3/2} A k-space description of an ordered structure starts with the selection of a low-order r  and a wave e−ik· is constructed at r throughout the real space crystal. 1/2} Cu: a{0. 1. 14. 1/2. 3/2} Fe: a{3/2. 1/2} Al: a{3/2.z ) . 0.15) = −1 if integer is odd . 0. 3/2. 3/2. 3/2} Al: a{1/2.z zˆ) . (14. 1} Fe: a{1.y 2 h= (14. 1. 0. 1/2.13) In more detail. 1/2} Fe: a{3/2.14 will be true if a integer . 17:32:20 . 1/2. 0} Fe: a{0. 1/2. OR (14.18) (14. wavevector. 0. 14. 3/2.y + lrk. 1/2. The wave is evaluated only at specific atom sites. 3/2} Al: a{3/2.1 Atom positions in selected ordered structures B2 Cs: {0. 0. however.13 includes 2π 2π (hˆx + kˆy + lˆz) · (rk. 0} Fe: a{1. 1/2} Au: {0. 1/2. and the wave has no meaning between the atom sites. 0. 0. 1} Fe: a{1/2.x 2 a integer . 1} Fe: a{1.

1.2 Concentration waves with selected k-vectors B2 (111) 2π c + η2 e−i a (111)·rl Cs: c + η2 Cl: c − η2 L10 (001) 2π c + η2 e−i a (001)·rl L12 (001)+(010)+(100)  2π  2π 2π c + η4 e−i a (001)·rl + e−i a (010)·rl + e−i a (100)·rl Au: c + η2 Au: (+1 + 1 + 1) η4 = c + 34 η Au: c + η2 Cu: c − η2 Cu: c − η2 Cu: (+1 − 1 − 1) η4 = c − 14 η Cu: (−1 + 1 − 1) η4 = c − 14 η Cu: (−1 − 1 + 1) η4 = c − 14 η l= a integer .22) 2 These two equations were evaluated at all the {rl } of the B2 and L10 ordered structures listed in Table 14.x . so conveniently. 14. For example. Before calculating thermodynamic free energies. rk.x . for the B2 structure in Table 14. An fcc crystal is 17:32:20 . and results are listed in Table 14. the order parameter is zero above a critical temperature. 2. reflections. that are rational fractions of a. rk. The amplitudes of the concentration waves are given by the parameter η.z 2 (14. but L or η grows larger with temperatures below Tc .x = a/2.20)  There are other ways to ensure that Eq. Note in Table 14. the concentration wave has intermediate amplitude.x .18.14 is true (so that e−ik·rk = ±1. r2.2 Symmetry and the star The symmetry elements of a crystal typically include rotations.Method of concentration waves and chemical ordering 338 Table 14.21) cB2 (r ) = c + e−i a (111)·rk . and there is no modulation in composition. and the composition is real). 14. rk.1. 2 2π η cL10 (r ) = c + e−i a (001)·rk . the structures shown in Fig. which is an order parameter. h = integer. As was shown in Fig. When 0 < η < 1. Examples of concentration waves used to describe B2 and L10 ordered structures are 2π η (14. for example.3 are generated.2. and identify the minimum  number of k-vectors that define an ordered structure. and inversions. the same types of atoms are at the same positions in space. we develop further the ideas of crystal symmetry in k-space. When η = 1. When η = 0. the chemical composition is simply c throughout the crystal. but the standard practice in this chapter will be to work with a standard cubic unit cell of edge length a. (14. and Cartesian components rk. When the atom arrangements in an fcc crystal are rotated by 90◦ about a Cartesian axis. 14.2 that the crests and troughs of the real part of the concentration waves (giving the ±η) are at the sites of the two chemical species in the ordered structure.

One First Brillouin zones of square lattice (bounded by squares). 14. Lines of reflective symmetry (through center) are indicated in gray. Horizontal or vertical gray lines are reciprocal lattice vectors (2π/a)ˆx and (2π/a)ˆy.7 To be systematic. Four e. (f) Wavevector along symmetry line. An example is shown in Fig.4b and c.8 Generating the star for these three cases of Fig. Dashed lines denote wavevectors removed from the star because they create identical atomic structures as other wavevectors. touching special point at corner of zone boundary.4a. Given one wavevector. 7 This sounds nonintuitive to many people because ordered structures display prominent and beautiful symmetries. we collect all the unique wavevectors that generate the ordered structure. as shown next. a 90◦ rotation about the x-axis will change the atom arrangement into one with periodicity along yˆ . if there is a periodic arrangement of atoms along the (001) direction as in the L10 structure shown in Fig. 14.14. The formation of the L10 structure has caused the crystal to lose an element of rotational symmetry possessed by the parent fcc lattice. Two f. 14. the other wavevectors in the star are generated from it by use of the point symmetry operations of the parent lattice. star contains two vectors. On the other hand.4a–c was fairly intuitive. star contains four vectors. (a) Arbitrary wavevector. and we call this collection the “star”. touching special point at zone boundary. the others are obtained from it by symmetry operations.c).4a. star contains one vector (there are reciprocal lattice vectors on all four sides of the square). as in Figs. (c) Wavevector along symmetry line. Four d. (e) Wavevector along symmetry line. 17:32:20 . 14. where symmetry operations on a typical vector on a two-dimensional square lattice generate seven other vectors. star contains eight vectors. 8 Otherwise the number of vectors in the star will not be unique – numerous equivalent vectors can be created by adding reciprocal lattice vectors.3. 14. for example. Chemical ordering causes a loss of symmetry. (d) Arbitrary wavevector orientation. the reflection symmetry operation across the line generates the same vector. touching special point at zone boundary. Four c. There are eight vectors in the star of Fig. 14.2 Symmetry and the star 339 “invariant” under such a 90◦ rotation. star contains four vectors.4b. star contains four vectors. All unnecessary vectors must be eliminated from the star (giving four in Figs. 14. The initial vector is bold. Eight b.4 a. t Fig. (b) Wavevector along symmetry line. When the initial vector lies on a line of symmetry.

(111) Systematically subtracting the other seven vectors from the (111) gives (002).1. 14. (100) 17:32:20 . 14.2.4d and e. (111). The gray lines are reciprocal lattice vectors of length 2π/a. (111). (111). These translations must be considered in k-space. (100) = (100) The equality symbols mean that the two vectors differ from each other by an (002)-type vector. Therefore half of the six vectors are eliminated from the star. • For the (111) wavevector of the B2 structure. (111). We therefore say that wavevectors connected by the gray reciprocal lattice vectors in Fig.340 Method of concentration waves and chemical ordering Translational symmetry is a distinguishing feature of crystals. special points in the Brillouin zone. (100) • The L12 structure involves the same (001)-type vectors and the same fcc parent lattice as does the L10 . 14.1. which is a very special situation with only one vector in the star. Only one of them is included in our collection called the star. We typically use only wavevectors in a bounded region of k-space (the first Brillouin zone). so all eight variants of the (111) wavevector generate the same ordered structure. (222) Every one of these is a reciprocal lattice vector of the parent bcc lattice. as we saw in the discussion of the Brillouin zone in Sect. This eliminates half of the wavevectors in Figs. (022). Now enumerate systematically the stars for the ordering wavevectors presented in Table 14. first list all eight variants along different directions: (111). and k = +2π/a. 14. (202). Consider Fig. which is a reciprocal lattice vector of the parent fcc lattice. 14. (200). (010). the six variants are: (001) = (001). • For the (001) wavevector of the L10 structure. they specify identical atom arrangements. (020).4f. These special points come up frequently for ordered structures because ordering wavevectors are usually along directions of symmetry of the parent lattice. especially edges and corners. and for k = −2π/a.1 and Fig.4a were extended to the edges of the Brillouin zone. (111). There is only one vector in the star for the B2 ordered structure. for which (110)-type vectors are not reciprocal lattice vectors. Otherwise we will generate identical atom arrangements with wavevectors that have differ This same issue arises for ordered structures when their wavevectors are at ent values of k. Whenever two wavevectors are separated by a reciprocal lattice vector. and therefore has the same star: (001). 14. (101) The parent lattice of the L10 is fcc. allowing all atoms to be accounted for by lattice translations of a unit cell. 14. as for example in Fig. We conclude that there are three vectors in the star for L10 ordering: (001). k = 0. which span the Brillouin zone. (010). 14. (111).1 for k = −π/a and k = +π/a (separated by 2π/a). which we select as (111). Systematically subtracting the remaining two vectors from the remaining (001) gives (011). in which the vectors of Fig. (220). and eliminates three of the four wavevectors in Fig.4 are the same vector. and their lengths are often submultiples of the spacings in the parent lattice.4d. (010) = (010).

3 is to analyze the thermodynamics of second-order phase transitions. r  . n(r ) = c. r ) = D(r. =0 δ4F δ(r ) δ(r  ) δ(r  δ(r  ) (14. where the energy parameters are (14. r  ) (r)(r  )(r  ) 3!   r r r r r 1  + D(r. The difference in free energy between an alloy with a small concentration wave and a disordered solid solution with none is functional9 F((r )) = Ford − Fdis .28)  .3.25) r 1  + C(r. Our effort in Sect. r  . (14.24) is a of the concentration wave. r  . A(r ) must vanish. A concentration wave in an A–B alloy sets the probability for a B-atom on a site r as n(r ) = c + (r ) .3 The free energy in k-space 341 14. which is itself a function. 17:32:20 . Ford (c + (r )) − Fdis (c) = F((r )) . r  ) =     C(r. (14. for which we focus on small order parameters near the critical temperature.29)  .1 Using concentration waves to construct a free energy functional Compare the free energies of an alloy with and without a concentration wave. We therefore perform a Taylor series expansion around the disordered free energy.27)  . 14. r  ) =  δF δ(r )  .26)  A(r ) = B(r. r  . If thermodynamics 9 Again. r  ) (r)(r  )(r  )(r  ) 4!    r + ··· r .30) =0 Near the critical temperature for ordering.3 The free energy in k-space with concentration waves 14. a functional is a function of a function. r .  1  F((r )) = A(r ) (r ) + B(r.23) Without the concentration wave. (14. =0 δ3F δ(r ) δ(r  ) δ(r  ) (14. The difference in free energy. Here the free energy function depends on the concentration wave. or an increasingly large concentration wave (of the right sign) would be increasingly favorable.14. r  . r  ) (r )(r  ) 2!  r r (14. (r ). =0 δ2F δ(r ) δ(r  ) (14.

In an infinite crystal. and this will occur for the ordering wavevec however.e. so r  = r − R Define R 1   ) (r )(r − R ) . r − r  . r  ) and D(r. r  .33) (14. so shifts of the limits of the sum are not important when changing variables.Method of concentration waves and chemical ordering 342 favors a large concentration wave.10 Neglecting for a moment the higher-order terms gives 1  B(r − r  ) (r )(r  ) . 14. 2!  (14.26.34 is 1   r ∗ +ik·(  r−R )  ) η e−ik· B(R η e .  since then e−ik·r = e−i2π integer = +1 for all r (this includes k = 0). r ∗ ∗ +i k· to ensure reality. F((r )) = η∗ η 2! r F((r )) = (14.35) 2! r  R where all sums are over an infinite crystal. and Eq.31) F((r )) = 2!  r r The form for the concentration wave  (r ) = η e−ik·r (14. but we still need to be careful that the concentration is real.36) F((r )) = 2! r   R  Recognizing e−ik·r e+ik·r = +1. specifically B(r. the temperature is not near the critical temperature of a second-order phase transition. (14.37)  R N 2  |η| b(k) .32) has served us well. We know that (r ) will be real if k is a vector of the reciprocal lattice vector of the parent crystal. The form of Eq. allowing us to avoid the complex conjugates. we will focus on the problem for one ordering wavevector k. 14. Our plan is to first use the following form of F((r ))  (r ) = η e for a discussion of possible types of phase transitions at arbitrary k F((r )) = F((r )) = 1  B(r − r  ) (r )∗ (r  ) . It is also fine to  have shorter k if e−ik·r = e−iπ integer = −1. exponentials are ±1.38) 10 The absolute value of  r cannot have physical importance because the origin could be anywhere. i. 2 (14. We call on the complex conjugate tor. There can be problems with arbitrary k. 14. (14. 2!  r r r r 1     B(r − r  ) η e−ik·r η∗ e+ik·r . making it easier to work with the higher-order terms with C(r. the first sum over N atoms is simplified  1  R   ) e−ik· (+1) B(R .31 becomes  = r − r  . F((r )) = B(R (14. can depend only on the distance between r and r  . a function of two positions. however.34)  for which the Later. 17:32:20 .  . r  ) in Eq. r  )..

5 The function b(k) near the critical temperature. where there is some region in k-space with negative b(k). we know that the disordered solid solution is stable. Figure 14. (This is also true for an ordering alloy above its critical temperature in Fig.5a.and B-atoms.38 is positive. (14. Note the zero crossing of the lowest point on b(k) for the two cases: (a) b(k) for unmixing alloy. alloy. Spinodal decomposition typically occurs at temperatures  The thermodywell below Tc . and with decreasing temsame reason.3.5a. Below  is negative for the critical temperature.5a shows how  to change with temperature for an unmixing alloy. Eq. We now consider a few possibilities for  and their temperature dependences. 14.3 The free energy in k-space  is defined as the Fourier transform of B(R ) where the important quantity b(k)  R   =  ) e−ik· b(k) B(R . so the wavevector for fastest unmixing will tend to become larger with increased undercooling. which give chemical unmixing of the A. t Fig. 14.39)  R  14. 14. the b(k) With larger k the atoms need to move only short distances. 17:32:20 . with lowest point at k = ±k0 . 14. whereas the thermodynamics favor smaller k. F ord pertains to an alloy with a nonzero concentration wave of k = 0. With deeper undercooling. such as the temperature just above Tc labeled Tc in Fig. namic preference is for smaller k to minimize the gradient energy and strain energy in the  is increasingly negative over a wider range of k. Tc . respectively. the shapes of b(k) In Chapter 12 we analyzed the case for alloy instability against long wave modulations. (b) b(k) for ordering alloy.343 14. At high temperatures we expect b(k) above the critical temperature. The origin with b(0) = 0 is labeled with “0. Tc .” Temperatures Tc+ and Tc− are just above and below Tc . The minima of b(k) icities that are most favorable thermodynamically. so  F((r )) = Ford − Fdis of Eq. 14. 11 For unmixing. with lowest point at k = 0. 14. the minimum in b(k) perature this minimum comes down to touch zero at the critical temperature Tc .38 shows how b(k)  is a k-space function that depends on depends on atom arrangements in the alloy.) For unmixing.11 Since |η|2 is positive. b(k)  some region of k around k = 0.2 Physical interpretation of b(k)  contains all information about how the free energy For small η. Kinetics tends to favor a chemical modulation with the largest k. above Tc the b(k) +  is therefore positive for all k. such as at the temperature Tc− in Fig. The b(k)  give the chemical periodperiodicities of concentration modulations.5b for the  is at k = 0.

38. there is a discontinuity in the entropy and a discontinuity in the first derivative of the total free energy with respect to temperature (cf. so a continuous transformation seems possible.40) k0 ∂ k  The places to look for such a condition are along lines of symmefor any direction of k. Their full enumeration is beyond the scope of this text. one for the new phase and one for the parent phase. We  seek a point k0 that gives an extremum of b(k) ∂  b(k) = 0 (14. and the wavevector remains independent of temperature. Although there is no discontinuity in the total free energy function at the critical temperature Tc . didates for minima of b(k).38. 14. so nonzero values of η will be expected in equilibrium because this lowers the alloy free energy in Eq. For spinodal decomposition. concentration waves of wavevector ±k0 give negative b(k0 ). although only two are shown in this one-dimensional k-space.4. and they are candidate wavevectors for ordering on the crystal lattice.5b shows how we expect b(k)  alloy. 14. we say that 17:32:20 . of two lines of symmetry. Nevertheless. consistent with Eq. Fig. 14. There There are “special points” in k-space where we expect b(k) are several such points for a crystal structure of high symmetry. so a concentration wave at Tc+ will cause an increase in the alloy free energy. look at the very “special points” where there is a crossing in b(k). the wavevector tends to lock to the periodicity of the parent crystal. The value of b(k) reflected across such a line of symmetry. The line therefore defines a valley (or mountain)  To ensure an extremum. There are minima in b(k) for all wavevectors in the star of the ordering wavevector. Recall that for nucleation and growth we used two separate free energy curves.2).344 Method of concentration waves and chemical ordering  to change with temperature for an ordering Figure 14. such as the lines of reflection in Fig.4 Symmetry invariance of free energy and Landau–Lifshitz rule for second-order phase transitions We now discuss whether an ordering transition can occur as a continuous second-order phase transition. At Tc− . our analysis predicted an infinitesimal amount of unmixing at Tc . again b(k0 ) > 0 for all k0 . 14. 14.  must be the same when try. 1.3 Special points  to be an extremum. but it is not hard to understand the essential idea of where to look for them. when ordering occurs at special points. With no discontinuity in the first derivative of the free energy. or if it must be a discontinuous first-order transition. This discontinuity could be eliminated if there were a continuous variation of structure from the parent phase into the new phase. At Tc+ . The difference is similar to the difference between spinodal decomposition and nucleation and growth. but these points are can Not all ordered structures occur at special points. k0 . Of course there could also be a maximum or a saddle point at these special points.3. A minimum exists at the crossing of two valleys.

26 F − Fdis = star N   b(k0j ) |η(k0j )|2 2 j + 1 3! star  C(k0j1 .1 Concentration waves and lattice translations Consider the free energy in the presence of only one type of concentration wave. k0j3 . and r  (and includes the exponentials). 14. 14.29 by summing over all r. 14. k0j2 . k0j3 ) is related to C(r ) of Eq. k0j4 ). Our interest is in η at temperatures near the critical temperature.4 Symmetry and second-order phase transitions 345 spinodal decomposition is a “second order” transition.45 shows how to account for the lattice translation by a simple substitution   η → η e−ik0 ·T .42) Rewriting (r ) of Eq. which shifts the concentration wave F((r )) = F((r + T )) .44) (14. We focus on the wavevector k0 for which b(k0 ) = 0 when the temperature falls to Tc .j4 where C(k0j1 . and again with a translation by T  (r ) = η e−ik0 ·r (14.5b has two vectors in the star at ±k .j3 star 1      D(k0j1 . and similarly for D(k0j1 . whereas nucleation and growth is “first order.46) .45) Comparing Eq. A fundamental physical requirement is that F is invariant after a lattice translation T .j3 . 14. 14.41 12 The one-dimensional axis of Fig.14. That is. k0j2 .32 into the free energy expression of Eq.5b. 0 17:32:20 (14. k0j2 .4.32. k0j3 . . + 4! (14.j2 . k0j2 . the b(k0 ) will vanish for all vectors in the star of k0 . . 14. (14. r  . assume we already know the type of ordered structure that forms from the disordered solid solution.43 and 14.j2 .43) −ik0 ·(r+T ) (r + T ) = η e      (r + T ) = η e−ik0 ·T e−ik0 ·r (14. 14.” This section develops an elegant rule to predict if there must be a discontinuity in order parameter at Tc . Actually. k0j4 ) η(k0j1 )η(k0j2 )η(k0j3 )η(k0j4 ) + . k0j3 ) η(k0j1 )η(k0j2 )η(k0j3 ) j1 . which is easily done for the free energy expression of Eq.12 After substituting the concentration wave expression Eq. 14. so the picture is much like that of Fig.41) j1 . 14.

14. k0j3 ) e−i(k0j1 +k0j2 +k0j3 )·T η(k0j1 )η(k0j2 )η(k0j3 ) 3! j1 . k0j3 ) η(k0j1 )η(k0j2 )η(k0j3 ) 3! j1 .Method of concentration waves and chemical ordering 346 F − Fdis star N   = b(k0j ) |η(k0j )|2 2 j + star 1      C(k0j1 .j3 + 1 4! star       D(k0j1 . . Possibility I gives.47 must be equal.j2 .4. both at Tc and near Tc . (14. { k0j1 + k0j2 + k0j3 = g.j3 .j4 × η(k0j1 )η(k0j2 )η(k0j3 )η(k0j4 ) + . Eqs. k0j2 . At Tc . .j2 . with b(k0 ) = 0 at Tc . k0j3 . so we examine the higher-order terms in Eq. F − Fdis = 0 (14. k0j2 . (14. k0j4 ) e−i(k0j1 +k0j2 +k0j3 +k0j4 )·T j1 . { e−i(k0j1 +k0j2 +k0j3 )·T = 1 AND      e−i(k0j1 +k0j2 +k0j3 +k0j4 )·T = 1 } OR II. k0j2 .2 The Landau–Lifshitz criterion Because crystals are invariant against lattice translations. AND k0j1 + k0j2 + k0j3 + k0j4 = g } where g is a vector of the reciprocal lattice of the disordered alloy. with b(k0 ) = 0 at Tc . { C(k0j1 .49) 17:32:20 . k0j2 .j2 . k0j3 . k0j4 ) = 0 or > 0 } Possibility I is the same as I.47) 14.48) when k0j1 + k0j2 + k0j3 = g. 14.j4 × η(k0j1 )η(k0j2 )η(k0j3 )η(k0j4 ) + . Two important possibilities are:     I.41 and 14. k0j4 ) j1 . b(k0 ) = 0. F − Fdis = star 1  C(k0j1 . k0j2 .j3 .j3 + 1 4! star  D(k0j1 . k0j2 .47.j2 . . . AND k0j1 + k0j2 + k0j3 + k0j4 = g Possibility II gives. k0j3 . k0j3 ) = 0 AND D(k0j1 .

4 Symmetry and second-order phase transitions Box 14. The equilibrium amplitude of the concentration wave η(k0 ) must go to zero continuously – the equilibrium free energy is unstable against an infinitesimal concentration wave at a temperature infinitesimally below Tc . Possibility II does not forbid a second-order phase transition. the free energy can be minimized by infinitesimal concentration waves at temperatures just below Tc . 14. but beautiful. Figure 14. AND k0j1 + k0j2 + k0j3 + k0j4  = g The following logic is subtle. The free energy is minimized by finite concentration waves. presented in the classic text by Landau and Lifshitz [150]. The required condition for a secondorder phase transition is that there is no discontinuity in the order parameter at Tc . When Possibility I is true.3 Applications of the Landau–Lifshitz criterion Crystallization from the melt A liquid is isotropic. not infinitesimal ones. the characteristic functions are those of Bloch’s theorem (Sect. In a periodic crystal.50 is true. so a crystal can grow with its precise atomic periodicity in any direction. 14. and g is a reciprocal lattice vector of the disordered alloy. Possibility I therefore forbids a second-order phase transition. we can choose the signs of the η so that the third-order term (with C(k0j1 . not a finite concentration wave. Its star has vectors of all orientations.48 is negative. This criterion evolved from work by Landau in 1937 [149] using the key idea that the free energy must have the same symmetry as the atom distribution. we can choose larger and larger η to minimize further the free energy. it must be impossible to construct a third-order term in the free energy of the disordered phase using the characteristic functions of the group of the ordered phase. shows that for a second-order phase transition to be possible. When Possibility II is true. the phase transition must be first-order. We arrive at the Landau–Lifshitz criterion for a second-order phase transition (Box 14. Unless Eq.2). A group theoretical analysis. (14. as we show next. The concentration wave approach allows for quick identification of first-order phase transitions in a number of phase transitions in materials.4.1 Landau–Lifshitz criterion The Landau–Lifshitz criterion for a second-order phase transition is k0j1 + k0j2 + k0j3  = g . the method of concentration waves gives an explicit construction to test if the cubic term must be nonzero. 14. If so. 6. For the ordered structure.1).50) where {k0j } are wavevectors in the star of the ordered structure. when k0j1 + k0j2 + k0j3  = g.347 14. k0j2 . k0j3 )) in Eq. Although group representation theory gives a general criterion for the possibility of a second-order phase transition. the phase factors of the Bloch functions have wavevectors that form the star of the ordered structure.6 shows how it is possible to construct 17:32:20 .

6 Method of concentration waves and chemical ordering A reciprocal lattice vector. 0) + (0. 3) do not give any vectors of the bcc reciprocal lattice. It can start with infinitesimal concentration waves having wavevectors that are small. Crystallization and melting must be first-order phase transitions. a reciprocal lattice vector with three wavevectors from this star of the ordered (crystal) phase. 1. Spinodal decomposition Unmixing is isotropic. a reciprocal lattice vector. the thermodynamic unmixing occurs with k = 0. 1) + (1. 14. Although Fig. and of course 0 + 0 + 0 = 0.6 would seem to forbid a second-order transition. 1)}. (0. 0. but not zero. the difference with crystallization is that there is an infinite number of possible lengths for the unmixing wavevector in spinodal decomposition. 1) + (1. (It takes a little effort to try all combinations of the ±1. 1) gives a vector of the fcc reciprocal lattice. Note. Therefore all three components of (h. (0. however. In this case. 1. L12 structure The star of the L12 structure is {(1. 0). that the sum of three odd integers gives an odd integer. Sums like (1. 14. 0. however. spinodal unmixing can be continuous. 1. so spinodal decomposition is similar to the earlier example of crystallization. 0) + (0. Unmixing is therefore first order in a proper thermodynamic treatment. 17:32:20 . The kinetic process of spinodal decomposition can show characteristics of a second-order phase transition.6 is therefore infinitesimal. l) are odd. 0). B2 structure The star of the B2 structure is {(1. 1. The L12 structure must form by a first-order phase transition. 1) = (1. Nevertheless. 0. 1)}. 14. The chance of finding three that meet the criteria of Fig. so they cannot sum to an even integer as required for a bcc reciprocal lattice vector. 1. 1.348 t Fig. 1.) The B2 structure may form by a second-order phase transition. 0. showing characteristics of a second-order phase transition. 1) = (3. 2π/a[0. 0. k. The sum (1. constructed from three wavevectors of the disordered liquid phase. 1. 0].

Since concentration waves are parameterized by periodicities.10 and 7. defined as c(r ) = 0.5 Ordering transitions with long-range interactions 349 14. and calculates the critical temperature for long-range order.8. the entropy remains at the same level as for the Bragg–Williams approximation. and not real space positions.1 Thermodynamic formulation The present section develops a thermodynamic theory of ordering.5 Thermodynamics of ordering in the mean field approximation with long-range interactions 14. if r is occupied by a B-atom An ensemble average of c(r ). the effort to include distant nearest-neighbor pair energies is essentially the same as the effort to include first-nearest-neighbor pair energies. never both. an average occupancy of B-atoms. but the present approach with concentration waves allows an important extension beyond the first nearest-neighbor bond energies to account for interatomic interactions over all nearest-neighbor shells. With only long-range order parameters. Assume a lattice site is occupied by either an A-atom or a B-atom. the method of concentration waves does not extend easily beyond the point approximation. and the site is never empty. This was done previously in Sections 2. The occupancy variable for the site at r is c(r ). if r is occupied by an A-atom c(r ) = 1. is calculated as c(r ). Unfortunately. and short-range order parameters are not included in the analysis.5.14.

= e B − k AT 0e c(r ).

The eA and eB have the same meaning as in Eq. = e B − k AT e e B − k BT + 1e e B − k BT +e . 2.52 is essentially the same as the Fermi–Dirac distribution for electrons. 14. (14. not just the z-atoms in the first nearest neighbor shell. (14.54) r  eB − eA =  r  eAB (r − r  ) − eAA (r − r  )    + c(r  ) eAA (r − r  ) + eBB (r − r  ) − 2eAB (r − r  ) .13.53) e −e − Ak TB B e r  eB =   (1 − c(r  )) eAB (r − r  ) + c(r  ) eBB (r − r  ) .51) 1 . Notice how Eq.   eA = (1 − c(r  )) eAA (r − r  ) + c(r  ) eAB (r − r  ) .52) +1 The denominator of Eq. which arises from the analogous assumption that each electronic state can accommodate either zero or one electron. (14. 2.51 is the Z1site of Eq. 14. but here we sum the chemical bond energies from all surrounding atoms.55) . (14. 17:32:20 (14.14.

but it is not confined to the first nearest neighbor shell.56) Substituting Eq.55.Method of concentration waves and chemical ordering 350 3 4 5 V(r–r1) V(r–r2) V(r–r3) V(r–r4) 2 A/B r t Fig. Figure 14.7 1 Geometry for ordering problem with arbitrary-range interactions.24 we define a quantity V. 14. Modulation of gray shading of neighboring sites represents a small concentration wave.52 becomes  c(r ). 14. (14. 14.56 into 14.7 depicts how the central atom at r has bonds of various strengths to surrounding atoms at r  . As in Eq.   V(r − r  ) ≡ eAA (r − r  ) + eBB (r − r  ) − 2eAB (r − r  ) /4 . 2. (r  ). Eq.

(14. c0 . 14. or may contain a small concentration wave (r ).57 into this compact form c(r ). = exp r−r  )−eAA (r−r  )] r  [eAB ( kB T 1   exp r  4V(r−r  ) c(r ) kB T  +1 . define E0 and δ to recast Eq.5.57) 14.58) For convenience in what follows.2 Small-amplitude concentration waves We consider an alloy that either has uniform composition. (14. We write c(r  ) at the neighboring sites {r  } as c(r  ) = c0 + (r  ) .

(14. and another for the small concentration wave (r  )     r − r  ) − eAA (r − r  )] r − r  ) c0 r  [eAB ( r  4V( exp . E0 eδ + 1 (14. (14.61) kB T  r 17:32:20 .59) which separates factors involving the concentration at the neighboring atom sites into one for constant composition c0 . = 1 .60) E0 = exp kB T kB T 4  δ= V(r − r  ) (r  ) .

then use Eq. E0 (1 + δ) + 1 1  .63) To make further progress.59. such as near Tc . i.63 1 . so we expand eδ in Eq. we assume the concentration wave amplitudes are very small.14.62) (14. for which there are no concentration waves. 14. E0 = c0 c0 = (14. and Eq.59 becomes 1 . E0 + 1 1 − c0 .5 Ordering transitions with long-range interactions 351 As a reference state we will use a random solid solution. Therefore δ is very small. (r ) = 0 and δ = 0.e.. 14. c(r ). 14.

=  1−c0 (1 + δ) + 1 c0 c0 c(r ).

= . 1 + δ(1 − c0 ) c(r ).

65) (14.66) For small δ   c(r ). = (14.64) (14.

67) c(r ). = c0 1 − δ(1 − c0 ) . (14.

58. (14. using Eq. 14. 14.61.68) Substituting from Eq. recognizing that c0 c for small . and recognizing that the ensemble average (r ). = c0 − c0 (1 − c0 ) δ .

14. we can readily get the critical temperature for an ordering alloy with first-neighbor interactions but a general composition c.69) r when (r ) is small. kB (14.49. Equation 14. we will transform this r V  (r ) to v˜ (k0 ).69 can be used to obtain Tc . which is a more general result than Eq. This case includes the chessboard and B2 structures where the 17:32:20 .3 Critical temperature Our assumption of very small (r ) is true near the critical temperature. We want to put this variation into V(r ) to define a V  (r ). 14. and show a systematic way to go from V to V  for an ordered structure with the star k0 . however. valid when (r ) is small in equilibrium. Different ordered structures will have different values of  at different neighboring sites. and Eq.70) r  In what follows. kB T  (14. Right now. 14.5.57. = (r ) (r ) = −c(1 − c) 4  V(r − r  )(r  ) . 2.69 is a linearized form of Eq. and obtain a simple expression for Tc Tc = −c(1 − c) 4   V (r ) .

75 is Tc Tc = − 4 c(1 − c)  v˜ (k0 ) . the temperature defined by Eq. 14. v˜ (k) δ( kB T (14. . with a star containing one vector.74) r and the convolution in Eq.71) For the equiatomic alloy with c = 0.5. 14. Our (r ) does this quite well. . and it is often not possible to consider instabilities by other small concentration waves. kB (14. but it is necessary to use an order parameter to quantify the difference of atom concentrations on the two sublattices. as of v˜ (k).69 is much more general. (Once this concentration wave of k0 has grown. kB (14.5. but it will be easier to determine the critical temperature. Tc = zV1 . 2. (14. which exist near Tc . the alloy is no longer a disordered solid solution. k0 = 2π/a(1.49.69. and this structural information is mathematically convoluted with the interaction energy in Eq. . 14. 1.77) r1 = a 2 2 2 2 2 2 13 It is not necessary to use concentration waves to obtain Eq. 14. We start by Fourier transformation of the quantities V(r ) and (r )    = e−ik·r V(r ) . 0) .4 Formulation in k-space The method of concentration waves can be used to recast Eq.69 becomes a multiplication in k-space  =− ˜ k) δ( 4 c(1 − c)   ˜ k) . shown in Fig.73) v˜ (k) r  = ˜ k) δ(   e−ik·r (r ) . We evaluate v˜ (k0 ) of Eq.69 into an interesting form in k-space. First. r2 = a(1. .70. since it allows pairwise interatomic interactions of arbitrary range. There are z identical first-nearest-neighbors. It does require information about the ordered structure. so13 Tc = −c(1 − c) 4z (−V1 ) . 0) . 17:32:20 . kB (14. since this wave is first to achieve b = 0 upon cooling. r4 = a 1 .75 pertains to small concentration waves.76) We work the example of the critical temperature for B2 ordering. .72) which agrees with Eq. r3 = a(1. 14. 14. Equation 14. 0. (14. so (r1 ) = −(0). the exponential phase factors for the concentration waves are determined.75) The critical temperature can be obtained for the concentration wave with the lowest value  here denoted v˜ (k0 ).73 with two steps. 14.) Since Eq. The nearest-neighbor vectors are     1 1 1 1 1 1 . 14.352 Method of concentration waves and chemical ordering first neighbors are the other species of atom. 1). . (14. however. 1.5. 14. The result will still be confined to small-amplitude concentration waves.

show that the average chemical interaction from the solutes in the 1nn shell is 8cV1 by evaluating the sum P(N. but it is still without an analytical solution for δ( Problems 14. positive V1 favors B2 order for equiatomic alloys. too. e −i2π(111)·r3j = +1 . 14.Problems 353 giving e−i2π(111)·r1j = −1 . n. (b) For a bcc alloy with solute concentration c. 6.1).69 and 14. n.14.82) (a) Explain why. We then need to use the full. so Eqs. .75 are valid at temperatures infinitesimally below Tc .75 are inaccurate. the probability of finding n solute atoms in a nearest-neighbor shell having N sites is the binomial probability P(N.79) gives v˜ (k0 ) = −8V1 + 6V2 + 12V3 − 24V4 + 8V5 + . 12. (14. c) = N! cn (1 − c)N−n .57 which. (14. . 8. 6 . Equations 14.57. Other possibilities such as a strong. is transcendental and has no simple analytical form. 14.5. 14. 14.1 For a random alloy of solute concentration c.83) n=0 (c) By a similar method. a strong. and we count the number of sites in the shells. . positive V4 might promote B2 order. Far below Tc− .81) +1 ˜ kj ). . these equations have the trivial solution (r ) = 0 because the solid solution is stable and concentration waves do not form. the assumptions of Sect. The k-space version of Eq. 2. The product of these two parts (numbers of neighbors and Eqs.69 and 14. . e−i2π(111)·r2j = +1 . nonlinear expression of Eq. (N − n)! n! (14. 14.78. the V(r ) is decomposed into terms for the different nearest-neighbor shells. but is not the most stable structure for the alloy with strong V4 (see Sect. 24. but other concentration waves would have to be considered – perhaps the B2 structure is stable. c) = 8  P(8. n. which for bcc is {8. (14. evaluate the variance n2 . e −i2π(111)·r4j (14. (14.79) Second. is simpler than Eq. At temperatures above Tc . . 14.80) As expected. like Eq.}. c) nV1 = 8cV1 .78) = −1 .57 c0 + star  j ˜ kj ) = δ(  exp 1 V0 c0 + star j ˜ kj ) v˜ (kj ) δ( kB T  .2 are not valid. 19.46.

− n.

17:32:20 (14.84) .2 .

14. Show that 6 8   P(8.38. we expect 6 8   P(8. especially pages 1–64. c) jV . m.3 Figure 14. Determine whether this ordered structure can or cannot form by a second-order phase transition. (14. 14. (14. n.8 shows the fcc-based ordered structure of CuPt (L11 structure). n. when used in Eq. (b) What is the potential v˜ (k0 ) through fourth neighbors. c)(nV1 + mV2 ) = 8cV1 + 6cV2 .85) n=0 m=0 Now if V1 = V2 = V. c)P(6. c)P(6. (d) For bonding effects from the 1nn and 2nn shells of the bcc structure.2 For the B2 structure: (a) Evaluate the concentration wave at positions of the first through fourth neighbors of an A-atom at the origin. where k0 is the ordering wavevector? 14. (r ) = 17:32:20 .) 14.33.4 The concentration wave (r ) is sometimes written as a complex exponential plus its complex conjugate 1  −ik·    (14. gives a result like that of Eq. m. (Hint: Reference [148] is useful for this problem. each of the 14 sites of the 1nn and 2nn shells makes the same contribution to the chemical bonding. c)(nV + mV) = n=0 m=0 14  P(14.86) j=0 14. j.Method of concentration waves and chemical ordering 354 a t Fig.87) η e r + η∗ e+ik·r 2 Show that this expression for (r ). 14.8 The L11 structure of CuPt.

Because the atoms do not move with individual independence. and from the viewpoint of the vibrational entropy of the transformation. the atoms move independently (often by a vacancy mechanism). Part III has described melting. 17. (These descriptions of second-order and first-order displacive transitions find use again for ferroelectric transitions in Sect.15 Diffusionless transformations So far. A fully satisfying theory of martensitic transformations is not offered. sometimes with magnetic entropy as in the case of iron. in which the atoms in a crystal move cooperatively. Some macroscopic and microscopic features of martensite are described. these all require the diffusion of atoms over moderate distances. followed by a two-dimensional analog to the crystallographic theory for the martensite habit plane. Ordering transformations also require the movements of individual atoms. and does not require a vacancy mechanism for the motions of individual atoms. so twinning is not a “true” phase transition. For alloys.7. since it is not yet available. The entropy of martensitic transitions is primarily vibrational. and unmixing phase transformations. from the viewpoint of soft modes in bcc structures that may be a step along the path of a martensitic transformation. precipitation.) Finally phonons are discussed. This chapter begins with a discussion of dislocations and how their glide can be used in mechanisms of crystallographic shear. however. Displacive phase transitions are explained more formally with Landau theories having anharmonic potentials and vibrational entropy.” in which a crystal transforms into a different variant of the same type of crystal. but again without long-range diffusion. Nevertheless. Diffusionless transformations include “twinning.1 Martensitic transformations are changes in crystal structure that occur by shears and dilatations. and the configurational entropy can undergo large changes with only a few jumps per atom. This chapter describes diffusionless transformations. for which continuum diffusion equations provide much of the essential behavior. The atoms may not move at exactly the same time. Vacancy migration is not important for either twinning or martensitic transformations. and the crystal is distorted into a new shape. the change in configurational entropy is small. 17:32:22 . but during ordering the atom movements are over such short distances that a diffusion equation is not appropriate. but the transformation is very fast. 1 The new twin phase and its parent have the same atomic structure and have the same volume free energy.

There are two types of “pure” dislocations. (Giant.1 Dislocations and mechanisms By crystal symmetry. and left. up. there are usually multiple variants of the product phases of a diffusionless transformation. 15. Somehow. On the figure is drawn a circuit of five atom spacings.) The agent of change in diffusionless transformations is probably the dislocation.1. and their role in shear transformations is then explained. and probably slower. This communication is expected to be no faster than the speed of sound.1. An edge dislocation is the easiest to illustrate. Dislocations carry large displacements.Diffusionless transformations 356 15. and glide of a dislocation are first reviewed.” does not close perfectly when it encloses the dislocation line. notice how an extra half-plane of atoms has been inserted in the upper half of the simple cubic crystal. of course. nonlinear vibrations are discussed in Sect. 15. Transmitting the shear to surrounding regions by sound waves has some plausibility.1 x a Edge dislocation in a cubic crystal. 15. This circuit. and can move at a good fraction of the speed of sound. known as a “Burgers circuit.7. energy. All atoms in a domain of one variant must undergo the same movements with respect to their neighbors. A large body of knowledge has formed around dislocations because their movement is the elementary mechanism of plastic deformation of many crystalline materials. Dislocation line is parallel to yˆ. The structure. the atoms at the initiation point communicate their selection of variant to neighboring atoms. however. and it also closes perfectly when it is drawn in a dislocated crystal around a region that does not contain b z y t Fig. In Fig.1 Structure of a dislocation A dislocation is the only type of line defect in a solid. right. b = a 100. (It does close in a perfect simple cubic crystal. but the atom displacements in the shear process are much larger than the atom vibrations in a phonon. 15. down. This extra half-plane terminates at the “core” of the edge dislocation line.

. 17:32:22 . and b is perpendicular to the dislocation line.

15.3 z y x Screw dislocation in a cylinder of cubic crystal. b = a 001.2 t Fig. 15.15. Dislocation line is parallel to ˆz.1 Dislocations and mechanisms 357 Cut surface b t Fig.

and glide of dislocation from left to right.2 is also along zˆ.” The applied shear stress drives the edge dislocation from left to right along a “glide plane” that is halfway down the crystal.2. They are called “mixed dislocations.3 shows how an edge dislocation moves under stress in a process called “glide.2. In Fig. dislocations are neither pure edge dislocations nor pure screw dislocations. Each step occurs as the distorted bonds on one plane are broken at the glide plane. b. and a half-plane of the lower crystal joins to the half-plane of 17:32:22 . It is a “screw dislocation. the dislocation core. b is parallel to the line of the dislocation. Finally. (g) Final state of crystal without applied stress.” Whenever a dislocation line is curved.) The vector from the end to the start of the circuit is defined as the  “Burgers vector” of the dislocation. After completing a Burgers circuit in the x–y plane in Fig. a b c d e f g (a) Initial state of crystal without applied stress. 15. but the character changes because the dislocation line can be curved (and usually is). part of the dislocation must have mixed character. the b for a mixed dislocation is the same everywhere along the line of the dislocation. and b is parallel to the dislocation line.1.. (b–f) Applied stress indicated by arrows. the crystal planes form a helix. Dislocations are characterized by their Burgers vector and the direction of their dislocation line.2 Dislocation glide Figure 15. Around the core of a screw dislocation. the magnitude of the Burgers vector b is the “strength” of the dislocation – dislocations with larger b cause larger crystalline distortions. the vector from finish to start lies along zˆ.1 the b = aˆx. and the line in the center of the cylinder in Fig. 15.” and is illustrated in Fig. and the dislocation line is along yˆ . In general. In general. 15. 15.” The other type of “pure” dislocation has its Burgers vector parallel to the dislocation line. 15. For a screw dislocation. but rather have their Burgers vectors at some intermediate angle to their line direction. This is an “edge dislocation.

3a and 15. therefore. Outside the core region. Neglecting the smaller core energy. 0 0 2 Eelastic = Gb K [J/cm] .4 shows how to accommodate two extra half-planes with either one dislocation of b = 2a.3b to 15. and the Burgers vector is the vector of the shear displacement across the cut crystal. Consider the energies.1) (15. First note that the cut itself requires no energy. comparing the macroscopic shapes of Fig. G. sometimes the “core energy” is estimated from the latent heat of melting.3f. The alternative of breaking all bonds in the glide plane simultaneously. The elastic energy stored in the crystal is obtained by integrating the force over the distance. it is reasonable to calculate the energy by linear elasticity theory. however. An elastic restoring force opposes the shear. at least until it leaves the surface of the crystal. Big 17:32:22 .358 Diffusionless transformations the edge dislocation. 15. Here K is a geometrical constant that depends on the size and shape of the crystal (and somewhat on the dislocation character).2a shows how a dislocation can be created by a cut-and-shear process. When the stress is removed. smaller dislocations is half as large. or two dislocations. and this restoring force is proportional to the shear times the shear modulus. of shear 1cmb Eelastic ∝ G x dxdz . The energy needed to make the dislocation is the energy required to make the shear across the cut surface. the energy cost of creating a unit length of edge dislocation is the Eelastic of Eq. The dislocation is preserved in this process.4 Dislocation reactions Because the self-energy of a dislocation increases as b2 . x. 15.1.3 Strain energy of a dislocation (self energy) A dislocation generates large elastic strains in the surrounding crystal. as is evident from Figs.2.1 and 15. Dislocation glide is usually the mechanism for the plastic deformation of crystals.2) for 1 cm of dislocation line along zˆ. Figure 15.3g shows that the crystal has undergone some plastic shear consistent with the applied stress. Because the atom configurations in the core are noncrystalline. The total elastic energy of a crystal with the two separate. the energy cost of making a unit length of dislocation line is equal to the elastic energy per unit length of the dislocation. each of b = a. since the atoms across the cut are properly reconnected after the dislocation is made. The dislocation line is located at the edge of the cut. and doing the shear in one step from Fig. Approximately. It turns out that this total elastic energy in the surrounding crystal is typically an order-ofmagnitude larger than the energy of the core region. The strain in the material in the dislocation core (usually considered  is so large that its excess energy cannot be accurately to be a cylinder of radius 5|b|) regarded as elastic energy. dislocations prefer Burgers vectors that are as small as possible. 15. The distance of displacement across the cut is b. 15. requires stresses that are far larger. Think of the cut crystal as a spring. (15. 15.2. 15. Figure 15.1.

Smaller Burgers vectors are usually not possible. single dislocations have the smallest possible Burgers vector. (d) hcp stacking of the two types of close-packed planes seen from above. C B A a b c d A B t Fig. Figure 15. The next layer will be an A-layer. dislocations therefore decompose into smaller ones. however.1. 15. and will locate directly above the dark A-layer at the bottom. 15. The shifts between the adjacent layers of the fcc structure are smaller than this. b = 1/2[110].5 Stacking faults in fcc crystals A special dislocation reaction occurs for dislocations on close-packed {111} planes in fcc crystals. and these small shifts can be obtained by creating a “stacking fault” in the fcc crystal. (c) hcp stacking of close-packed (0001) basal planes. Specifically.5 shows how the stacking of these close-packed planes determines whether the crystal is fcc or hcp. The two smaller dislocations try to move apart so their elastic fields do not overlap.1 Dislocations and mechanisms 359 a t Fig. with the cubic face marked with the square. perspective view of three layers.4 Accommodation of the same macroscopic distortion by two dislocations or by one dislocation. The next layer will be an A-layer. perspective view of three layers. and will locate directly above the dark A-layer at the bottom.5 C (a) fcc stacking of close-packed (111) planes. 15. but an exception occurs for fcc and hcp crystals. This lower limit is typically the distance between nearest-neighbor atoms.4 have a repulsive force between them. Note the error in stacking from placing an A-layer to the immediate right of a B-layer. The structure 17:32:22 . For the same reason. (b) Stacking of the three types of (111) planes seen from above.15. The “perfect dislocation” in the fcc crystal has a Burgers vector of the nearest-neighbor separation. 15. assume that the ABCABCABC stacking of the fcc crystal is interrupted by a small shift of a {111} plane as: ABCAB|ABCABC. so the two in Fig. The lower limit to the Burgers vector is set by the requirement that the atoms must match positions across the cut in the crystal.

.and a B-layer.). .AB|AB.Diffusionless transformations 360 is still close-packed. This shift is a vector of the type: a/6 112. which has a Burgers vector equal to the shift between an A. The hcp region can be bounded by a “Shockley partial” dislocation. This region of hcp crystal need not extend to the edge of the crystal. but there is a narrow region of hcp crystal (. . however. .

Equilibrium separations of Shockley partial dislocations offer a way to determine the stacking fault energy of fcc crystals. and the stacking fault energy tends to keep the partials from getting too far apart. 15. Note that the total Burgers vectors are equal before and after this reaction2 a/2[110] → a/6[121] + a/6[211] . C. 3 The atoms in the different registries are not all in the plane of the paper. which is a shift of a/6 [121]. but they project onto the positions in the figure. but must extend to the surface or form a loop.2 2 KG a4 12 + 12 + 0 Eperfect 3 = (15. for example. Three of these give a net shift of 3 × a/6 [121] = a/2 [101] + a[010]. for example. There is. 2 E2partials 2 2KG a 12 + 22 + 12 36 It is energetically favorable for a perfect dislocation in an fcc crystal to split into two Shockley partial dislocations. 15. is smaller for the two Shockley partials on the right than the single perfect dislocation on the left. performed one layer at a time. proportional to the square of the Burgers vector. In other words.3) The energy. a thin region of hcp crystal between these two Shockley partials (the stacking fault). B. The original crystal is shown in Fig.4) = .2 Twinning Figure 15.6 illustrates twinning in an fcc crystal by small displacements of close-packed (111) planes. which move apart to reduce the elastic energy. (An atom in a C-registry is returned to a C-registry.. This stacking fault energy is approximately (or at least conceptually) related to the free energy difference between the fcc and hcp crystal structures. (15.6b shows three sequential slips of upper planes of the crystal.6a. Consider a specific dislocation reaction when a perfect dislocation decomposes into two Shockley partial dislocations. 15. however. 2 The conservation of Burgers vector is equivalent to the fact that a dislocation line cannot terminate in the middle of a crystal. 15.3 Figure 15. This is verified by calculating the energies with Eq. with its planes in three different vertical registries indicated by shading and labels A. The first vector is a nearest-neighbor vector in the fcc structure.6b. and the second is a translation of the standard fcc unit cell.) These shifts are depicted by the short horizontal arrows in Fig. The magnitudes of the slips are in fact just what is obtained by passing a Shockley partial dislocation across a (111) plane. 17:32:22 . and the new registrations of the planes. these three shifts by a Shockley partial dislocation return an atom to its original vertical registry.

” The model includes two dislocations: a screw dislocation that serves as the pole. The macroscopic shapes of twins seem to be symmetrical. 15. At the same time. 15.6b. It is also possible to consider the twinning plane as a grain boundary between two fcc crystals of different orientation.7 is that the two segments of the partial dislocation must pass over each other in the first wrap. however. the B directly above B. the left segment of this dislocation can move down the pole causing appropriate shifts of the lower planes. but the energy is still a concern. followed by slip above the original A (now B). The pole mechanism is depicted in Fig.2 Twinning 361 t Fig. The result is a twin at the upper half of the crystal. One solution to the problem uses an out-of-plane orientation for one of the segments. and a Shockley dislocation that wraps around the pole and spirals upwards plane-by-plane. 15. Another solution is that prior deformation in the material has induced internal stresses that can help the passage of the dislocation segments 17:32:22 . (c) Reflection symmetry across the dashed twinning plane. or some character that includes the a/3[111] displacement between the close-packed planes. not just the first plane.15.7. (b) Slip of crystal planes above the middle C-layer as a block. and C above C. An issue with the pole mechanism as shown in Fig. the stacking of ABC above the first plane of shift is transformed to CBA (the grayscale sequence is reversed). followed by slip above original B (now A). The two segments probably do not remain straight lines that overlap at all radii simultaneously. and this involves significant elastic energy. The relationship between twinning and moving a Shockley partial dislocation on every close-packed plane suggests a process for a twinning called the “pole mechanism. 15. so there are matched planes of A–A. 15. Take a look at the systematics of using the short arrows with the notation for the ABC stacking in Fig. B–B and C–C planes equidistant from the twinning plane. The screw dislocation could have the Burgers vector of a perfect dislocation.6 a C B A C B A b C B A C B A A C B C B A A C B C B A C A B C B A B A B C B A B A B C B A c C A B C B A (a) Depiction of stacking sequence of close-packed planes of an fcc crystal.6b – each shift requires re-registering all planes above the shift.6c shows that there is a reflection symmetry across the twinning plane. As the working dislocation wraps around the pole. Finally. it moves up by one close-packed plane for each wrap. so likely the upper and lower planes are shifted symmetrically by the two counter-rotating segments. after the shifts shown in Fig. Figure 15. The vertical arrows indicate that A-layers are directly above A-layers.

15. Martensitic transformations also occur in Fe–Ni.3 Martensite 15. 15.7 A pole mechanism for twinning.1 General features of martensite A plain carbon steel (for example Fe–1 at. with reliance on the handy registry notation of Fig. 11. Problem 15. 15. twinning often competes with the process of dislocation glide. many aspects of martensitic transformations. A third solution is that the partial dislocation segments move at a very high speed.7 with the phase diagram of Fig.” Because of the technological importance of steel. The large helix denotes atomic surfaces intersected by the screw dislocation.6b and the shifts of letters using the right arrows. Twinning is sometimes favored at high deformation rates. The details are not fully understood. Ti–Nb. twinning can sometimes be driven by applied stress. Nevertheless. After rapidly quenching the steel to room temperature. Au–Cd. In this case there is a “relativistic” change to dislocation self energies that alters the elastic interaction between the two segments. it emits audible high-frequency sounds as twinning occurs.%C) at a temperature of 950 ◦ C has the fcc phase known as “austenite. The Fe atoms occupy the sites of an fcc lattice. over each other. and their elastic field does not keep up with them.” As a mechanism for plastic deformation.1 works through this case. 17:32:22 . it transforms into a body-centered tetragonal (bct) phase called “martensite. but this is material-specific. remain unclear. known as the “cry of tin. including the basic mechanism of atom displacements. for example.362 Diffusionless transformations Screw pole Shockley partial t Fig. Twinning tends to occur during the deformation of bcc iron at low temperatures. Cu–Zn. The deformation of tin is perhaps the most famous. making their overlap less unfavorable. When a bar of tin is bent or deformed at room temperature.13. constructed with two perpendicular dislocation lines. the martensitic transformation has received extensive study for many years. 11. In–Tl. which has made about 3 12 wraps about the vertical pole dislocation.3.” as discussed in Sect. Phase transformations between fcc and hcp structures have many similarities to the twinning process just considered for fcc crystals. Because twinning gives a reorientation of a crystal. and the interstitial C atoms are in octahedral sites such as the site in the center of the standard fcc unit cell.

as shown in Fig. One consequence is that the local chemical composition cannot change during a martensitic transformation. where the transformation occurs in bursts. martensite often forms in a series of bursts as shown in Fig. 15. Figure 15. 152]. 15.8 are very different.” Martensite plates grow at velocities of a fraction of the speed of sound. Cu–Al. Martensite initially forms as very thin plates that subsequently thicken during the rapid growth stage. 15. are thought to be the original nuclei of the martensite phase.10 −0. but other morphologies such as needles and laths are also observed. Martensite usually forms as thin plates in the austenite phase.3 0.9 is a schematic diagram of the plate formation process.10 shows an almost fully transformed Fe–Ni alloy. This is believed to be an issue with the nucleation of martensite in iron alloys. the reverse transformation upon heating. which reveal the presence of midribs running along the center line of the plate. Figure 15. V3 Si and Nb3 Sn. but at a much higher temperature than for cooling in panel a. (a) Heat release upon cooling. The entropy of the transformation originates from changes in vibrational modes. and this hysteresis often makes it difficult to deduce thermodynamic properties for the martensite transformation. The formation of a martensite platelet can trigger the formation of other plates.8a. is more continuous.05 Cooling −0. Eight characteristic features of martensite are enumerated below [125]. 1.11. the plates lie on distinct crystallographic planes in the parent phase. The thickening process is inferred from optical and electron microscopy. The preferred plane is called the “habit plane. as seen in Fig.00 0.1 0. the specimen surface or another martensite plate. The area in the exothermic peaks is approximately proportional to the amount of new phase that has formed.8b. 2.5 0. On the other hand.0 Heating 500 260 b 550 600 650 700 As Temperature [K] 750 Af Calorimetry measurements of martensitic transformation in Fe71 Ni29 .4 0. and growth is terminated when the plate encounters a grain boundary. 15. (b) Heat absorption upon heating. During cooling. For a given composition. 15. and all local atomic configurations change in the same way [151. 17:32:22 .20 180 t Fig. from changes in magnetic order.15.8 a 200 200 240 Ms Mf Temperature [K] ΔCP [Jg−1 K−1] ΔCP [Jg−1 K−1] 0. notice the successively smaller and smaller plates as more of the specimen is transformed to martensite. Midribs. eg. They occur in superconductors. showing a large hysteresis for the transformation. Martensitic transformations are “diffusionless. or in some Fe alloys. The transformation temperatures of the forward and reverse martensitic transformation in Fig. and in ceramic systems such as ZrO2 and BaTiO3 . The transformation therefore causes a negligible change in the configurational entropy of the alloy. and many other alloys.3 Martensite 363 −0. The reverse transformation from α to γ occurs continuously.2 0.” Atoms move by less than one interatomic distance and the product phase is formed by a cooperative motion of many atoms.15 −0.

the volume fraction of martensite increases with time. Intersecting martensite plates in an Fe–32%Ni specimen. From [153]. The straight line running through the center of the plates is known as a midrib. Ms . 15. with permission.364 t Fig.9 t Fig.14 mm. Reproduced. Some isothermal character to the 17:32:22 . and holding at temperature for any length of time does not cause the formation of more martensite. 3. The fine lines running across the midrib are twins. In the “isothermal” case. from [125]. with permission. from [153]. Image width is 0.10 t Fig. 15. Martensite plates in an almost completely transformed Fe–Ni specimen. Growth of plates is terminated at grain boundaries and at other martensite plates.11 Diffusionless transformations Sequence of martensite plate formation. the “martensite finish” temperature. 15. whereas the specimen is completely transformed for temperatures below Mf . The first martensite plates form at the “martensite start” temperature. Reproduced. The volume fraction may or may not change with time. For “athermal” martensites. the volume fraction of the product phase changes almost instantaneously at any temperature in the range Ms > T > Mf . the volume fraction of martensite is a function of temperature. Upon cooling.

so this transformation is mostly athermal. the [101] direction in austenite. There is a crystallographic relationship between the martensite and austenite crystals.% Ni. In addition. 4. In Fe–Ni alloys. the orientation relationship found by Nishiyama [155] is (111)γ || (011)α  [112]γ || [011]α  . with Ni content greater than 28wt.5) The two planes that lie parallel to one another are the close-packed planes in each structure. i. the plane separating the parent and martensite phases. but most of the transformation occurs promptly upon cooling to the low temperature.4 0.6 0. Using the standard notation of “γ ” representing the fcc austenite and “α  ” denoting the bct product phase.13b.0 t Fig. The macroscopic distortion caused by the formation of martensite is a homogeneous shear. shows the subsurface region. In plain carbon steels. The habit plane.2 223K 0.12 0 60 120 180 Transformation time (minutes) Diffractometry measurements of the remaining austenite as martensite forms in Fe71 Ni29 at different times. 15. (15. Most of the transformation occurs quickly (from 1. An illustration of the surface relief phenomenon. depicted in Fig. 6. (15. The inside of the martensite plate is sheared or twinned. but make a distinct angle with respect to the lines of the austenite. 15. which lies in the (111) plane.6) Other martensitic transformations have different orientation relationships.13a shows a series of straight parallel lines that were inscribed across the surface of the parent phase. the Kurdjumov–Sachs relationship is (111)γ || (011)α  [101]γ || [111]α  . and these need not involve low-index planes and directions [156]. however.15. is one of zero macroscopic distortion. 15.0 γ-phase fraction 0.0 to the fraction on the y-axis) but some transformation continues after times of hours.8 243K 0. Martensite transformations are often reversible in the sense that the product phase will transform back to the austenite phase upon heating.12. After a martensite plate forms and intersects the sample surface. Figure 15. 5.e. the lines remain straight in the martensite (labeled M). The temperature at which the 17:32:22 .3 Martensite 365 1. Kurdjumov and Sachs [154] determined that the (111) plane in the austenite phase lies parallel to the (011) plane in the martensite. martensitic transformation in Fe71 Ni29 is shown in Fig. is oriented parallel to the [111] direction in the martensite.

except only half as many Shockley partial dislocations move along the glide planes (i. Also.. From [125]. In contrast. the temperature is lowered again. The formation of this hexagonal   -martensite is common in some iron alloys. Notice how a clockwise shear is induced 17:32:22 .3. Suppose austenite is quenched to a temperature below Ms but above Mf .1 illustrates the details of how an fcc-to-hcp transformation can occur in almost the same way as twinning. The amount transformed after the holding time is less than would be obtained if the material had been directly quenched to this final temperature. 15.Diffusionless transformations 366 M t Fig.2 Transformation mechanisms Despite the enormous technological importance of martensite in steels. 15. 15. 7. and this simple mechanism of a coordinated glide of Shockley partial dislocations is believed responsible for its formation. The highest temperature at which the martensite can form during deformation is Md . (b) Surface upheaval due to the plate intersecting the surface of the sample. In general. In a study of the Cu–Al–Ni system. For any temperature in the range Ms > T > Mf . there is thermal hysteresis such that As > Mf and Af > Ms . however. and much discussion is framed in terms of these models. a Shockley dislocation glides across every other fcc closepacked plane). A plastic strain can also cause the formation of martensite plates even at temperatures above Ms . 8.13 a M Surface b (a) View normal to the surface of a specimen with a martensite plate. there is no consensus on the atomic-level mechanism (or mechanisms) of the fcc-to-bct martensite transformation in iron alloys. some detailed models have been proposed. reverse reaction starts is denoted by As .3. An applied stress or high magnetic field can also initiate a martensitic transformation. After the holding time.e. The first plate to form at Ms during cooling is the last one remaining upon heating. This section 15. the last plate to appear at Mf is the first plate to retransform at the temperature As . Nevertheless. but a macroscopic shear due to the transformation is observed. Problem 15. deforming the austenite at temperatures above Md often inhibits the martensite transformation. The magnitude of the hysteresis can be quite large.6. and the transformation finishes at the temperature Af . Martensite does not form immediately. as shown in Fig. Kurdjumov and Khandros [157] (see also Tong and Wayman [158]) demonstrated the microscopic reversibility of martensite reactions. 15.8. Parallel lines inscribed on the surface remain straight. This holding treatment is called “stabilization” of austenite [159]. and held at this intermediate temperature for a time.2 describes an important model that has similarities with the twinning transformation shown in Fig. the amount of martensite will usually be increased by plastic strain.

17:32:22 . The dislocation arrays may begin near the surfaces of the transformed particle. although they are spaced on every third glide plane. extended to a detailed dislocation mechanism as the Olson–Cohen model [161]. by showing two arrays of partial dislocations that cause clockwise shear.14 Arrays of dislocations in an fcc crystal that could produce a near-bcc structure in their zone of intersection. and the lower array may be described as a faulted fcc crystal. and should not be the full displacement of a Shockley partial dislocation on just every second or every third plane. The lines are different {111} planes in the fcc crystal. The top array would produce a band of hcp crystal. and a smaller shear occurs with the fcc-to-hcp transformation. includes the same set of partial dislocations that form hcp crystal from fcc. but there is no geometrical reason why the dislocation arrays need to intersect as bands (and in fact a band of hcp phase is not found after many martensitic transformations in many materials). either after the passage of the partial dislocations or simultaneously.6. Another feature of a dislocation mechanism is that the atom shifts should be the same on all planes. i.e.15. although the formation of dislocations inside a bulk material will require an energy barrier.3 Martensite 367 t Fig. An important proposed mechanism has the euphonious4 name “Bogers–Burgers double shear. The zone of intersection is the transformed bcc martensite. but transformation to bcc is possible by shears of different orientation. identical Shockley partial dislocations gliding across alternate close-packed (111) planes. Notice that the clockwise shear caused by the first array causes a larger deflection of the lower array of dislocations.” The mechanism of Bogers–Burgers double shear [160]. 4 Agreeable to the ear. 15. Intersecting this region of hcp crystal (either sequentially or simultaneously) is another array of partial dislocations. Depicting intersecting bands of glide planes is helpful to show the geometrical relationships between the dislocation arrays. Figure 15. One shear is insufficient to transform a 3D fcc crystal to bcc. These questions are difficult to address by experimental measurement because they are at such a small scale and occur so quickly. 15. by the twinning mechanism in Fig. 162]). The atoms in the intervening planes must therefore “shuffle” into position..14 gives the general idea (crystallographic details are given in [161.

” highlighted with an arrow. and reinsertion into the original austenite in Fig. There is. shown in Fig.15 (a) Two fcc unit cells. The unit cell is widened from a to a(1+). 15. The habit plane in Fig. 15. It requires a major flattening and expansion to become a bct (nearly bcc) crystal. The steps include the “Bain strain. Nature has found a way to do the martensitic transformation with minimal elastic energy by using a geometrical trick.15b and c.16a cannot be chosen arbitrarily.16.16c. The gray unit cell in Fig. 15. A direct transformation in shape would generate stresses in the surrounding austenite larger than the 10 GPa in Table 8. The trick is explained with Fig. (c) Bain distortion to transform bct cell into bcc cell. so the transformed t Fig. 15. This distortion is small if the width of the transformed region is small. but it is far easier to understand than the full three-dimensional crystallographic theory of martensite [163]–[167]. however.16 that involves the atoms in the unit cell.368 Diffusionless transformations 15. there is a rotation of the transformed band. 15. a shear of the austenite itself. This two-dimensional problem is incomplete. and could not be driven by a thermochemical process. 15. 15. 17:32:22 . with bct cell in gray. 15. The approach to finding the habit plane is to find the condition for zero strain energy.15a is not a cube. which is the only step in Fig. 15.16c.1.” that minimizes the elastic energy in the material. 15. Notice the perfect registry of atomic planes across the two interfaces where the transformed zone contacts the austenite. Finally.16c. Suppose this strain is  in the horizontal direction of the figure. Consider the differences in unit cell of the austenite and martensite. These steps include slip at the bottom of Fig.15 are so large that they could create elastic energies that are orders of magnitude larger than any discussed so far.) The geometrical requirements for the habit plane are described here for two dimensions. and martensite forms as plates to minimize this distortion.15. but they are then comparable to chemical and thermal energies of phase transitions. as shown in Fig. and in fact the habit plane changes with the Bain strain. The other steps are macroscopic ones to ensure that the transformed region fits well in the austenite. 15.4 The crystallographic theory of martensite An important feature of a martensitic transformation is how the individual plates of martensite are oriented in the parent austenite crystal. which was designed to ensure that the transformed band is macroscopically flat. The martensite plates lie on a “habit plane. This is too extreme. (b) Isolated bct cell from part a. The transformation strains shown in Fig. (There are still substantial elastic strains associated with martensite. as shown by the difference between Figs.

In three dimensions. The idea is shown in Fig. For this we need another degree of freedom. 15. which now fits with atomic registry in the original austenite. (d) Rotation of band. There was one subtly arbitrary decision in drawing Fig. 15. but with a different dislocation mechanism than Fig.6).15. and this can be calculated geometrically with the crystallographic theory of martensite [163]–[167]. Dark lines show the registry. this θ gives the habit plane of the martensite plate. The process of slip is the trick to make it fit again. 15. however.17. but ensured matching along the austenite interface by performing a smaller amount of slip than at the bottom of Fig.16 – twinning (as described in Sect. however. with additional degrees of freedom and additional constraints of atom matching.4 Crystallographic theory of martensite 369 a θ a(1 + ε) Bain strain Slip Rotate θ a b t Fig. We have assured the alignment of atom planes along the y-direction. The shear components of the strain are 17:32:22 .17a. This extra degree of freedom makes it possible to preserve crystal registry along the austenite interface for different Bain strains.16c.16d show its success in achieving the alignment of atom planes across the interface. they can make the composite square shown in Fig. region is too wide to fit in the original austenite. In three dimensions we have similar considerations. 15. θ = arccos 1+ (15. the idea is that twinning on a fine scale allows length matching across the habit plane and atomic registry. 15. Together. The slip shown at bottom generates a new surface of appropriate length to reconnect to the austenite.16 c d (a) Original austenite crystal.8) For our 2D problem. we could have worked with the same transformed region. θ  cos  1 . 15. If the Bain strain were smaller. 15.2. (b) Pre-selected region and its unit cell. with pre-selected region to be transformed.7) (15. There is an alternative to the slip process shown in Fig. but in two dimensions we also need to consider atom matching in the x-direction. 15. 15. 15. The thick lines in Fig. (c) Bain strain of martensitic transformation for unit cell and the region. Either one of them will cause substantial strains when reinserted into a square lattice. 15.16c – the slip was chosen to give a horizontal surface because this gave a length match to the habit plane by Eq. the Bain strain forces the selection of only one habit plane. and fitting is achieved if 1 1+ = . Again. Suppose a square lattice has the two variants of rectangular transformation products shown in Fig. The slip over the angle θ shortens the surface that needs to fit.17b.8. however.

This potential is graphed in Fig. eliminated.9) where κ2 and κ4 are positive constants that are assumed independent of temperature. 15. 15. (c) Plate composed of multiple units of twins. Note the square geometry of the composite. and the formation of a plate structure to minimize the distortion. At low temperatures.5. Usually there is some atomic volume mismatch after the transformation.17c. There are two minima of φ(x) at $ κ2 . 15. If the volume per atom were to remain the same in the square and rectangular crystals. the composite of Fig. In what follows. (15. Even powers of x are appropriate when the atoms in the crystal have inversion symmetry. 15.18a. This degree of freedom can accommodate a range of transformation strains. with vertical boundaries corresponding to the vertical line shown in panel b. 15. 15. Such a plate is illustrated in Fig. the average thermal displacement x.Diffusionless transformations 370 t Fig. and optimizing the fractions of twins is analogous to the selection of the shear at the bottom of Fig.10) xmin = ± 2κ4 (easily found by setting to zero the first derivative of Eq. as is often the case. It is often composed of twin boundaries between the variants of martensite. 15. Although equal fractions of the twins are shown in Fig. (15.9).5 Landau theory of displacive phase transitions 15. (b) Mutual arrangement of the two variants to minimize shear strains on a macroscale.17b would fit without strain into the parent square crystal.1 Displacive instability A theory for a displacive phase transition can be developed using the “Landau potential” φ(x) = −κ2 x2 + κ4 x4 . x is the displacement of an atom off a symmetric position in the center of a unit cell.17c.17 a b c (a) Two variants of a rectangular phase that are transformation products from a parent square lattice. it is possible to change the fractions of the twins by changing their widths.16c. 15.

15. however. and neither the atomic potential nor the surrounding atomic structure has inversion symmetry in ±x. At high temperatures. The unit cell is distorted at T = 0. The potential of Eq. is expected to be near one of the minima of φ(x).9 is sampled as an even function of x.18a is small in height compared with the thermal energy. 15. the thermal average displacement is x. so at high temperatures. the bump at the bottom of the Landau potential in Fig.

where 17:32:22 . A phase transition is therefore expected at an intermediate temperature. = 0.

(b) Least-squares fits of a parabola a + bx 2 to the curve of part a. The different ranges of the fits are indicated in both graphs by shaded end points. 15. and the crystal takes on a distorted configuration with two values of x.15. the symmetry of this high-temperature phase is lost.5 Landau theory of displacive phase transitions 371 4 Potential (φ(x)) 3 2 1 0 –1 –2 t Fig.18 a –1 0 1 Displacement (x) 2 –2 –1 b 0 1 Displacement (x) 2 (a) Graph of the function −x 2 + x 4 .

reaching x..

18a to a harmonic potential.18a. as at high temperatures. With an increasing range of x. as indicated by the gray endpoints in Figs.15) When κ is positive.5 The quadratic potential depends on the range in x over which we perform the fit. 15. 15.11) (15. and a normal oscillation occurs. Fr = − (15.b. $ κ ω= . the average restoring force vanishes. i. The atom vibrational frequency from the equation of motion.13) (15. Typical results are shown in Fig. dx Fr = −κx . Eq. Fits near x = 0 sample mainly the bump in the middle of the potential.12) At a temperature where the atom displacement is approximately between the dots in Fig.14) (15. 17:32:22 . we have the condition κ = 0. 15. ma = Fr .18b. 15.18a for the range −1. = ±xmin at T = 0. but the qualitative behavior is the same. This pathological vanishing of κ marks an instability. dt2 x(t) = x0 sin ω t . where κ is negative. In a big approximation that will be handy for calculating the vibrational entropy. the frequency ω is positive. is m d2 x = −κ x . When the range of x is large. At low temperatures. we fit the potential of Fig.08 < x < +1. emphasizing the higher parts of the potential (where the kinetic energy and velocity are smaller). the average potential is strongly concave upwards. giving a quadratic potential that is concave downwards.. 15.08.e. the average fit flattens out.15 shows that ω is 5 These were simple least-squares fits that gave equal weights to all values of x within the range. m (15. It would be more appropriate to weight by the amount of time spent at each value of x. The average restoring force Fr against the displacement x is dφ .

These additional terms are from atom vibrations 1 d2 x . where the structure does not oscillate about its original value.9. Imaginary frequencies are the signature of an unstable vibration. a kinetic energy and an entropy are needed. 15. but makes a displacive transition into a new structure. F = E − TS with the potential energy of Eq.2 Free energy To develop a thermodynamic free energy.5. 15.Diffusionless transformations 372 imaginary.

− κ2 x.

2 + κ4 x.

(15.16) m 2 dt2 The substitution x → x.4 − T Svib (x) .

where x..

but instead we seek the temperature where there is an instability of the high-temperature free energy against an infinitesimally small atom displacement. is a thermal average. reflects a change in how we use the Landau potential. We will not be comparing the free energies at different temperatures. x.

so the vibrational entropy is obtained from Eq. 14.4 and 14. We simplify to the Einstein model. repeated here for a three-dimensional material .. 7.52.5 for finding the temperature of instability against a concentration wave (r) with an infinitesimal amplitude η. This is the same approach used in Sects.

(15.ω (T) 3 kB ln ω F(T) = The dependence of ω of Eq. 15. k T  B +1 .15 on x.17) Svib.

is parameterized as   2 x.

γ 2 ω2 = ω02 + γ2 x.

6 We seek the difference in free energy with and without the small displacement x.18) which should be adequate if ω is not much different from ω0 .2 = ω02 1 + . ω02 (15.

16.. and since x. The kinetic energy terms cancel in Eq. 15.

is small. the small fourth-order term κ4 x.

4 is ignored F(T) ≡ F(T. x.

x.) − F(T.

F(T) = −κ2 x. = 0) .         kB T kB T +1 + T 3kB ln +1 .

2 − T 3kB ln ω ω0  ⎞⎤ ⎡ ⎛  2 ω02 1 + γ2 x.

ω02 ⎢ ⎜ ⎟⎥ ⎜ ⎟⎥ . ln F(T) = −κ2 x.

2 + T 3kB ⎢ ⎣ ⎝ ⎠⎦ 2 ω0 F(T) = −κ2 x.

2 + T 3kB γ2 x.

22) 6 Using a harmonic entropy is risky because φ(x) is not harmonic.20) (15. ω02 (15. 17:32:22 . where the frequency of the nonlinear oscillator increases with displacement. Nev- ertheless.19) (15.21) (15. this model captures the essential behavior above Tc . so quantitative results are not expected.2 .

3 Critical temperature and equilibrium phases Our goal is to find the temperature at which F(T) becomes negative (unstable) for an infinitesimal displacement x. 15.5 Landau theory of displacive phase transitions 373 The approximation ln(1 + δ) δ was made in Eq. and F is normalized to the number of atoms.5.22. 15.15. N.

We find the trivial solution x..

but at low T we see that F(T) can be negative with a finite x. = 0 for high T.

In both cases.22 Tc = + κ2 ω02 . which was Tc = zV/kB . 3kB γ2 (15.23 as kB ln(1 + γ2 x. The role of vibrational entropy is seen in Eq. We obtain the critical temperature by setting F(T) = 0 in Eq. 15. It seems intuitive that the critical temperature should scale with this potential. 14..23 is the analog of Eq. 15.23) Equation 15.72 for developing chemical order in an alloy. the critical temperature is proportional to the strength of the potential that drives the breaking of symmetry – the development of either chemical order or a small displacement.

2 /ω02 ) kB γ2 x.

If the displacement x.2 /ω02 .

This causes the displacement to be disfavored at finite temperature. 15. causes the vibrations to increase in frequency rapidly in Eq.18. A large ratio of γ2 /ω02 therefore suppresses the critical temperature for the displacive phase transition. the mean displacement x.7 Below the critical temperature. the vibrational entropy decreases rapidly.

and we include it to give a more complete form of Eq. We cannot ignore the quartic term. is no longer infinitesimal.22 F(T) = −κ2 x. 15.

2 + κ4 x.

4 + (Tc − T) 3kB γ2 x.

For a selected T.24) where T ≡ Tc − T is positive below Tc . we seek a minimum in F(T) with respect to x. ω02 (15.2 .

. The condition dF/d x.

23 for Tc gives % 1 3γ2 ) x. 15. = 0 along with Eq.

19 shows that x.25 is graphed in Fig.19. = kB T . (15.25) ω0 2κ4 Equation 15. Figure 15. 15.

goes to zero continuously at Tc .e. This result of Landau theory predicts that the displacive transition is second-order in free energy. when T = Tc ). The slope of x.19 agrees with Eq.8 Figure 15. 15..10 for the displacement at T = 0 (i.

15. is inappropriate at low temperature. For the harmonic fits of Fig. and the curve in Fig. the role of configurational entropy is less obvious because it contributed a numerical factor. The heat capacity of a material goes to zero at T = 0. 7 For concentration waves.18.(T) is not satisfactory near T = 0. 15.17 was used for convenience. so γ2 must also go to zero. too.6. however. 8 Many displacive transitions are first-order. in fact γ2 does go to zero for small displacements. but the Landau formalism can be adapted to handle them as described in Sect. 15. The high-temperature limit of Eq. but it. 17:32:22 .19 must flatten at low temperatures. 15.

0 –0.4 0.5 0.0 Temperature [Tc] 1.19 0.2 1.6 0.5 –1.5 Displacement <x> [(κ 2/2κ 4)1/2] 1. 15.0 t Fig.4 Temperature dependence of displacement parameter x.8 1.0 0.0 –1.2 0.Diffusionless transformations 374 1.5 0.

the breaking of symmetry below Tc gives two values for x.25. x. Because we considered one coordinate. 15. above and below Tc from Eq.

It was the positive quartic term that gave the positive potential at large displacements. The opposing signs of the terms in the free energy with quadratic and quartic displacements (involving the constants κ2 and κ4 ) were essential for generating the structure at the bottom of the free energy function in Fig. 15. 15. 15. In three-dimensional crystals there are frequently more symmetrically equivalent variants of displacements below Tc . Inside a solid material. x.6 First-order Landau theory We can adapt the Landau theory to accommodate first-order phase transitions.18. The trick to adapting to a first-order phase transition proves to be the addition of a positive sextic term in the displacement.26) .6. 17:32:22 (15. and switching the signs of the quartic and quadratic terms. 15. This leads to microstructural complexity like that discussed in Sect. as appropriate when the high-temperature phase is symmetric in the displacement coordinate. these different variants can interact through strain energy. essential for stability at high temperatures. (plus and minus).4.1 Potential energy and low-temperature displacements A Landau potential suitable for first-order phase transitions is φ(x) = +κ22 − κ4 x4 + κ6 x6 . The potential remains an even function of displacement.

27) making the substitution χ = x2 . A larger value of κ4 is interesting.20 –2 –1 0 Displacement (x) 1 2 Graph of the function φ(x) = +1 x 2 − κ4 x 4 + 1 x 6 for κ4 = {0. 1. 2.5 0 3 –1 –2 t Fig. {κ2 . With three independent constants. The values of x at the minima can be found by setting dφ(x)/dx = 0. 15. but some essential ones are shown in Fig.20.20.20. so we take the positive sign and obtain for x % 2κ4 x ± .28 κ 4 ± κ4 χ . 15.29) 3κ6 We have no interest in the trivial solution χ = 0. The shape at intermediate values of x depends on the relative strength of the quartic component. Eq. 17:32:22 . giving two more minima in the potential that are separated by energy barriers from the minimum at x = 0. This is especially evident for κ4 = 2 in Fig.20. These minima are away from x = 0 where the term κ2 x2 is important. (15. At small values of x the quadratic component dominates.20. 15. For small values of κ4 there is only the one minimum at x = 0.5. κ4 . and at large values of x the sextic component dominates. 15. 15. 15. κ6 }. 15. (15. 3} as labeled. Using the quadratic formula for a standard solution for χ ) κ4 ± κ4 1 − 3κ2 κ6 (κ4 )−2 . from which we obtain the quadratic equation 3κ6 χ 2 − 2κ4 χ + κ2 = 0 .28) χ= 3κ6 We are interested in the two deep minima in Fig. but this potential is not useful for modeling a phase transition. and we also assume that κ4 is relatively large. (15. the function φ(x) can take on a number of shapes.15. (15.6 First-order Landau theory 375 2 0 1 Potential (φ(x)) 1 2 2.30) 3κ6 For the larger values of κ4 in Fig. 15. We therefore ignore the second term in the radical in Eq. and a typical variation is shown in Fig. 2.30 is accurate to a few percent.

20. with a sextic term and the switch of signs of the quadratic and quartic terms described above F(T) = 1 d2 x . the Landau free energy is like that of Eq.2 Free energy minimization and critical temperature For first-order transitions. 15.Diffusionless transformations 376 15.6.

+ κ2 x.

2 − κ4 x.

4 + κ6 x.

The decrease of vibrational entropy with x. 15.2.5. where it was handled approximately by assuming three Einstein modes per atom.31) The vibrational entropy was discussed in Sect.6 − T Svib (x) . m 2 dt2 (15. with a frequency that increases with temperature as parameterized by γ2 .

Anticipating a first-order phase transition. disfavors this displacement at higher temperatures. we seek the condition of equal free energies for a phase with finite x.

and a phase with x.

x. = 0 F(Tc .

x.c ) = F(Tc .

= 0) .32) where we expect x. (15.

c to jump discontinuously from 0 to a finite value as the temperature is reduced below Tc . The difference between these two free energies is   γ2 x.

(15.33) F(T) = +κ2 x.2 2 4 6 .

− κ4 x.

+ κ6 x.

+ T 3kB ln 1 + ω02   γ2 F(T) = κ2 + T 3kB 2 x.

2 − κ4 x.

4 + κ6 x.

F(Tc ) = 0 for the finite displacement x.34) ω0 We seek two conditions for F(Tc ) at the critical temperature • First.6 . (15.

(The two phases are in equilibrium at Tc .c . this x.) • Second.

so dF(T)/dx| x.c corresponds to a minimum of F(T).

(Otherwise a different value of x.0 = 0.

would be more stable.) To find the displacements x.

c for the minima. set dF/d x.

(15. = 0 and consider only the dominant quartic and sextic terms. The minima of Eq.35) x.30 are recovered % 2κ4 . 15.

0 = ± 3κ6 In cases where the order parameter is not strongly dependent on temperature. this x.

20 are dominated by the negative quartic and positive sextic parts of the potential. 15.0 of Eq. but the quadratic part is important for the temperature dependence and the magnitude of F(T. The deep minima of Fig. x. 15.35 can be used to obtain the critical temperature.

15.). 15.34.36) Tc = 3kB γ2 κ6 17:32:22 . setting F(Tc ) = 0. (15. and after some rearrangements   ω02 2κ42 − κ2 . Substituting Eq.35 into Eq.

If there is a change of x. The coefficients of the potential also set the Tc in a similar way. 15.23 for the second-order transitions.and second-order Landau potentials.7 Crystal instabilities and phonons 377 Compared to Eq. although the sign of κ2 is switched for first. this Tc shows a similar role of vibrational entropy through ω02 /γ2 .15.

” and for a cubic crystal the conditions on the elastic constants are [64. 15. The other two criteria involve instabilities where shear distortions are favored to grow. there must be a positive cost in elastic energy when the crystal is distorted. from B = (C11 + 2C12 )/3 (Eq. 15. In other words. with temperature.97). (15.7. as is expected. B > 0. Of particular interest in the study of martensite is the stability with respect to smallamplitude shears on (110)-type planes in 110.36 can be semiquantitative. 6. so its atoms come apart. the value of Tc in Eq.37) Each of these stability criteria corresponds to a different mode of deformation. C11 − C12 > 0 .1 Criteria for crystal stability A stable crystal provides restoring forces against shear and dilatation strains that would change its shape. C11 + 2C12 > 0 . For example. we see that the second inequality demands a positive bulk modulus. An analysis of the elastic energy gives the “Born stability criteria. 168] C44 > 0 . A negative bulk modulus causes a severe instability – the crystal reduces its elastic energy by expanding.7 Crystal instabilities and phonons 15.

(15. 15. 15. η =  .39) If a particular eigenvalue vanishes at a given temperature. η =  . 6) . A stability analysis from a free energy with only the first summation in Eq. j = 1. the crystal shears spontaneously in a “soft mode instability.38) i=1 j=1 k=1 where F0 is the free energy of the strain-free material. If C11 − C12 vanishes at a given temperature. η =  and η =  . and Voigt notation for strains η is used. 1 11 2 22 3 33 4 23 5 31 6 12 17:32:22 . the material is unstable with respect to the deformation mode of the associated eigenvector.” The earliest explanations of martensitic transformations [169.9 Stability with respect to strain is governed by the sign of each of the six eigenvalues of the 6 × 6 matrix of second derivatives ∂ 2F = Fij ∂ηi ∂ηj (i. the stability 9 Here η =  . Later. η =  . directions. 170] were based on soft mode instabilities.37. Clapp [171] wrote the free energy of a crystalline solid as 1  1  Cij ηi ηj + Cijk ηi ηj ηk + · · · 2 3! 6 F = F0 + 6 i=1 j=1 6 6 6 (15. To account for a phase transformation that depends on temperature.38 will reproduce the inequalities of Eq.

15. For such vibrations. 15. The parameter ξ gives the magnitude of the wavevector as ξ [111]. This is.5. For a crystal with harmonic vibrations.40) Expressions such as Eq. 24. The transverse dispersion. has a large softening near the wavevector of 2/3[111] that depends on temperature.Diffusionless transformations 378 30 T Energy [meV] 25 L 20 15 10 5 0 0. an instability of the atom movements in this vibrational exp[+ωkj  mode.11 for individual atom vibrations contain factors of exp[−i(k · rl − ωkj  t)].6 0.0 t Fig.8 Reduced wavevector (ξ) 1. a general stability criterion is that the vibrational frequencies must be real 2 ωkj  >0  . t]. 15. has a normal stiffening with wavevector.7. but the longitudinal dispersion. ∀ kj (15. of course. 174. too. labeled “L”. Phonon instabilities can exist at short wavelengths. At long wavelengths. 15.21 0. This was the role of vibrational entropy in the Landau theory of Sect. against a lattice distortion must be temperature-dependent.2 Soft phonons in bcc structures In general. labeled “T”.0 Phonon dispersion curves for bcc Ti along [111] at elevated temperature.37.4 0. the bcc structure is weak against longitudinal waves with 2/3 111.2 0. this instability can be equivalent to the Born stability criteria of Eq. an imaginary ωkj  allows for amplification with time. 175]. After [173.

At temperatures near a transition to the hexagonal phase. Phonon dispersion curves of all bcc elements show a dip in frequency of their longitudinal dispersion near this wavevector. because the bcc structure is not stable with first-nearest-neighbor (1nn) central forces alone. The dip at the dashed vertical line is quite prominent. wavevectors.) Many bcc crystals also have phonon dispersions that soften at specific frequencies along the transverse [110] branch. becoming unstable. This behavior could originate with a temperature-dependent change of second nearest-neighbor force constants [172]. phonon dispersions of fcc crystals tend to increase monotonically to the edge of the Brillouin zone. or at least an anomaly. 15. these low-frequency transverse modes dip towards zero frequency. 10 In contrast.10 The softening of [111] phonons in bcc Ti is shown in Fig.21. but the specific wavevector of these soft modes varies for different alloys. 17:32:22 .

the restoring forces will be nonlinear functions of displacement. the three atoms with up arrows and the three with the horizontal lines make a hexagonal arrangement.379 15. notice that pairs of them move together. A crystal structure with these six atoms on the same plane is the ω-phase. the shortest wavelength would have the stationary three atoms moving downwards. The diffusionless ω-phase transformation in Zr is probably driven by this phonon. Three atoms have horizontal lines indicating they are stationary. CC BY28. What is interesting about this mode is that its displacements can collapse the bcc structure. The mechanics of how a soft mode causes structural changes is not yet well understood. 15.23. If the mode shown in Fig. and one downwards.7 Crystal instabilities and phonons Fig. 15. the interatomic force constants are not expected to behave as for small displacements. We do know that the interatomic forces in bcc transition metals are stiffest along the 111. so the repeat distance would be two planes instead of three as labeled in the figure (hence the 2/3[111] designation for the soft mode). Again. The atom displacements in a 2/3[111] phonon are shown in more detail in Fig. which is one unit cell of the bcc structure balanced on its corner.0. and “ω-phase collapse” is an instability of the bcc structure.e. i. and is a type of martensitic transformation.. with arrows showing atom displacements in a soft 2/3[111] longitudinal phonon. For comparison. although they are separated by planes that are stationary. Three arrows show atom displacements upwards. these six atoms will briefly form a hexagonal plane.22. This is a longitudinal vibration (phonon) of wavevector 2/3[111].22 becomes large in amplitude. With large atom displacements. Planes of equivalent atoms extend out of the plane of the paper. 15. however. The horizontal lines in the figure identify planes of atoms with the same displacements.22 Large model of one unit cell of a bcc crystal. Figure courtesy of André Waterkeyn. t The soft atom displacements in the bcc structure are shown in Fig. 15. Viewed from above.

23 to move together.3 Phonons and entropy Many bcc metal crystals are unstable energetically at low temperatures. This would be consistent with the tendency for all vertical chains in Fig.7. where a closepacked structure is preferred. directions of first-nearest neighbors. the bcc–fcc transformation cannot cause 17:32:22 . 15. 15. In pure elements.

24 were obtained from inelastic neutron scattering. Friedel suggested some time ago that soft transverse modes stabilize the bcc structure at elevated temperatures [176].50 it was found that the vibrational entropy of the martensitic transformation was quite close to the full entropy measured by calorimetry. Other considerations of entropy changes during martensitic transformations and twinning in alloys were described by Cahn [177]. Cu–Al–Ni.5 ± 0. and Cu–Al–Be alloys. From the measured heats from Cu–Al–Zn. The phonon DOS curves of the austenite and martensite of Fig. the entropy was obtained as S = H/T. Notice how close-packed columns along [111] move as units (as do entire (112) planes). A typical value of S was 0. with white atoms in the layer above the black. Mañosa et al. 15. H. with Eq. and the temperatures of transformation. From these data. Owing to the weakness of bcc structures against various distortions. Planes.2 kB /atom [178]–[180].23 (a) Standard bcc unit cell (plus two centered atoms from adjacent cells). the vibrational entropy of austenite was found to be larger than that of martensite because the transverse acoustic (TA(2)) and longitudinal acoustic modes of austenite are softer than those of martensite. After [173]. with wavefronts shown as horizontal lines. They argued that the martensitic transformation temperature depends on the presence of low-energy modes in the high-temperature phase that help stabilize it. as measured by calorimetry [181].380 Diffusionless transformations Fig. Ortín. which may also be responsible for the soft modes involved in the mechanism of the martensitic transformation. which tend to have good thermodynamic reversibility. performed a number of calorimetric studies of the heats of martensitic transformations in shape-memory alloys. t any change in configurational entropy because the shear and shuffle processes of atom movements create the product phase deterministically. Simulations suggest that this originates with a large and negative 1nn transverse force constant.05 kB /atom. and the transformation is promoted when these modes are depopulated at lower temperatures. 7. Note how the transverse modes of austenite are prominent at low energies around 10 meV. A transverse mode with k = 1/3 [112] and a [111] polarization has the same atom displacements. (b) Atoms in two stacked (110) planes. oriented on end and almost rotated into the position for correspondence to panel b. From simulations and the measured DOS. 15. 17:32:22 . The martensite-to-austenite transformation in NiTi has a latent heat of 0. Arrows show displacements of the atoms in a longitudinal phonon of k = 2/3 [111].

(a) Derive a form of the Clapeyron–Clausius equation that predicts how much the temperature of γ /α  equilibrium is shifted by a magnetic field.24 0 10 30 20 E [meV] 40 50 Phonon DOS curves of NiTi at different temperatures and different phases as labeled [181].7. (b) For a magnetic field of 100. how much of a change is expected in the temperature of γ /α  equilibrium? Assume the entropy of the austenite phase is 0.2 (a) Calculate the angle of shear for twinning in an fcc crystal. and the fcc austenite phase (γ ) is not. where Mα  is 2. Is is possible to make a twin in an hcp structure by our pole mechanism and close-packed planes? Why or why not? 15. (c) Twinning is possible in hcp crystals.2 μB /atom (the Bohr magneton μB = 0. 15. (b) A pole mechanism may be designed for this fcc→hcp transformation. whereas Mγ is nearly 0.Problems 381 60 × 10–3 Martensite 150 K Austenite 573 K g(E) [meV] 40 20 Austenite 300 K 0 –20 t Fig. You can do this conveniently by using the registry notation of letters and arrows written for the three cases of slip in Fig. 15.5 kB /atom greater than that of martensite.1a.6b. (c) Do you expect an effect of the magnetic field on the nucleation rate of martensite? Why or why not? 17:32:22 . Problems 15.3 In many steels the bct martensite phase (α  ) is ferromagnetic. 15. (b) Calculate the angle of shear for an hcp to fcc transition as described in Problem 15.1 A phase transformation from fcc to hcp may occur by a mechanism similar to that of twinning in fcc structures. In a magnetic field H. and it will look much like that of Fig. Describe the differences of the pole mechanism from that for twinning in fcc crystals.000 Oe (10 T). 15.92 × 10−20 erg/oersted). the difference in magnetization M can lower the free energy of the martensite with respect to austenite by the amount Mα  H. (a) Show that the ABAB stacking of the hcp structure is generated by the passage of a Shockley partial dislocation on alternate close-packed planes.

15.18.382 Diffusionless transformations 15. going instead as ω2 = ω02 + β x.4 Suppose the vibrational frequency ω were more sensitive to the displacement than in Eq.

4 .16d. differ from Eq.16 is that the Bain strain from b to c would have some unit cells with tetragonality along x and others along y. (15. 15.41) How will the critical temperature.) (a) By alternating unit cells with the two tetragonalities.34. The modification to the diagram of Fig. (They are all along x in Fig.16c. 15. 15. show that the change with temperature in the minimum of the free energy at x.23? Why? 15. 15.16 included a slip process in part c that allowed refitting the transformed region with crystallographic registry in part d.6 Using the free energy expression for the first-order Landau theory of a displacive transition in Eq. An alternative to slip is twinning. 15.5 The martensitic transformation in Fig. using the same fraction of each variant (1/2). Tc . (b) Is the angle of the habit plane θ the same as if you had performed a shear process instead of twinning? 15. construct a transformed region that can fit back in the original crystal as in Fig. 15.

(15.42) F = 2kB (T − Tc ) 2 ω0 κ 6 17:32:22 .0 is approximately γ 2 κ4 .

An overview of phenomena in nanomaterials is well beyond the scope of this text (see [184]).75 kB /atom would null the free energy difference between the nanostructured material and the bulk material. including surface relaxation and reconstruction processes that are driven by chemical energy. Chapter 16 discusses the thermodynamics of interfaces. and a large fraction of atoms located at. A more basic question is how the thermodynamics of nanostructured materials differs from conventional bulk materials. interfaces. magnetics. The confinement of structural defects such as dislocations or internal interfaces alters their energies and interactions. a nanostructured material with a grain boundary enthalpy of 3 kJ/mole [182.. i.6 and 11. • Nanomaterials confine electrons. surfaces. optics. Some basic issues for the confinement of electrons in nanostructures are presented here.16 Thermodynamics of nanomaterials Nanostructured materials are of widespread interest in science. 17:32:37 . and technology. The unique properties of nanomaterials are being studied with an eye towards applications in electronics. or composition gradients in nanostructures. and elastic energy. These entropy contributions tend to stabilize a nanomaterial at finite temperatures. it is useful to define nanomaterials as materials with structural features of approximately 10 nm or smaller. For the purpose of thermodynamics. but Sections 6. mechanics. but three examples are given of thermodynamic phenomena concerning surface reactions. phonons. chemistry. or near. In short. The free energy of nanostructured materials is altered by the entropy from the configurations of nanostructural degrees of freedom and their excitations. the material might find a different crystal structure possessing some of the local atomic arrangements of the nanocrystal. and medicine. engineering. Although such a thermodynamic stabilization has not yet been demonstrated.1 entropic effects alter the stability of nanomaterials. biology. Important physical properties of nanomaterials originate from one or two basic features: • Nanomaterials have a high surface-to-volume ratio. or polarons to relatively small volumes. altering their energies.2 covered important aspects of surface energy. At a temperature of 500 K. tens of atoms across. magnetism. 183]. Consider. A practical question is whether nanostructures are adequately stable at modest temperatures. an entropy of 0.e. for example. too. their internal energy is raised by the surfaces. 1 If the nanostructure were favored thermodynamically.

1b shows in more detail the extra half-planes of the six edge dislocations between two tilted simple cubic crystals. for example.13b. 16.1 Grain boundary structure The interfaces between adjacent crystals. For a small misorientation between two crystals. The gray dots are sites of the “coincidence lattice.1b and 16. Grain boundaries between crystals of larger misorientation have geometrical complexities owing to crystallographic registry. “grain boundaries. Nevertheless. provide terminating half-planes that offer a tilt like the vicinal surfaces of Fig. however. (c) Coincidence grain boundary for symmetric 53.1c show such conditions for the integer ratios 1/4 t Fig. This dislocation array creates a low-angle tilt boundary when the atoms relax their positions. a rotation of 90◦ around a fourfold axis of a cubic crystal cannot produce a grain boundary because the crystals are matched perfectly by symmetry. A special tilt condition. Other types of boundaries can be constructed from patterned dislocations. such as twist boundaries comprising arrays of screw dislocations.1a are in a stable configuration because the compressive stress above each dislocation cancels partially the tensile stress below its neighboring dislocation.1 a b c (a) Stable structure of edge dislocations. and make a major contribution to the free energy of nanocrystalline materials. occurs for boundaries of tilt angles θ when tan(θ/2) is a ratio of small integers.Thermodynamics of nanomaterials 384 16. but it is usually assumed that once the orientations are fixed. The specification of an individual grain boundary requires a minimum of five coordinates – two for the orientation of each crystal. easily seen in two-dimensional figures.07◦ . (b) Grain boundary constructed with edge dislocations for two cubic crystals tilted by 28.13◦ tilt. Atoms are located on all corners of the dark lattices.” 2 Displacement coordinates may also be required for a full specification. cannot bring two crystals into crystallographically equivalent orientations. The six edge dislocations in Fig. 17:32:37 . Figure 16. Figures 16. For example.2 This complexity allows for numerous types of grain boundaries. Read and Shockley proposed a grain boundary structure composed of an array of dislocations [185]. chemical bonding will select the displacements by nudging the atoms into a favored configuration at the boundary. special tilts can bring a high degree of crystallographic registry at the interface itself. and one for the rotation in the plane of the boundary. Edge dislocations. 6. more dislocations are necessary geometrically.” are important features of polycrystalline microstructures. 16. Most tilt angles. For a higher angle of tilt.

0 bcc Cr-Fe Mo-Fe Fe-Ti Grain boundary spectral fraction 0. In general. neighbor distances.4 fcc 0. 16.07◦ and 53. for example. respectively.2 shows the fraction of 1.6 0. From [190]. and for nanocrystals there can be a substantial fraction of atoms in these unfavorable grain boundary sites. versus the characteristic grain size of the nanocrystals.16.1c. Most crystal orientations cannot accommodate special grain boundary orientations as in Fig.1b. there are crystal orientation angles out of the plane of the paper. Accommodating small deviations from these perfect tilts can be accomplished by adding an occasional extra dislocation. and the orientation of the boundary itself can be varied). More geometric detail is shown in Fig. comprising a “coincidence lattice” for this case of tan(θ/2) = 1/2. Additional positive edge dislocations can accommodate higher tilt angles than that of Fig. and the best fits for fcc and bcc crystals are shown as solid lines. The energy of the grain boundary increases with the numbe