Perfect crystals would be too strong
Bonding in solids ended with a puzzle. Estimate the strength of a metal from its bonds and you get several gigapascals; measure it and you find a thousandth of that. A single crystal of very pure copper starts to deform permanently under a shear stress of about one megapascal, roughly the pressure of an adult standing on a postage stamp. Something lets the atoms slide far more easily than the bonds would suggest.
That something is defects: places where the perfect order of Crystal structures breaks down. Every real crystal contains them, and most of materials engineering is the art of controlling them. Missing atoms let other atoms diffuse, which is how steel is case-hardened and how dopants are driven into silicon. Foreign atoms one in a million change silicon from an insulator to a useful conductor. Line defects called dislocations let metals bend instead of shattering, and blocking them is how every strong alloy is made strong. Boundaries between grains make metals stronger still, and also give corrosion and cracks a path.
Defects by dimension
Defects are classified by how many dimensions they extend in.
| Dimension | Kind | Examples |
|---|---|---|
| 0 | point defects | vacancies, self-interstitials, substitutional and interstitial impurity atoms |
| 1 | line defects | dislocations (edge, screw, mixed) |
| 2 | planar defects | grain boundaries, twin boundaries, stacking faults, phase boundaries, the free surface |
| 3 | volume defects | pores, cracks, inclusions, precipitates |
Some defects are a matter of thermodynamics and some are not. A crystal at any temperature above absolute zero wants a certain number of vacancies, because they raise its entropy, so vacancies are equilibrium defects. Dislocations and grain boundaries cost far too much energy for entropy to pay; they are left over from how the crystal was grown, cast, worked and heated, and annealing reduces them. So point defects are controlled mostly by temperature and composition, and line and planar defects mostly by processing (Processing and heat treatment).
Point defects
Definition— vacancy, self-interstitial
A vacancy is a lattice site from which the atom is missing. A self-interstitial is an extra atom of the crystal's own kind squeezed into a space between sites. In close-packed metals a self-interstitial distorts its surroundings so much that its formation energy is two to four times that of a vacancy, and in equilibrium it is far rarer.
Definition— substitutional and interstitial impurities
A substitutional impurity occupies a normal lattice site in place of a host atom: zinc in copper (brass), nickel in iron. An interstitial impurity is a small atom sitting in a gap between host atoms: carbon, nitrogen, hydrogen or boron in iron.
Definition— Frenkel and Schottky defects
In an ionic crystal, defects must keep the crystal electrically neutral. A Schottky defect is a set of vacancies in the right ratio, for example one missing Na⁺ and one missing Cl⁻ in sodium chloride. A Frenkel defect is an ion moved from its site into an interstitial position, leaving a vacancy behind; it is common for small cations, as for Ag⁺ in silver bromide.
Which impurities dissolve, and how much, depends on size, structure and chemistry. William Hume-Rothery's rules of thumb say that extensive substitutional solubility needs atomic radii within about 15 per cent of each other, the same crystal structure, similar electronegativity and similar valence: copper and nickel, both FCC with radii within 3 per cent, mix in all proportions. Interstitial solubility depends on the size of the holes: carbon dissolves far better in FCC austenite than in BCC ferrite, as Crystal structures noted, and that difference is the basis of steel heat treatment.
How many vacancies? An equilibrium calculation
Making a vacancy costs energy: bonds to the missing atom are broken, and the atom must be put somewhere, for example on the surface. Call this the vacancy formation energy , typically 0.5 to 2 eV in metals. Why would a crystal make any? Because there are enormously many ways to place a few vacancies among many sites, and that disorder lowers the free energy .
Theorem— Equilibrium vacancy fraction
In a crystal of atoms at temperature , with vacancies that do not interact, the number of vacancies that minimises the free energy satisfies
where is the formation energy and the extra vibrational entropy of one vacancy.
Proof
With vacancies among sites, the number of arrangements is , and the configurational entropy is . The free energy relative to the perfect crystal is . By Stirling's approximation, , whose derivative with respect to is . Setting gives , and rearranging gives the result. The second derivative, , confirms a minimum.
∎
The program evaluates this for aluminium and copper, using formation energies of about 0.67 eV and 1.28 eV, the values given by positron-annihilation and dilatometry measurements, with the entropy factor set to 1. It then does what an experimenter does: fits simulated measurements on an Arrhenius plot, against , whose slope is .
import numpy as np
kB = 8.617333e-5 # eV/K
def vacancy_fraction(E_v, T, S_over_k=0.0):
"""Equilibrium fraction of vacant sites: exp(S/k) exp(-E_v / kT)."""
return np.exp(S_over_k) * np.exp(-E_v / (kB * T))
# Approximate measured vacancy formation energies (eV) and melting points (K)
metals = {"aluminium": (0.67, 933), "copper": (1.28, 1358)}
temps = [300, 600, 900]
print("metal E_v/eV " + "".join(f"T = {T} K " for T in temps) + "at melting point")
for name, (E, Tm) in metals.items():
row = "".join(f"{vacancy_fraction(E, T):10.2e} " for T in temps)
print(f"{name:10} {E:5.2f} {row} {vacancy_fraction(E, Tm):.2e}")
# How many vacancies in a cubic centimetre of copper at 1000 C?
sites = 4 / (0.3615e-7) ** 3 # atoms per cm^3 (FCC, a = 0.3615 nm)
print(f"\ncopper: {sites:.2e} sites per cm^3; at 1273 K about {sites * vacancy_fraction(1.28, 1273):.1e} vacancies per cm^3")
print(f"quenched to 300 K, the equilibrium number would be {sites * vacancy_fraction(1.28, 300):.1e} per cm^3")
# Recovering E_v from (simulated) measurements: fit ln c against 1/T (an Arrhenius plot)
rng = np.random.default_rng(7)
T = np.array([700.0, 750, 800, 850, 900])
c_measured = 3.0 * vacancy_fraction(0.67, T) * (1 + 0.03 * rng.standard_normal(T.size)) # 3% noise
slope, intercept = np.polyfit(1 / T, np.log(c_measured), 1)
print(f"\nArrhenius fit: E_v = {-slope * kB:.3f} eV, exp(S/k) = {np.exp(intercept):.2f} (true values 0.67 and 3)")At room temperature the equilibrium fraction is tiny: about for aluminium and for copper. Near the melting point it rises to to (and, with the entropy factor, which is usually between about 2 and 20, to to ): roughly one site in a few thousand is empty. Simmons and Balluffi's 1960 experiment on aluminium, which compared the expansion of a bar with the expansion of its lattice, found about one vacancy per thousand sites at the melting point.
Dislocations: why metals are weaker than ideal
How strong would a perfect crystal be in shear? In 1926 Yakov Frenkel imagined sliding one whole plane of atoms over the one below. As the top plane moves by , the stress needed rises, then falls to zero when the atoms reach the halfway point between sites, and the pattern repeats with the spacing of the atoms in the slip direction. Model it as . For small this must agree with elasticity, with the spacing between the planes and the shear modulus, which fixes
For copper, with GPa, that is about 8 GPa. Real annealed copper crystals slip at about 1 MPa, nearly ten thousand times less. The answer, proposed independently in 1934 by Geoffrey Taylor, Egon Orowan and Michael Polanyi, is that crystals do not slip a whole plane at a time. They slip one line of atoms at a time, by moving a dislocation.
Analogy, not a formal construction
Moving a large rug across a floor by pulling one edge is hard: you must overcome the friction of the whole rug at once. Instead, make a ruck at one end and push the ruck along. Only the small part of the rug inside the ruck moves at any moment, yet when the ruck reaches the far side the whole rug has shifted by the ruck's width. A dislocation is a ruck in a crystal. The analogy is about the ease of moving a localised disturbance, not about friction: crystal planes do not rub.
Definition— dislocation, Burgers vector
A dislocation is a line in a crystal along which the atoms are mismatched, the boundary between a region that has slipped and one that has not. Draw a closed loop of atom-to-atom steps around it, the same number of steps in each opposite direction, a Burgers circuit. In a perfect crystal the circuit closes; around a dislocation it fails to close, and the vector needed to close it is the Burgers vector . (Which end the vector runs from is fixed by a convention, which books state differently; the magnitude and line are what matter here.)
Definition— edge, screw and mixed dislocations
In an edge dislocation, is perpendicular to the dislocation line: the crystal looks as if an extra half-plane of atoms had been inserted, ending at the line. In a screw dislocation, is parallel to the line: the atomic planes form a helical ramp around it, like a multistorey car park. A dislocation whose line meets at any other angle is mixed. A dislocation line cannot simply end inside a crystal: it ends at a surface or a grain boundary, joins other dislocations at a node, or closes on itself as a loop, and is the same all along it.
The dislocation glides on its slip plane, the plane containing both the line and . Each step breaks and remakes only the bonds along the line, so the stress needed is small; when the dislocation leaves the crystal, the top half has been displaced by exactly . Dislocations glide most easily on the densest planes in the densest directions, the slip systems of Crystal structures: in FCC metals, with (0.256 nm in copper), and in BCC, with (0.248 nm in iron).
How easy is "easy"? The program first applies Frenkel's estimate to three metals, then simulates a one-dimensional model of a slip plane due to Frenkel and Tatiana Kontorova: a chain of atoms joined by springs, lying on a wavy substrate that represents the plane below. Pushing every atom with a force , it finds the smallest push that keeps the chain moving, first for a perfect chain, then for a chain with one extra atom squeezed in, a one-dimensional dislocation.
import numpy as np
# 1. Frenkel's estimate: to slide one perfect plane of atoms over another, the shear stress must
# reach about G b / (2 pi a), roughly G / (2 pi) when b ~ a.
for metal, G in [("copper", 48), ("aluminium", 26), ("iron", 82)]: # shear moduli, GPa
print(f"{metal:9}: G = {G} GPa, ideal shear strength ~ G/(2 pi) = {G / (2 * np.pi):.1f} GPa")
print("pure annealed single crystals actually start to slip at about 0.001 GPa (1 MPa)\n")
# 2. A one-dimensional model (Frenkel-Kontorova): atoms joined by springs of stiffness k, lying on a
# wavy substrate V(x) = (1/2)(1 - cos 2 pi x) of period 1, whose largest restoring force is pi.
def forces(x, k, L, F):
nxt = np.roll(x, -1); nxt[-1] += L # a ring of length L
prv = np.roll(x, 1); prv[0] -= L
return k * (nxt - 2 * x + prv) - np.pi * np.sin(2 * np.pi * x) + F
def relax(x, k, L, F, steps, dt):
for _ in range(steps): # overdamped motion: velocity = force
x = x + dt * forces(x, k, L, F)
return x
def slides(k, n_atoms, n_wells, F, steps=2000):
"""Does a push F on every atom keep the chain moving?"""
dt = 0.4 / (4 * k + 2 * np.pi**2)
x = np.arange(n_atoms) * n_wells / n_atoms + 0.01
x = relax(x, k, n_wells, 0.0, steps, dt) # the resting chain
x = relax(x, k, n_wells, F, steps, dt) # push, and let transients die away
x1 = relax(x, k, n_wells, F, steps, dt)
return np.mean(x1 - x) > 0.02 / n_atoms # still moving?
def threshold(k, n_atoms, n_wells):
lo, hi = 0.0, 1.01 * np.pi # bisection on the push
for _ in range(10):
mid = (lo + hi) / 2
lo, hi = (lo, mid) if slides(k, n_atoms, n_wells, mid) else (mid, hi)
return hi
N = 24
F_perfect = threshold(5.0, N, N)
print(f"perfect chain ({N} atoms in {N} wells), k = 5: force to slide = {F_perfect:.3f}")
print("chain with one extra atom (a 'dislocation'):")
for k in [2.5, 5.0, 10.0, 20.0]:
Fc = threshold(k, N + 1, N)
width = np.sqrt(k / (2 * np.pi**2)) # rough width of the squeezed region, in wells
print(f" k = {k:4}: width ~ {width:.2f} wells, force to move it = {Fc:.4f}, {F_perfect / Fc:.0f} times less")The perfect chain needs a push of , the full restoring force of the substrate, because every atom must climb its hill at the same moment. With one extra atom the chain slides under a far smaller push, and how much smaller depends steeply on how widely the misfit is spread: doubling the spring stiffness from 10 to 20 cuts the force from about 0.30 to about 0.04. In a real crystal this force, the Peierls stress, is tiny in FCC metals, whose dislocations are wide, and large in covalent crystals such as silicon and diamond, whose directional bonds keep dislocations narrow. That is why copper is ductile at any temperature, while silicon is brittle at room temperature and deforms plastically only when hot.
Dislocation energy, density and hardening
A dislocation strains the crystal around it, and that strain stores energy, roughly per metre of line. The dislocation density is the total length of dislocation line per unit volume, measured in m/m³, or m⁻². Annealed metals have about to m⁻²; heavy cold work raises this to m⁻² or more. Dislocations crossing a slip plane act as obstacles (a "forest") to those gliding on it, so the stress to keep deforming grows with density, as with about 0.2 to 0.5: this is work hardening, the reason a paper clip gets harder to bend back and forth.
import numpy as np
eV = 1.602176634e-19
G, b = 48e9, 0.2556e-9 # copper: shear modulus (Pa) and Burgers vector a/sqrt(2) (m)
# Energy of a dislocation line, roughly G b^2 / 2 per metre
line_energy = G * b**2 / 2
print(f"line energy ~ G b^2/2 = {line_energy:.2e} J/m = {line_energy * b / eV:.1f} eV per atom length along the line")
# Dislocation density: total line length per unit volume (m per m^3 = m^-2)
print("\n state rho/m^-2 line length in 1 cm^3 stored energy/(J/m^3) Taylor term/MPa")
alpha, M = 0.3, 3.06 # Taylor constant and Taylor factor for FCC polycrystals (typical values)
for state, rho in [("well annealed", 1e10), ("annealed", 1e12), ("lightly worked", 1e14), ("heavily worked", 1e15)]:
length_km = rho * 1e-6 / 1e3
energy = rho * line_energy
sigma = M * alpha * G * b * np.sqrt(rho) / 1e6
print(f" {state:15} {rho:9.0e} {length_km:14.3g} km {energy:12.3g} {sigma:8.1f}")
print("(the Earth-Moon distance is about 384 000 km)")
print(f"stored energy at 1e15 per m^2: {1e15 * line_energy * 7.11e-6:.0f} J per mole of copper (molar volume 7.11 cm^3)")
# Grain boundaries: the Hall-Petch term grows as d^(-1/2)
print("\ngrain size d d^(-1/2) relative to 100 micrometres")
for d_um in [100, 25, 10, 1, 0.1]:
print(f" {d_um:7} um {np.sqrt(100 / d_um):5.1f}")
# Doping: one phosphorus atom per million silicon atoms
n_Si = 8 / (0.54310e-7) ** 3 # silicon atoms per cm^3 (diamond cubic)
n_P = n_Si * 1e-6
n_i = 1.0e10 # intrinsic electron concentration at 300 K, per cm^3 (approximate)
print(f"\nsilicon: {n_Si:.2e} atoms per cm^3; 1 ppm of P gives {n_P:.1e} electrons per cm^3,")
print(f"about {n_P / n_i:.0e} times the {n_i:.0e} per cm^3 of pure silicon at room temperature")Each atomic length of dislocation line costs about 2.5 eV, far more than the thermal energy (0.026 eV at room temperature) and far more than any entropy gain could repay, which is why dislocations, unlike vacancies, are not equilibrium defects. A heavily cold-worked cubic centimetre of copper contains about a million kilometres of dislocation line, more than twice the distance to the Moon, and the dislocation term in its yield stress rises from about 1 MPa when annealed to about 360 MPa (here the shear stress is converted to a tensile one with the Taylor factor for FCC polycrystals). The stored energy, about 11 J per mole, is small, but it is the driving force that makes cold-worked metals recrystallise when they are annealed.
Grain boundaries and other planar defects
Most engineering metals and ceramics are polycrystals: aggregates of grains, each a single crystal in a different orientation, typically 1 to 100 micrometres across. Where two grains meet there is a grain boundary, a layer a few atoms thick in which the atoms cannot satisfy both lattices. If the misorientation is small (below about 15°), the boundary is a neat wall of dislocations, a low-angle boundary; larger misorientations give high-angle boundaries with a more open, disordered structure. Some special orientations fit unusually well: a twin boundary is a mirror plane across which the lattice is reflected, and in FCC metals it is a simple change in the stacking sequence, ABCABC to CBACBA. A stacking fault is a single mistake in that sequence, ABCABABCABC, a thin slice of HCP inside FCC.
Grain boundaries matter in several ways at once:
- Strength. A dislocation cannot easily cross into a grain whose slip planes point elsewhere, so dislocations pile up at boundaries. The more boundaries, the stronger the metal: the Hall–Petch relation (Eric Hall, 1951; Norman Petch, 1953) says the yield stress rises as with grain size . As the program shows, refining grains from 100 μm to 1 μm multiplies the boundary term by ten. Grain refinement is the only common strengthening mechanism that also improves toughness, which is why controlled rolling of fine-grained steels became so important. Below roughly 10 to 20 nm the trend reverses, because grains become too small to hold dislocation pile-ups and the boundaries themselves start to slide (Glasses and nanomaterials).
- Diffusion and corrosion. The open structure of boundaries makes them fast paths for diffusion and preferred sites for chemical attack, which is why etching a polished metal reveals its grains and why some stainless steels corrode along their boundaries if wrongly heat-treated (Fatigue, creep and corrosion).
- High-temperature behaviour. At high temperature boundaries slide and let the metal creep, so the most demanding turbine blades are cast as single crystals with no grain boundaries at all.
- Electrical behaviour. In ceramics and semiconductors boundaries can trap charge and block current, a defect in solar cells and the working principle of the zinc oxide varistors that protect circuits from voltage surges.
How defects control properties
The engineer's view is that defects are knobs to turn.
Doping semiconductors. Replacing one silicon atom in a million with phosphorus, which has one more valence electron, adds about free electrons per cubic centimetre, some five million times the intrinsic concentration of pure silicon at room temperature. Boron, with one fewer electron, creates holes instead. Every transistor and solar cell is built from such deliberately placed point defects; see semiconductors and diodes in Lumenary EE and Semiconductor materials.
Strengthening metals. Every strengthening mechanism works by making dislocations harder to move: solute atoms that distort the lattice (solid-solution strengthening), more dislocations to tangle with (work hardening), more grain boundaries (grain refinement), and fine particles of a second phase that dislocations must cut through or bow around (precipitation hardening, which makes aircraft aluminium alloys and nickel superalloys strong). These are the subject of Elasticity, plasticity and strengthening.
Diffusion. Atoms in a crystal move almost entirely by jumping into neighbouring vacancies, or, if small, from one interstitial hole to the next. The vacancy concentration computed above is therefore one factor in every diffusion rate in a metal, the subject of Diffusion.
Ionic conduction. Zirconia doped with yttria contains one oxygen vacancy for every two yttrium ions, and oxide ions hop through these vacancies. Hot yttria-stabilised zirconia is the electrolyte of solid oxide fuel cells and of the oxygen sensors that tune the fuel mixture in car engines.
Colour and quantum technology. An electron trapped at an anion vacancy in an alkali halide absorbs visible light, giving irradiated salt crystals colour; these were named Farbzentren, colour centres. In diamond, a nitrogen atom next to a vacancy forms the nitrogen–vacancy (NV) centre, whose electron spin can be read out with light at room temperature; as of 2026 NV centres are used as nanoscale sensors of magnetic fields and are studied, with similar defects in silicon carbide, for quantum networks (Two-dimensional and quantum materials).
Seeing the invisible
Dislocations were invented on paper before anyone could see one. After Frenkel's estimate (1926) exposed the gap between ideal and real strength, Taylor, Orowan and Polanyi proposed the edge dislocation in 1934, and Johannes Burgers added the screw dislocation and the vector named after him in 1939. For twenty years the evidence was indirect. Then in 1956, at the Cavendish Laboratory in Cambridge, Peter Hirsch, Robert Horne and Michael Whelan filmed dislocations moving in thin aluminium foils in a transmission electron microscope, and Walter Bollmann in Geneva independently imaged them in stainless steel. In the same year S. S. Brenner, at General Electric in the United States, pulled tiny iron "whiskers", single crystals a few micrometres thick that grew nearly free of dislocations, and found strengths of several gigapascals, approaching Frenkel's ideal: the clearest proof that ordinary metals are weak because of their defects.
The professional view
Defects are measured and modelled with a whole toolkit. Transmission electron microscopy images individual dislocations and stacking faults, and aberration-corrected instruments resolve the atomic structure of dislocation cores and grain boundaries; electron backscatter diffraction in a scanning electron microscope maps the orientation of every grain over square millimetres (Microscopy). Positron annihilation spectroscopy detects vacancies, because positrons are trapped in them and live longer there. X-ray peak broadening, as in X-ray diffraction, estimates dislocation densities.
On the computational side, the formation energies of point defects, including charged defects in semiconductors and oxides, are routinely computed by density functional theory in large periodic supercells, with corrections for the artificial interactions between periodic images (Density functional theory). Molecular dynamics with interatomic potentials simulates dislocation cores and their interactions with solutes and precipitates (Atomistic simulation), and discrete dislocation dynamics follows millions of dislocation segments to predict work hardening, bridging to the continuum models of Multiscale modelling.
In industry, defect control is everyday work: silicon wafer makers grow crystals with almost no dislocations and control the oxygen and vacancy clusters that remain; steelmakers control grain size by rolling schedules and small additions of niobium, titanium and vanadium; and the makers of turbine blades eliminate grain boundaries altogether.
Common misconceptions
Misconception
"Defects are flaws that a good material would not have." Many are essential. Without dislocations metals could not be shaped; without dopants there would be no electronics; without vacancies atoms could not diffuse, so no heat treatment would work. The aim is control, not elimination.
Misconception
"A perfect crystal can be made by cooling slowly enough." Slow cooling and annealing remove most dislocations and grain boundaries, but vacancies are equilibrium defects: at any temperature above absolute zero a crystal has a non-zero equilibrium concentration of them. Kinetics, not effort, limits how close to that equilibrium a real crystal gets.
Misconception
"A dislocation is a line of missing atoms." No atoms are missing: an edge dislocation is the edge of an extra half-plane, and a screw dislocation has no extra atoms at all. The defect is in the arrangement, which is why a dislocation can move through a crystal by small shuffles of atoms without any of them travelling far.
Misconception
"Plastic deformation is the bonds stretching further than in elastic deformation." Elastic strain stretches bonds; plastic strain is permanent because dislocations glide and atoms change neighbours. After the load is removed, the bonds return to their normal length, but the crystal has slipped.
Exercises
1. Defects and dislocations
Warm-upHumanWhich of these are planar (two-dimensional) defects? Select all that apply.
2. Defects and dislocations
CoreHumanCopper has a vacancy formation energy of about 1.28 eV. Ignoring the entropy factor, by what factor does its equilibrium vacancy fraction increase when it is heated from 900 K to 1200 K? Use eV/K and round to the nearest whole number.
3. Defects and dislocations
CoreHumanAluminium has a shear modulus GPa. Using Frenkel's estimate , what is the ideal shear strength of a perfect aluminium crystal in MPa, rounded to the nearest hundred?
4. Defects and dislocations
CoreHumanA dislocation line runs along in an FCC crystal, and its Burgers vector is . What kind of dislocation is it?
5. Defects and dislocations
CoreHumanSilicon and copper both contain dislocations. Why is silicon brittle at room temperature while copper is ductile?
6. Defects and dislocations
CoreComputationalWrite arrhenius_fit(T, c) that fits to temperatures T (in kelvin) and vacancy fractions c, and returns (E, A) with in eV. Accept lists or NumPy arrays. Use eV/K and a straight-line fit of against with np.polyfit.
7. Defects and dislocations
StretchComputationalCheck the vacancy theorem without Stirling's approximation. For atoms and vacancies, the free energy in units of is
Write best_vacancy_count(N, e) that returns the whole number that minimises this. Compute with math.lgamma (since lgamma(m + 1)), and search from 0 up to about ten times the estimate , plus a margin.
Further reading and resources
- The mechanism of plastic deformation of crystals. Part I. Theoretical · G. I. Taylor · 1934One of the three papers of 1934 that proposed dislocations to explain why crystals are weak.
- Measurements of equilibrium vacancy concentrations in aluminum · R. O. Simmons and R. W. Balluffi · 1960Vacancies counted directly by comparing a bar's length with its X-ray lattice parameter.
- Direct observations of the arrangement and motion of dislocations in aluminium · P. B. Hirsch, R. W. Horne and M. J. Whelan · 1956The first films of dislocations moving, seen in a transmission electron microscope.
- DoITPoMS teaching and learning packages · University of CambridgeFree interactive packages, including ones on dislocations, slip and point defects.
- The Materials ProjectComputed elastic constants, from which shear moduli and ideal strengths can be estimated.