Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

The pseudopotential panacea

Marvin L. Cohen

Citation: Phys. Today 32(7), 40 (1979); doi: 10.1063/1.2995618


View online: http://dx.doi.org/10.1063/1.2995618
View Table of Contents: http://www.physicstoday.org/resource/1/PHTOAD/v32/i7
Published by the American Institute of Physics.

Additional resources for Physics Today


Homepage: http://www.physicstoday.org/
Information: http://www.physicstoday.org/about_us
Daily Edition: http://www.physicstoday.org/daily_edition

Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms
The
pseudopotential
panacea

Webster's dictionary defines panacea as surfaces and interfaces, chemisorption, TODAY, December 1978, page 44). Here
a "remedy for all ills or difficulties; a electronic transport, lattice vibrations, the electrons are assumed to be nearly free
cure-all." A pseudopotential is an ap- bonding and crystal structure, and ap- (a Fermi gas), but subject to an ad hoc
proximation to the real potential an plications even beyond solid-state physics weak perturbing potential. In the 1950's
electron feels in a solid. In what sense (for example, molecular structure). The this model appeared to explain many de-
can it be called a cure-all? What are the predictive power of theories involving tailed properties of solid metals (for ex-
ills it cures, the difficulties it overcomes pseudopotentials is very impressive. In ample, their Fermi surfaces) and even
or the problems it solves? And how does many cases where predictions from such liquid metals.
it help solve problems? Our object here a theory have seemed to conflict with the However, there are two major reasons
will be to supply some answers to these data, later experiments produced clear why these successes were unexpected.
questions and to describe the growing agreement with the theory. Such suc- First, the kinetic energy of the conduction
influence of theories involving pseudo- cesses have given pseudopotentials the electrons is about the same as their mu-
potentials on solid state or condensed aura of a panacea. tual Coulomb energy. In sodium metal,
matter physics. for example, the average distance between
In the last twenty years, the significant Nearly free electrons conduction electrons is only 2 A, so that
success of solid-state theory in explaining A major conceptual problem of con- the Coulomb energy is around 3.6 eV,
the detailed properties of real materials densed-matter physics is the question of about the same as the kinetic energy.
is due to a large degree to the pseudopo- why the nearly-free-electron model works. The electrons should thus be strongly
tential approach.1 A partial list of ap- The classical free-electron model used by correlated by their electrostatic interac-
plications would include: electronic Paul Drude at the beginning of the cen- tions. How can they then behave as
structure of solids and liquids, electron- tury explained many of the properties of nearly free electrons? Second, the elec-
lattice interactions, superconductivity, metals. With quantum extensions, in the tron-ion interaction is expected to be on
late 1920's, Felix Bloch and Arnold Som- the order of a Rydberg (~13 eV) in energy.
Marvin L. Cohen is professor of physics at the merfeld solved many of the remaining How then can the system behave as a gas
University of California in Berkeley. "outstanding problems" (see PHYSICS of nearly free electrons with the elec-

40 PHYSICS TODAY / JULY 1979 >-U))»U.JU ® IOf 0 iWncan Institute ol Physics


Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms
Potentials for describing the behavior of valence electrons in solids
have considerably advanced our understanding of real materials, including
their optical spectra, bonding, superconductivity and surfaces.

Marvin L. Cohen

tron-ion interaction acting as a small nique for solving for the properties of the when electrons move to fill the vacant
perturbation? These problems were then system is ordinary quantum mechanics. spot, in turn leaving another vacancy.3
probably as perplexing as quark confine- The difficult aspects of the problem are Another example arises when electrons
ment and asymptotic freedom are in the determination of the electron-core act collectively to produce long-wave-
particle physics today. potential, and the large number of degrees length oscillations, as in a plasma. The
of freedom: 10 2 ' particles in a typical oscillations of the plasma can be quan-
Quasiparticles and pseudopotentials three-dimensional material. tized and the collective excitations, or
The absence of strong correlations has In the elementary-excitation model the "plasmons," can be pictured as being re-
its explanation in the Landau "quasi- solid is taken to be a "system responding sponsible for the responses of the system.
particle" picture or the theory of Fermi to probes." A probe such as electromag- Other elementary excitations3 include
liquids. The relative weakness of the netic radiation, a magnetic field, or tem- phonons, magnons, quasi-electrons, pol-
electron-ion interactions is explained by perature change excites the system and an arons, excitons, Cooper pairs, rotons,
the pseudopotential concept.2 Before experimental measurement then deter- Bogoliubons, polaritons, and so on. Each
discussing these conclusions, let us ex- mines the appropriate response function of these types of quasi-particles or col-
plore two of the mental pictures used by such as the dielectric function, the mag- lective excitations exists (has been de-
many solid-state theorists as their models netic susceptibility or the heat capacity. fined) only to suit a particular system and
of microscopic behavior in a solid: the In formulating a theory of the response a particular probe. The interactions of
standard model and the elementary-ex- function, it is often most useful to com- the elementary excitations are again dy-
citation model. pute the responses of the system as if they namical Coulomb interactions, and the
In the standard model, the atomic cores were caused by fictitious particles, the technique for solving for the properties of
(nucleus plus electrons not involved in "elementary excitations." These parti- the system is essentially quantum elec-
bonding) form a periodic array; the va- cles are called "quasi-particles" (usually trodynamics. Quantum electrodynamics
lence electrons in metals move throughout fermions) or "collective excitations" was introduced into physics for relativistic
the solid, behaving like free particles, (usually bosons). One of the best-known problems where particles can be created
except for their weak interactions with the examples of a type of quasi-particle is the and destroyed, so it is also very useful for
cores. The "elementary" particles of the holes that are responsible for many of the the elementary-excitation model, where
standard model are cores and valence electrical properties of semiconductors. the particles are fictitious and therefore
electrons, the force is the Coulomb in- Physically a hole is a vacant space in the created and destroyed as the state of the
teraction which is now treated dynami- periodic electron structure of a crystal; it system changes.
cally because of screening, and the tech- moves like a positively charged particle The elementary-excitation model is

WAVE VECTOR q

s
-%EF

A typical real space pseudopotential (black) is weaker than the ionic or A typical reciprocal space pseudopotential. Values of V(q), for wave
all-electron potential (colored line) near the core region. The thin black vectors equal to the reciprocal lattice vectors, G, are indicated by the
line is a simple extrapolation for small r while the solid line extrapolation dots. For very small q the potential approaches (—2/3) times the Fermi
is more representative of most pseudopotentials. Figure 1 energy, which is the screened-ion limit for metals. Figure 2

PHYSICS TODAY / JULY


Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms 1979 41
modelled after the Landau picture of in-
teracting fermions. Landau showed that
the excitations of a system of strongly
interacting fermions were quasi-particles
resembling the free fermions making up
the system: The masses of the quasi-
electrons may differ from the free elec-
trons but the charge, spin, and statistics
are the same. So the excitations of the
electron system in sodium, for example,
can be described as a gas of free quasi-
electrons. This is the basis of the answer
to the puzzle of how supposedly highly
correlated electrons can act free.
What about the problematic weakness
of the electron-core or electron-ion in-
teractions? We have mentioned that
pseudopotentials provide an answer to
that question. Electrons in a solid or
dense liquid experience a potential made 1 2 3 4 5 6 7
up of two very different components. In
or near the cores they feel a strong atomic ENERGY E(eV)
potential, so they have wavefunctions that Reflectivity of silicon. The theoretical curve (color) was obtained using a pseudopotential calcu-
are very much like those of electrons lation.9 The experimental curve (black) is due to H. R. Philipp and H. Ehrenreich. Figure 3
bound in individual atoms. Between the
cores the electrons see an almost constant
potential, and the corresponding wave- time-independent Schrodinger equation in a crystal of cores and free electrons. It
functions are nearly plane waves. This thus provides us with a basis for explain-
schizophrenic behavior of the electrons ing why the nearly-free-electron model of
led Conyers Herring5 around 1940 to treat with a Hamiltonian, H, that is a sum of a Drude, Bloch and Sommerfeld works.
the wavefunctions of electrons that free-electron part, p'2/2m, and the core The remaining problem is to find the
propagate through the crystal as a linear potential, Vc. Using the values of 6C from pseudopotentials and elementary exci-
combination of core states and plane equation 2, we can write Schrodinger's tations that will serve as a basis for cal-
waves that he made orthogonal to the equation (after some algebraic manipu- culating the properties of a real sub-
filled core states. This "orthogonal- lation) as stance.
ized-plane-wave" approach produces a
wavefunction that has plane-wave char- Ec)<j>c (4>c\<t>) = E<f (3) Finding pseudopotentials
H<p + Z(E-
c
acter between the cores and resembles the Several methods have been used to
next excited atomic state near the cores. The sum over the core states represents construct pseudopotentials, including
These wavefunctions formed an excellent an operator acting on the smooth part of • Determination of potentials from
basis for calculations of response func- the wavefunction. We identify this op- models that mimic the form desired for an
tions and other properties. In the late erator with the repulsive potential analysis of equation 5. These models
1950's James C. Phillips and Leonard usually involve a few adjustable parame-
Kleinman 6 showed that one can derive VR4> = Y.(E-Ec)4>c(4>c\<t>) (4) ters, which are fitted to atomic data.
c
Herring's wavefunctions as solutions for • Forcing the pseudopotential to repro-
a problem in which a repulsive potential Schrodinger's equation thus becomes duce some measured properties of the
VR cancels much of the ion or core po- 1
(p2/2m + V)4> = EQ (5) solid. This potential can then be used to
tential V o producing a net weak pseudo- calculate other properties and analyze
potential. This pseudopotential provides where the potential V is the sum of the experimental data. This approach, which
the answer to the puzzle of how the sup- core potential Vc and the effective re- was developed by Phillips, David Brust,
posedly strong ion potentials can be re- pulsive potential VR. Franco Bassani and me is called the Em-
garded as weak. Note that the energy E in equation 5 is pirical Pseudopotential Method.
the exact energy of the state, so that we • Computing the pseudopotential di-
Phillips cancellation theorem can solve for eigenvalues using smooth rectly from information about atomic
The calculation of Phillips and Klein- wavefunctions via a plane-wave basis set structure, such as atomic spectra or
man provides a simple explanation of how and weak potentials. The repulsive po- atomic calculations. The part of the po-
a pseudopotential works, so we repeat it tential partially cancels the attractive core tential that arises from electron rear-
here. potential leaving a weak pseudopotential. rangement (as the atoms condense into a
Assume the real wavefunction \p has a This is the Phillips cancellation theorem. solid) can be calculated using the wave-
smooth part <p (the pseudo-wavefunction Morrel Cohen and Volker Heine later functions or charge density obtained from
that can be expanded in plane waves) showed that this conclusion is quite gen- solving Schrodinger's equation for the
plus a core part (which can be expanded eral and that the cancellation can be made condensed state. When the problem is
in core states, 4>c): almost complete in the core region. solved self-consistently, this is called
There are a few assumptions implicit in the Self-consistent Pseudopotential
bc4>c (1) this calculation. We use the standard Method.
model and assume the Born-Oppen- For many of the methods used, the
Following Herring's approach we make yp heimer approximation, which allows us to pseudopotential outside the core region
orthogonal to the core states and obtain picture the cores as stationary when dis- normally resembles the schematic curve
the coefficients bc from cussing the electron dynamics, and we shown in figure 1. Far from the nuclei, at
assume a Hartree or one-electron model large r, the potential behaves like an ionic
when we treat the potential Vc as an av- Coulomb potential. At smaller r the po-
The wavefunctions must satisfy the erage potential an electron feels anywhere tential deviates from the ionic one. The

42 PHYSICS TODAY / JULY


Downloaded 12 Nov1979
2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms
spectra for silicon. The energy positions
of the peaks are in excellent agreement;
peak heights are in fair agreement. The
peak heights are related to optical tran-
sition matrix elements and hence wave-
functions, which may not be as accurately
determined as the energy levels. We also
did not consider the effects of excitons, or
electron-hole bound states, which may
affect the accuracy of the calculations.
Although the spectrum does not appear
to be rich with structure, many energy
levels have to be consistent with experi-
ment for agreement on this level. The
success of these calculations is illustrated
more dramatically in modulated reflec-
tivity measurements, which, in essence,
determine the derivative of the spectrum.
Fine structure caused by optical transi-
tions is much more obvious here. Figure
ENERGY E(eV) 4 shows a comparison between experi-
mental results and a theoretical calcula-
Modulated reflectivity spectra reveal fine structure in optical spectra. The calculated pseudo-
potential results (color) are used to determine the origin of the observed structure. The experimental
tion for germanium. The calculation in-
results (black) are due to R. R. L. Zucca and Y. R. Shen. Figure 4 dicates that almost all the structure
shown can be interpreted in terms of op-
tical transitions.
change becomes large at about one-half a ciprocal lattice vectors are G = m2w/a, A wealth of uninterpreted data from
bond length (nearest-neighbor distance), where n and m are integers.) For the the 1960's and 70's has been analyzed with
where the repulsive force caused by the pseudopotential the Fourier expansion in such empirical pseudopotentials. These
orthogonality of the wave function to the terms of reciprocal lattice vectors is studies have explained the properties of
core-state wavefunctions becomes im- dozens of crystals and have resulted in
portant. (Typically, a bond length in a V(r) = Z V(G)S(G)eiG-r (6) general topological arguments for each
G
semiconductor is on the order of 2.5 structural class of solids; these in turn
A—for silicon it is 2.35 A). A good model The coefficients V(G) are called "form have given detailed predictions of optical
potential would consist of an ionic Cou- factors" of the potential; S(G) is the spectra. It was finally possible to cast
lomb potential cut off at around one-half "structure factor" that locates the atoms aside the old idea that, unlike atomic
of a bond length as shown by the hori- in the unit cell; it can be determined by spectra, the spectra of solids are too broad
zontal line in figure 1. A more accurate x-ray or neutron diffraction analysis. to be interpreted.
potential is given by the solid line. The Often V(q) is small at large q; so it is A host of other properties can be cal-
potential inside the core region is often only necessary to determine the form culated from the energy levels, wave
not well defined in many pseudopotential factors for a few reciprocal lattice vectors. functions and pseudopotentials deter-
schemes. Because we are normally only For example, for silicon the first G's have mined by these calculations, for ex-
interested in valence electrons outside the magnitude V3, V8, V11. With the three ample
core, this lack of information causes numbers V(V3), V(\'8) and V(v 11) one • the wave-vector and frequency-de-
few problems for valence electron energy can determine the electronic energy levels pendent dielectric functions
bands. One can also treat the core region or bands of silicon over a range of 1 Ryd- • lattice vibrations and electron energy
more exactly using "hard-core" pseudo- berg with an accuracy of about 1 or 2%. losses
potentials, as we discuss later. • temperature and pressure dependence,
It is convenient in dealing with periodic Electronic properties of energy levels
systems like crystals to work in "recipro- Let us now explore how to apply the • band gaps
cal space" whose points are wave vectors, Empirical Pseudopotential Method. We and many other spectral and transport
q, and to deal with the Fourier-analyzed first choose values for the three form properties.
potential V(q). At small q (or large r) the factors based on estimated or crude cal- One of the most interesting applica-
potential is attractive and looks like a culations. These are put into a pseudo- tions is to a study of the chemical bond in
screened Coulomb potential. For a metal potential Schrodinger equation, whose solids. By using the wavefunctions cal-
this implies that as q becomes small the solutions yield the energy eigenvalues and culated through the empirical pseudo-
potential should approach —2/3 of the corresponding wavefunctions. These potential we can obtain the electronic
Fermi energy. For larger q the potential eigenvalues and wavefunctions are then charge density as a function of position in
becomes positive because of the repulsive used to compute various response func- the crystal simply by squaring the wave-
nature of the core, and as q becomes very tions like the reflectivity or the density of function for each state and summing the
large the potential dies out. Figure 2 states. Comparison of the calculated and contributions of all the occupied states.
shows a typical potential, like those found the measured reflectivity, or the density Figure 5 shows the results for silicon. The
for many elements in the periodic table. of states and the energy distribution colored circles represent the cores, and the
Because of the periodic structure of curves from photoemission experiments contour lines indicate the charge density
crystals, Fourier transforms can be ex- reveals the accuracy of the form factors. for the valence electrons, which are de-
pressed as sums over reciprocal lattice We then adjust the V(G)'s until there is rived from the rearranged atomic 3s23p2
vectors, G, instead of integrals over all q. a good fit between the calculated and ex- electrons. The covalent nature of the
(The reciprocal lattice vectors determine perimental results. Once a fit is found Si-Si bond is clearly displayed by the
the Bragg scattering planes and are re- and the pseudopotential fixed, a host of "ball" or "ellipsoid" of charge located
lated to the symmetry of the crystal. In other properties can be calculated. between the cores. The experimental
a one-dimensional crystal with spacing a, Figure 3 contains a comparison of results given in figure 5 were derived from
the lattice vectors are R = na and the re- computed and measured reflectivity an analysis of x-ray data, in which the core

PHYSICS TODAY / JULY 1979


Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms 43
atoms and charges rearrange themselves
considerably from the bulk material.
Furthermore, a surface, like a defect
within the bulk of the crystal, breaks the
fundamental periodicity. Localized
configurations—like atoms, molecules,
defects, and surfaces—present real
problems for calculations that depend on
crystal periodicity. One solution, used by
Joel Appelbaum and Donald Hamann, is
to match bulk pseudopotential solutions
to surface-state-like solutions. Another
approach is to retain our precious peri-
odicity by introducing the concept of
"supercells."
A supercell is simply a large "unit" cell
of a crystal that can contain many lattice
sites of an ordinary crystal. The struc-
ture factor, S(G), of equation 6 accounts
for the positions of the atomic cores, and
we can use it to describe supercells with
any arrangement of atoms we like. If, for
example, we wish to study an atom or a
molecule we can use the structure factor
The covalent bond of silicon is clearly displayed in this electronic charge-density-contour plot for to create a large cell with the atom or
the valence electrons. The core electron contribution is represented by the large circles.
Agreement between theory9 (black) and experiment (color) is very good. The experimental deter-
molecule in the center of the cell and
mination (from x-ray analysis) is due to L. W. Yang and P. Coppens. Figure 5 empty sites or space in the rest of the cell.
This supercell can then be repeated infi-
contributions to the charge density were pseudopotentials compare very well with nitely and the problem to solve is of the
removed by subtracting calculated atomic experiment. same type as before, but with a large unit
charge densities. Although there are It is intriguing that pseudopotentials cell; the empty space ensures that there
some uncertainties with respect to detail derived from optical properties of semi- will be little interaction between the
in both the experimental and theoretical conductors can be used to determine su- atoms or molecules. If we wish to study
results, the agreement between theory perconductive properties of metals. For surfaces we can use a supercell that has
and experiment is striking. The elec- example, analysis of the optical spectrum the regular crystal periodicity in the x-
tronic charge density is concentrated in a of semiconducting indium antimonide and y-directions, but has slabs of finite
manner that was anticipated for the determines not only electronic structure thickness with space between them in the
crystal, but was never revealed before the of InSb, but also the potentials for indium 2-direction; the "crystal" consists of an
advent of pseudopotential calculations.7 and antimony individually. Some ex- infinite number of slabs of material
Other covalent crystals have been exam- trapolation is necessary, but it is possible stacked in the 2-direction, with space to
ined and systems with partially covalent to use the potentials derived from semi- isolate them. The new geometry retains
and partially ionic bonding have also been conductor data to analyze the electronic the periodicity we found useful earlier,
analyzed with this approach. By exam- structure and Fermi surface of indium but the much larger unit cell also intro-
ining the contributions of individual and antimony, and even to compute the duces new reciprocal lattice vectors, G,
electronic states one can further reveal the superconducting transition temperature and forces us to extrapolate the V(q)
s, p, d bonding or antibonding nature of a of indium.1 These results can be ob- curves.
state with a straightforward picture. tained knowing only the lattice constants The pseudopotentials needed to study
Calculations based on pseudopotentials and structure of In and Sb with no other surfaces have two parts: the ionic (or
and elementary excitations also permit us experimental measurements on these el- core) contribution, which is essentially the
to explore the interactions between the ements. same in the atom or the bulk as it is at the
electrons and the vibrating lattice. For In spite of this successful calculation, surface, and the electronic contribution,
example, the electron-lattice or elec- one must be cautious in transferring which depends on the electronic arJ
tron-phonon interaction is responsible for pseudopotentials from compound to rangement. One can estimate the elec-
binding pairs of electrons into Cooper compound, because the electronic envi- tronic part of the potential fairly accu-
pairs. Using these interactions as a ronment may be very different in differ- rately from the charge density, using the
mechanism, John Bardeen, Leon N. ent materials. For example, layer com- density-functional approach which allows
Cooper and J. Robert Schrieffer devel- pounds, metals, diamond-like semicon- the determination of the Hartree and
oped one of the most successful theories ductors, ionic insulators, and so forth, all exchange-correlation potentials directly
of physics, usually called the "BCS theo- have their own peculiar electronic fea- from the electronic charge density. To
ry," which answers most of our questions tures because valence electrons are ar- begin the calculation, we use an empirical
about superconductivity. To calculate ranged differently and are thus sensitive pseudopotential to compute the charge
the superconducting transition temper- to the pseudopotential in different regions density, fr6m which, in turn, we calculate
ature, Tc, it is necessary to know the of r (or q) space. There are a number of the electronic contribution to the poten
strength of the electron-lattice interac- fortuitous cases like In, Sb and InSb, tial. We then add the ionic potential and
tion, which in turn depends on the pseu- where one can transfer the potentials di- structure and use the new potential to
dopotential. With current versions of the rectly, but there are many other cases, begin the calculation over again. Onc<
BCS theory and modern computers, the such as crystal surfaces, where charge re- the input and output potentials are th<
major stumbling block to accurate com- arrangement is critical. same, that is, once the calculation is self
putations of Tc is a knowledge of the consistent, the problem is considere<
normal state properties of the supercon- Surfaces and supercells solved. For a surface, for example, durinj
ductor. For simple metals, however, The surface of a crystal is not just the the first few iterations, electronic charg-
transition temperatures derived from end of a perfect crystal. Rather, the begins to escape from the bulk into th

44 PHYSICS TODAY / JULY 1979


Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms
THE NEW SPECTROSCOPY

DERIVATIVE TECHNIQUES
IN THE INFRARED A. Transmission Spectrum - r2 band of NH 3
Pressure: 5.4 torr.
with a Laser Analytics' Model LS-3 Path length: 16.4 cm
Laser Source Spectrometer
Using,the unique electronic tunability of Tunable Diode
Lasers, the LS-3 Laser Source Spectrometer easily and rapidly
1 2
generates first and second derivative spectra as well as
conventional absorption spectra. Derivative spectra facilitate
experimental studies such as line position calibration and
extremely low level gas detection (e.g., <0.1 ppb of NH3 in
air). The line position calibration data at the right were B. First Derivative
recorded in less than an hour, illustrating the convenience
and versatility of the LS-3 as well as its exceptional resolution
capability(~10^ cm-1).
A. Transmission of NH3 near 1248 cm-1 showing 8 weak
absorption lines (pressure x path length product = 88.56
cm Torr).The sloping background is due to increase in
source intensity.
6. First derivative spectrum of A illustrating accentuation of
weaker NH3 lines.
C. Second derivative of A showing further accentuation of
weak NH3 lines. Note that the second derivative signal has C. Second Derivative
been used for exceptionally low level detection of NH3
(<0.1 ppb).
D. Frequency calibration with 0.0297 cm-1 fringe spacing.
E. Calibration results using adjacent 10°0-00°0 N2O lines as ,
reference frequencies. i

Explore the many unique capabilities and applications of the


LS-3 by calling Bill Diamond at Laser Analytics. He has
available a list of 70 recent applications papers in areas
ranging from molecular beam studies to heterodyne detection.
D. Etalon Trace

LASER ANALYTICS, INC.


25 Wiggins Avenue
Bedford, MA 01730 (617) 275-2650

E. Line Positions and Tentative Assignments

Line ID# Line Pos. (cm-1) J AJ K


1 1247.317 14 1 6
2 1247.325 14 ' 1 9
3 1247.393 14 1 5
4 1247.412
5 1247.441
6 1247.493 14 1 4
7 1247.604 14 1 3
8 1247.642 14 1 10
A SUBSIDIARY OF SPECTRA-PHYSICS INC.
Circle No. 19 on Reader Service Ca
Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms
vacuum, but then the resulting change in they also reveal that the dangling bond bond" states. Furthermore, some sur-
the potential acts to pull it back; at self- does not "dangle;" the surface "heals," in faces exhibit more complex geometries.
consistency a compromise is achieved. that the charge readjusts and smooths For example, we used an idealized model
Several localized configurations have been out, explicitly showing that the surface is for the Si (111) surface in that we assumed
analyzed using the above method; they not just the end of a bulk crystal. The the geometry of the bulk extended to the
include diatomic molecules, surfaces, figure also demonstrates that the per- surface. However, it is believed that the
steps on surfaces, adsorbates and vacan- turbation of the surface affects the total atoms on the surface "reconstruct," that
cies. charge only a few layers into the crystal. is, form an array different from the ideal
The (111) surface of silicon has been Near the surface there are evident chan- shown in figure 6. Low-energy electron
one of the most studied8 surfaces during nels in the total charge density, and it is diffraction (LEED) studies on Si(lll)
the recent renaissance in surface physics, likely that impurities can migrate into the indicate that one likely reconstruction is
and it provides a good example of what crystal through these channels. a "buckled" surface. The topmost atoms
the self-consistent method can achieve. As for surface states, these would not in figure 6 rearrange so that one atom
Figure 6 shows a ball-and-stick com- show up in the total charge density unless pushes out along the (111) direction while
puter-drawn model of a silicon crystal they had amplitudes comparable with the next one pulls in. This creates a so-
near its (111) surface. Because of the bulk states. It is possible to search for called (2 X 1) surface. Applying the
surface, the topmost silicon atoms have these states by comparing the electronic self-consistent procedure to this case re-
only three nearest neighbors whereas bulk states (using a density-of-states function) veals that the dangling-bond surface state
silicon atoms have four. How do these at the surface layer with those of the bulk. associated with the "in" atom loses its
"cut" or "dangling" bonds rearrange once The analysis is straightforward and shows electron to the "out" atom. It is even
they are cut—do they dangle like cut that there is indeed a prominent surface possible to calculate the optical spectrum
rop^es, pull back, or smooth out? Another state, whose energy lies in the forbidden for electron transitions from states asso-
important question concerns the exis- semiconductor gap region. This state has ciated with "out" atoms to states of "in"
tence, suggested by Igor Tamm in the been called the "dangling bond" surface atoms.
early 1930's, of surface states, which are state, because it is believed to be associ- Other semiconductor, insulator, sim-
electronic states localized near the surface ated with the cut (111) bond. The ple-metal and even transition-metal sur-
that decay both into the vacuum and into charge-density contour plot for this state faces have been explored using the self-
the bulk solid. A further point of interest appears in figure 7, which shows that it is consistent method. Even metal-semi-
is how far the perturbations caused by the a bona-fide surface state, spreading into conductor interfaces (Schottky barriers)
surface extend back into the bulk region both the vacuum and the bulk, while and semiconductor-semiconductor in-
of the solid. being localized close to the outermost terfaces (heterojunctions) have been an-
A charge-density plot for the surface atom on the Si(lll) surface. Experi- alyzed in detail. Steps on surfaces, bulk
can answer some of the questions. One mentalists have explored the properties vacancies and adsorbates, such as hy-
expects that such a plot would be similar of this state with photoemission and drogen or chlorine on silicon or hydrogen
to figure 5 for the bulk, but with differ- electron energy-loss techniques and on molybdenum or palladium, have also
ences arising from the charge redistribu- theorists have analyzed it with self-con- been studied. The applications in this
tion at the surface. The cover of this sistent potentials using both supercells area have only just begun.
issue of PHYSICS TODAY shows the total and the matching approach of Appel-
electronic charge density near the Si (111) baum and Hamann. Other new developments
surface. The results resemble the ball- Other surface states are localized in the While many recent applications of
and-stick model of figure 6. However, second or third layer—the so-called "back- pseudopotentials have been in surface and
interface research, progress in bulk cal-
culations and recent work on the con-
struction of basic theory and further ap-
plication of pseudopotentials have been
considerable. A major effort has gone
into calculations for complex systems.
Layer and chain compounds, complex
polytypes of simple materials like silicon
or germanium that simulate amorphous
systems, materials with d-electrons, and
complicated crystal structures are sam-
ples of the effort in this area. A good ex-
ample of the last item is the high-tem-
perature superconductor NbaGe. Pseu-
dopotentials have permitted not only the
computation of the complicated energy
bands and densities of states, but also the
analysis of the charge density. This anal-
ysis illustrates how charge is distrib-
uted among niobium, germanium and
the region between these atoms. The
dominant effect is that charge leaves both
Nb and Ge atoms to fill the space between
them, yielding a "metallic" configuration
for the charge density. One hopes that
calculations of this kind will supply useful
information for understanding why the
superconducting transition is at a high
A ball-and-stick model of the silicon (111) surface reveals the "dangling bonds" on the outermost
temperature in this material.
atoms. The colored balls represent the cores that are used to display the charge-density config- In all of the calculations described here
urations on the cover and in figure 7. (From reference 8.) Figure 6 it was first necessary to know the struc-

46 PHYSICS TODAY/JULY 1979


Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms
and Phillips have expanded on the Si-
mons-Bloch approach and developed
more flexible analytic forms for the
A closer look..
pseudopotential. Alex Zunger and I have
explored another avenue which we call the
"first-principles pseudopotential ap-
proach" (almost contradictory terms). In
this scheme, the pseudopotential is ob-
tained directly from atomic wave func-
tions without empirical fitting. These
pseudopotentials have the proper be-
havior near the core as well as giving ac-
curate results in the atomic and molecular
limits. First-principles pseudopotentials
also give the Simons-Bloch results for
quantities such as the core radii. While
the hard-core pseudopotentials are useful
for a large class of problems they also have
a disadvantage in that the potential is
...at
The dangling-bond surface state associated
with the Si (111) surface. The charge density
large within the core, so that the compu-
tations require a more sophisticated basis
EG&G
for this localized state decreases rapidly into the
vacuum and bulk solid. This state has been
observed experimentally. Figure 7
set than plane waves for the wavefunc-
tions and hence large computer facili-
Lock-Ins
ties. tells the
ture of the system considered, that is, the
In many calculations, the old "soft-
core" pseudopotentials are being replaced real story...
bulk crystal structure or the surface by a variety of "hard-core" pseudopo-
atomic arrangement. Often, as in the tentials. Although the hard-core poten-
case of surfaces, this arrangement is dif- tials can be more difficult to manage, they
ficult to determine. If the pseudopo- are often more accurately determined,
tential could be used directly to calculate especially in the core region. So the
total energies, minimizing the energy with pseudopotential domain is now being
respect to atomic configuration would expanded to include phenomena that Take dynamic reserve for instance...
yield a direct determination of the crystal depend on electronic structure near the the instrument's ability to handle
and surface structure. This has been a core region—a region which at the outset noise...the Model 186A allows a
difficult area; success has only recently was forbidden. But shouldn't we expect signal to be measured in the presence
been achieved for some bulk solids, and this from a true panacea? of up to 300.000 times full scale
only partial success is achieved for the noise. NO competitive instrument
silicon surface . The reason for the diffi- comes close to this performance. And
frequency range...0.5 Hz to 100
culty is that the energy differences to be This article is based in part on a talk given at kHz...is packed into every
calculated are very small, and extremely the meeting of The American Physical Society instrument...not into boxes of color-
accurate pseudopotentials are needed. on 20 March 1979 in acceptance of the 1979 coded boards to be changed! And
As we have seen, one does not need Oliver E. Buckley Prize in Solid State Physics. you don't have to give up 100
detailed knowledge of the pseudopoten- The author wishes to acknowledge support nanovolt sensitivity...or time constant
from the National Science Foundation; the or zero offset when you change
tial in the core region to expose most of Division of Materials Sciences, Office of Basil-
the physics involved in problems of elec- dynamic reserve...once you set
Energy Sciences, US Department of Energy, it...you get it!!!! No wonder EG&G
tronic structure calculations. However, and a Guggenheim Foundation Fellowship. Lock-In Amplifiers are the instruments
for crystal-structure analysis, for deter- He also wishes to thank the department of more researchers select for critical ex-
mination of the conditions for alloy for- physics at the University of Hawaii, Honolulu, periments. Shouldn't you take a
mation, for detailed molecular studies in for their hospitality while much of this article closer look?
the core region and for other problems, was written.
one must treat the region close to the Other points you'll see on closer
atomic core with care. References inspection of EG&G Lock-Ins...
To examine the core region, Gary Si- 1. A review of pseudopotential research is • True differential, floating or
mons and Aaron N. Bloch have intro- given in: M. L. Cohen, V. Heine, Solid grounded inputs
duced an analytic form of the pseudopo- State Physics 24, 37 (1970), as well as in • Full programmability
tential that contains a repulsive 1/r2 term. references 2, 3 and 4. • True 10-turn phase & zero offset
This term would cause the pseudopoten- 2. W. A. Harrison, Pseudopotentials in the controls
Theory of Metals, Benjamin, New York • Full frequency range narrow band
tial curve in figure 1 to pass through zero tracking
and become positive for small r. Because (1966)".
• Card-free, alignment-free fre- 1
of the strong repulsive force near r = 0 3. C. Kittel, Introduction to Solid State quency changing
these potentials are often called "hard- Physics, 5th ed., Wiley, New York (1976).
core" potentials. The zero of the poten- 4. J. C. Phillips, Bonds and Bands in Semi- For more information on EG&G
tial can be identified with the core radius. conductors, Academic, New York (1973). Lock-Ins, write or call today.
Phillips and James R. Chelikowsky used 5. C. Herring, Phys. Rev. 57, 1169 (1940). EG&G PRINCETON APPLIED
this approach to refine the semi-phe- 6. J. C.Phillips, L.Kleinman, Phys. Rev. 116, RESEARCH, P. O. Box 2565,
287 (1959). Princeton, NJ 08540;
nomenological theory of A. R. Miedema Phone: 609/452-2111.
on alloy formation. This research has 7. M. L. Cohen, Science 179, 1189 (1973).
opened up new vistas both for pseudo- 8. M. Schluter, J. R. Chelikowsky, S. G. Louie,
potential development and for the theory M. L. Cohen, Phys. Rev. B12, 4200 (1975). PRINCETON APPLIED RESEARCH
of metallic alloy formation. Wanda An- 9. J. R. Chelikowsky, M. L. Cohen, Phys. Rev.
dreoni, Alfonso Baldereschi, F. Meloni B14,556(1976). n
Circle No. 22 on Reader Service Card
PHYSICS TODAY / JULY 1979
Downloaded 12 Nov 2012 to 152.14.136.96. Redistribution subject to AIP license or copyright; see http://www.physicstoday.org/about_us/terms
47

You might also like