Documente Academic
Documente Profesional
Documente Cultură
l
M -o
e -oa/
M -z
e -.oa/
|
~
u
z
O
'
'
|
.
m
u
O
J
~
J
z
!P!L
'
~~~~ 0
O
u
2
2
O
o
'
z
>
O
C
MAGNETIZATION of a ferromagnet has a sudden onset at the Curie temperature. Above
this temperature the average numbers of up spins and down spins are equal and so the magneti
zation is zero. At any temperature below the Curie point two states of magnetization are possi
ble, depending on whether the up spins or the down spins are in the majority; in the absence of
an external magnetic feld the two states are equally likely. The susceptibility of a ferromagnet
measures the change in magnetization induced by an arbitrarily small applied magnetic feld.
At the Curie point the susceptibility becomes infnite. Near the Curie point a small change in
either the temperature or tbe external feld gives rise to a large change in the magnetization.
cause they have an overall excess of one
spin direction. Thus an ocean of spins
that are mostly up may have within it an
island of spins that are mostly down,
which in turn s urrounds a lake of up
spins with an islet of down spins. The
progression continues to the smallest
possible scale: a single spin.
When the temperature is precisely
equal to the Curie temperature, the cor
relation length becomes infnite. Any
two spins are correlated, no matter what
the distance between them is. Neverthe
less, fuctuations persist at all smaller
scales of length. The system remains un
magnetized, but it is exquisitely sensi
tive to small perturbations. For exam
ple, holding a single spin fxed in the up
orientation creates a di sturbance that
spreads throughout the lattice and gi ves
the entire system a net magnetization.
Below the Curie temperature the sys
tem becomes magnetized even in the ab
sence of an outside perturbation, but
there is no immediate change in the ap
pearance of the lattice. Smaller-scale
fuctuations persist; they are remnants
of the lakes and islets of opposite spin
direction. Merely by looking at the lat
tice one cannot detect the magnetiza
tion. Only when the system is cooled
further does the bias become obvious, as
the increasing coupling strength coerces
more of the spins into conformity with
tpe majori ty. At zero temperature com
plete uniformity is attained.
In fui ds t he fuctuations in density
near the critical point are closely analo
gous to the fuctuations in spin direction
observed in ferromagnets. In fui ds,
however, the presence of fuctuations at
all possible scales of l ength can be ob
served directly. When the correlation
length frst reaches a few thousand ang
strom units, which is comparable to the
wavelength of light, the fuctuations be
gin to scatter l ight strongly and the fuid
turns milky, a phenomenon called crit
ical opalescence. Signifcantly, when the
temperature comes still closer to the
critical point and the maximum scale of
the fuctuations becomes much larger
(millimeters or centimeters), the critical
opalescence is not reduced, indicating
that the smaller fuctuations persist. The
same phenomenon takes place in spin
systems, but because ferromagnetic ma
terials are not transparent to l ight it can
not be demonstrated as readi ly. The crit
ical opalescence of ferromagnets has
been detected, however, in the scattering
of neutrons from a magnetic material
near the Curie temperature.
T
he model I have been describing is
not my own invention. It is a version
of one introduced in the 1920' s by the
German physicists Wilhelm Lenz and
Ernest Ising, and it i s now called the
Ising model. The properties of a system
of Ising spins on a two-dimensional lat
tice are known in complete detail be
cause the model was solved exactly in
1944 by Lars Onsager of Yale Universi-
D
1979 SCIENTIFIC AMERICAN, INC
/SyCUrir!rCCUC!iCr!C
The Library of Science
hOO888lh8|
this SG76DClassic
f
or only S3.9
A saving o
f
94%
Edited by Douglas N. Ccrsicire. MeorIycLL
experts have contributed to thisthoroughly revisec
and greatly expanded fifth edition of the most
authoritative single-volume source of scientific
knowledge ever assembled. Enormous Y4x 1<
volume contains r.rmillion words, cc<pages,
<Lphotographs, drawings and charts, and
LBOGS. rBIL|C|GSCOVGImathematics, from
|OOIOBL|OOSC|GOCGSLOphysics and chemistry .
``...BOBOBZO@OOOH...OIboth the general and
scientific reader. " -The New York Times
any other 3 books for only $3.95
|yOu wiII [cnnOwOI8I8! 6IO0 8H08gI6608K6 d 00I60OOK5-8h8H05OU60i5LOun5OVe| IheneII2ccnIhs.
(values 0$75.00)
(Pub|is|ers'Prlcesshown)
b/
54995. THE ILLUS
T RATED ENCY
CLOPEDIA OF .R
CH AEOLOG Y. Clyn
Daniel. $17.95
44905. THE ENCY
CLOPEDIA OF HOW
IT'S MADE. $14.95
55000. T HE ILLUS
TRATED ENCY
CL O P EDIA O F
ASTRONOMY A ND
SPACE. $16.95
74300-2. ROM AND
HER EMPIRE. Bar
Cunliffe. A magnifi
cently illustrated volume
t h a t m ake s t h e
history of one of the
world's greatest civiliza
t i ons c ome a l i ve .
Counts as 2 of your J
books. $50.0
|O600
ExtmItt8bm| t.ta
la6" M
F
w
74771. MES SA GES
FROM THE S TA RS!
THE RUNAWA Y
UIVERSE. A report
on the effors of scien
tists to locate traces of
interstellar communica
tion. Plua, the gripping
and dramatic story of the
birth and death of the
cosmos from the Big
Bang to titanic future
holocaust. The ? coulll
as one book. $19.95
63340-2. MYSTERIES
OF THE PAST. Cap
tivating investigation of
prehistory and great rd
dles. Outsized. Sumptu
ously illustrated. Cottits
as ? of your J books.
$34.95
1979 SCIENTIFIC AMERICAN, INC
3 47 40. ATLAS O F
MAN . Covers 400
groups of peopl es,
tribes, and nations. 120
color photographs, 350
maps and line illus
trations. Covers geog
raphy, land use, cultur,
climate, religion, much
more. $2 5.0
4 65 802 . T HE
EVOLVING CONTI
NENTS. Br i a n F.
Windley. An integrated
global overview of the
geosciences. Counts as
Z of your J books .
$34. 95
8057 0. ST E L LAR
ATMOSPHERES.
Dimitri Mihalas. A au
thoritative rpr on the
stella envelop, moes
of energy transpr in
this atmosphere, rtes of
sta mass loss, and much
$2.95
4883 0: FRACTALS,
FORM, CHANCE
AND DIMENSION.
Benoit B. Mandelbrot.
A brilliant geometrical
investigation into the
fagmented and i gula
pat t e rn s of na
W. $17 .50
422 85. DIRECTIONS
IN PHYSICS. Paul ,
A.M. Dirac. One of the '
greatest t heoretical !|
physicists of the twen-
tieth century presents his _
views on cosmological
t h eori e s , q uant um
t h e or y , q u a n t u m
mechanics and quantum
electrodynamics.
$1 2.95
662 75-2 . THE ORI
GIN OF THE SOLAR
SYSTEM. Edited by S.
F.Dermott. A compila"
tion of the views of 28
leading astronomers, as
trbphys i ci st s, cos
mologists and mathema
ticias covering theores
of stellar formation.
Counts as Z of your J
books. $39. 0
THEANASlS
OFlNFONMATlON
S
ySTEM
S
""
t
:+ + t
+--*'"``
81403. SU, MOON,
AND STANDING
STONES. John Edwi n
Wood. I ncludes the latest
findings from resear
chers m achaeology, as
tronomy, climatology,
and ph ysi cs. il l us
tated. $14. 95
397462 . COMPAN
ION TO CONCRETE
MATHEMATICS.
Volume I and U.Z.A.
Me/zak. Mathematical
techniques and various
applications. The set
counts as Z of your J
books. $4. 45
41 610 . DAT A
ANALYSIS FOR
SCI ENTISTS AND
ENGINEERS. Stuart
L. Meyer. 513-page vol
ume covers all the statis
tical methods, concepts,
and analysis techniques
that any experimenter
will probably ever need.
$19. 95
40167 2. THE CON
DENSED CHEMI
CAL DICTIONARY.
Updated to meet toay's
needs. Over 18,000
entries. Counts as Z of
your Jbooks . $32.50
3330. THE ANALY
SIS OF INFORMA
TION SYSTEMS. Re
vised Edition. Charles
!. Meadow. A distin
guished computer au
thority offers an all
inclusive treatment.
$20.95
3 7 34 7 . CASTE AND
ECOL OGY IN THE
SOCIAL INSECTS.
Oster and Wilson.
$20. 0
34210. T HE AR
CHAEOLOGY OF
NOKH AME RICA.
Dean Snow. Prehistoric
Indian cultures, from
the first crossings into
t his continent to t he
20t h ce nt u r y. 195
photos , char t s and
maps. $18.95
7 3940. THE REST
LESS UNI VERSE.
Henr L. Shipman. A
highly acclaimed 450-
page tour of the ideas,
principles, and press
of modern astronomy.
$16.95
3 83622. CLIMATIC
CHANGE. Dr. John
Gribbin. Povides an A
to Z repor on the basic
pieces and patters of
t h e e a r t h's r i ch
climatological puzzle.
Counts as ? of your J
books. $32.50
74590. THE ROOTS
O F CIVILIZATION.
Alexander Marshack. A
remarkable re-creation
of the life of prehistoric
man. Illustrated. $1 7.50
87610. TH WORLD
ENERGY BOOK.
Crabbe and McBride.
Over l500 alphabeti
cally aranged entries on
everything from fossil
fuels to power from
sewage. Incl udes J5
wor l d maps and 40
pages of t ables, dia
grams, and chas.
If the reply card has been removed,
please write to The Library of Science
Dept. 2 -B6A, Riverside, N. J. 08370
$25.0
to obtain membership information and application.
UJ
1979 SCIENTIFIC AMERICAN, INC
ty. Since then sol utions have also been
found for several other two-dimension
al models (whereas no three- dimension
al model has yet been solved exactl y) .
Nevertheless, the problems of describ
ing two-dimensional systems are far
from trivial. In what follows I shall
apply the methods of the renormaliza
tion group to the two-dimensional I sing
model as if it were a problem still out
standing, and Onsager's sol ution will
serve as a check on the results.
What does it mean to solve or to un
derstand a model of a physical system?
In the case of the Ising system the micro
scopic properties are known completely
from the outset, since they were speci
fed in bui l ding the model. What is need
ed is a means of predicting the macro
scopic properties of the system from the
known microscopic ones. For example,
a formul a gi ving the spontaneous mag
netization, the suscepti bility and the
correlation length of the model as a
function of temperature would contri b
ute greatly to understanding.
It is not notably di fcult to calc ulate
the macroscopic properties of any given
confguration of the spins in an Ising
model. The magnetization, for exampl e,
can be determined si mpl y by counting
the number of up spins and the number
of down spins and then subtracting. No
one confguration of the spins, however,
! ` !0
2 !0
2 ` !0
0
z
O
6 ` !0
c
L
T
* !0
z
O
O
z
o
2 !0
0
U
O
c
65,526 u
z
5!2
!6
2
l
determines the macroscopic properties
of the system. Instead all possible con
fgurations contribute to the observed
properties, each in proportion to its prob
ability at a gi ven temperat ure.
In principle the macroscopic proper
ties coul d be cal culated directly as the
sum of all the separate contri butions.
First the magnetization woul d be found
for each confguration and then the cor
responding probabi lity. The act ual mag
netization would be obtained by mul ti
pl ying each of these pairs of numbers
and adding up all the results. The sus
ceptibility and the correl ation length
could be found by procedures that are
not much more elaborate. The common
element in all these calculations is the
need to determine the probabil ities of
all possible confgurations of the spins.
Once the distri bution of probabi lities is
known the macroscopic properties fol
low directly,
As I pointed out above, the probabi li
ty of any two adj acent spins' being par
allel is determined solely by the cou
pling strength K, whi ch I have defned as
the reciprocal of the temperature. If the
probability of two neighboring spins in
isolation being parallel is denoted p,
then the probabi lity of their being anti
parallel must be 1 p. From these two
val ues alone the relative probabi lity of
any specifed confguration of a lattice
! ! 2 2 2 2 4 4 5 5 6 6 !0 !0
LATICE SIZE
NUMBER OF SPIN CONFIGURATIONS rises steeply as the size of a lattice grows. For a
system of II spins, each of which has two possible values, the number of confgurations is equal
to 2. When the lattice is large, it becomes impractical to calculate the probability of all the con
fgurations. The limit of practical computation is a lattice somewhat larger than the six-by-six
array of J6spins. In order to observe the critical behavior of the system near the Curie temper
ature an array of about lOO-by-lOO spins would be needed, which has 2 confgurations.
D4
can be eval uated. All that is required i s
to multiply together the separate proba
bi lities for every nearest-neighbor pair
of spins, in each case taking the val ue as
] when the spins are parallel and as
! when they are anti parallel .
\
onsider a spin system that is made up
of j ust four spins arranged at the
corners of a square. Such a lattice has
four nearest-neighbor couplings, corre
sponding to the four sides of the square.
Each coupling is considered in t urn and
is assigned a probability of either p or
I ]according to whether the spins are
parallel or antiparallel; then the four
separate probabilities are multi pli ed. In
the confguration with all four spins ori
ented up all four pairs are parallel, and
so the relative probability is given by the
prod uct ] X p X p X ]. If three spins are
up and one is down, the relative proba
bi lity is ] X ]X (I ]) X (1 p).
The cal cul ation must be carried out
for every confguration of the spins; for
a system of four spins there are 1 6 con
fgurations. A fnal step is to convert
the relative probabilities into absol ute
ones by adj usting each value so that the
total of all 1 6 val ues is equal to exact
ly 1 . Since the temperature determines
the coupl ing strength and the coupl ing
strength in turn determines the values of
p and 1 p, the entire sequence of 1 6
cal culations would also have t o b e re
peated for every temperature of interest.
This plan of attack on the Ising model
is ambitious but impractical. If the prob
abi l ity of every spin confguration could
be calculated, the magnetization and the
other macroscopic properties coul d be
eval uated for any specifed temperature.
The problem lies in the number of spin
confgurations. For a system made up of
H spins, each of which can take on two
values, there are 2n possible confgura
tions. This exponential function grows
rapidly as H increases. As I have men
tioned, four spins have Z,or 1 6, confg
urations. A three-by-three bl ock of nine
spins has 512 confgurations and a four
by-four block has 65 , 5 36. The practical
limit of computation is not much larger
than a six-by-six bl ock of 36 spins, for
which there are approximately 7 X 1 0
!U
con fg ura tions.
What size lattice would be needed in
order to determine the critical proper
ties of the two-dimensional Ising model?
The array must be at least as large as the
largest fuct uations observed at the tem
perat ure of interest. At a temperat ure
reasonably close to the Curie point the
correlation length, in units of the lattice
spacing, might be about 1 00 and the
largest fuctuations would cover about
1002, or 10, 000, lattice sites. A bl ock of
spins that large has 2 1
0,000
possible con
fgurations, a number that is somewhat
greater than 1 03,
00
. The fastest comput
er conceivable coul d not carry out s uch
a cal culation. Even if the computer had
been working conti nuously since the
1979 SCIENTIFIC AMERICAN, INC
"bi g bang" with which the universe be
gan, it would not yet have made a signif
cant start on the task.
The need to carry out an almost end
less enumeration of spin confgurations
can be circumvented for two special
conditions of the lattice. When the tem
perature of the system is zero (so that
the coupl ing strength is infnite), all but
two of the confgurations can be neglect
ed. At zero temperature the probability
that a pair of spins will be anti parallel
falls to zero, and therefore so does the
probability of any confguration that in
cl udes even one antiparallel pair. The
only confgurations that do not have
at least one anti parallel pair are those
in which all the spins are up or all are
down. The lattice is certain to assume
one of these confgurations, and all oth
er confgurations have zero probability.
At infnite temperature, where the
coupling strength is zero, the probability
distribution is also much simplifed. Ev
ery spin is then independent of its neigh
bors and its direction at any instant can
be chosen at random. The result is that
every confguration of the lattice has
equal probability.
Through these two shortcuts to the
determination of the probability distri
bution it is a tri vi al exercise to calculate
exactly the properties of the Ising model
at absol ute zero and at infnite tempera
ture. Acceptable methods of approxi
mation are also available for any tem
perature low enough to be considered
close to zero or high enough to be con
sidered close to infnity. The trouble
some region is between these extremes;
it corresponds to the region of the crit
ical point. Until recently there was no
practical and direct method of calculat
ing the properties of a system arbitrarily
close to the critical point. The renormal
ization group provides such a method.
T
he essence of the renormalization
group method is to break a large
problem down into a sequence of small
er and more manageable stages. Instead
RENORMALIZATION-GROUP approach
to a model ferromagnet consists in breaking
down an intractable problem with multiple
scales of l ength into a sequence of smaller
problems, each of wbicb is confned to a single
scale of length. One version of the renormali
zation-group method, called the block-spin
transformation, has three steps. First the lat
tice is divided into blocks of a few spins each,
in this case nine. Then each block is replaced
by a single spin whose value is the average of
all the spins in the block; here the average is
determined by majority rule. In this way a
new lattice is created, with three times the
original lattice spacing and one-third the den
sity of spins. Finally the original scale is re
stored by reducing all dimensions by a factor
of J. The procedure must be carried out for
all confgurations of a few spins in the origi
nal lattice, so that a probability can be found
for every confguration of the block spins.
|
l
|
l
|
|
l
X
/
X
\
l
|
X
|
/
|
FORMATION OF BLOCKS
|
REPLACEMENT OF I NDI VI DUAL SPI NS BY BLOCK SPI NS
|
|
RESCALI NG
OF LATTICE
|
,
l
|
|
l
I65
1979 SCIENTIFIC AMERICAN, INC
PROBABILITIES OF NEAREST- NEIGHBOR CONFIGURATIONS IN ORIGINAL LATTICE
l
e * .3655 e .1345 e .1345 e * .3655
PROBABILITIES OF SIX-SPIN CONFIGURATIONS IN ORIGINAL LATTICE
'
e .2943
e .0147
e .0020
e .0007
e .0398
e .0054
e .0002
e .0020
e * .0147
e .0020
e .0007
e .0054
e * .0054
e .0147
e .0020
e .0398
e .0054
e * .0020
e * .0054
e * .0020
I
e .0007
e * .0007
e * .0007
e * .0007
P * .0020
P * .0020
P * .0020
P * .0020
P * .0007 P * .0147
P * .0007
I
P * .0147
P * .0147
e * .0020
P * .0020
P * .0020
P * .0020
e * .0007
e * .0007
P * .0007 e .0007
e .0020
e * .0054
e .0020
e .0054
e * .0398 e .0020
e .0147 e * .0054
e .0054
e .0007
e .0020
=
e .0147
e * .0020
_
e .0002
e .0054 e * .0398
e * .0007
_
e .0020
e .0147
e * .2943
t
e .4302 e .0697 e .0697 e .4302
PROBABILITIES OF NEAREST- NEIGHBOR BLOCK -SPIN CONFIGURATIONS
00
1979 SCIENTIFIC AMERICAN, INC
of keeping track of all the spins in a re
gion the size of the correlation length,
the long-range properties are deduced
from the behavior of a few quantities
that incorporate the efects of many
spins. There are several ways to do this.
I shall describe one, the block-spin tech
nique, in which the principles of the
method are revealed with particular
clarity. It was introduced by Leo P.
Kadanof of the University of Chicago
and was made a practical tool for cal
culations by Th. Niemeijer and J. M. J.
van Leeuwen of the Delft University of
Technology in the Netherlands.
The method has three basic steps,
each of which must be repeated many
times. First the lattice is divided into
blocks of a few spins each; I shall em
ploy square blocks with three spins on a
si de, so that each block incl udes nine
spins. Next all the spins in the block are
averaged in some way and the entire
block is replaced by a single new spin
wi th the value of the average. Here the
averaging can be done by a simple pro
cedure: by following the principle of
majority rule. If fve or more of the orig
inal spins are up, the new spin is also up;
otherwise it is down.
The result of these two operations is
to create a new lattice whose fundamen
tal spacing is three times as large as that
of the old lattice. In the third step the
original scale is restored by reducing all
dimensions by a factor of 3.
These three steps defne a renormali
zation-group transformation. Its efect
is to eliminate from the system all fuc
tuations in spin direction whose scale is
smaller than the block size. In the model
given here any fuctuation of the spins
over a range of fewer than three lattice
units wi ll be smeared out by the averag
ing of the spins in each block. It is as if
one looked at the lattice through an out
of-focus lens, so that the smaller fea
t ures are bl urred but the larger ones are
unafected.
It is not enough to carry out thi s pro
cedure for any one confguration of the
original lattice; once again what is
sought is a probability distribution. Sup
pose one consi ders only a small region
of the initial lattice, consisting of 36
spins that can be arranged in four
blocks. The spins in thi s regi on have Z`',
or about 70 bi llion, possible confgura
tions. After the block-spin transforma
tion has been applied the 36 original
spins are replaced by four block spins
with a total of 16 confgurations. I t i s
j ust within the limit of practicality to
compute the probability of each of the
confgurations of the original 36 spins.
From those numbers the probabilities of
the 16 block-spi n confgurations can
readily be determined. The calculation
can be done by sorting all the confgu
rations of t he original lattice into 1 6
classes according to which confgura
tion of the block spins results in each
case from applying the principle of ma
j ority rul e. The total probability for any
one confguration of the block spins i s
then found by adding up the probabili
ties of all the confgurations of the origi
nal lattice that fall into that class.
It may well seem that nothing i s
gai ned by thi s procedure. If the com
pl ete probability distri bution can be cal
culated for a system of 36 spins, nothing
new is learned by condensing that sys
tem into a smaller lattice of four block
spins. Near the critical point it is still
necessary to consider a much larger lat
tice, with perhaps 1 0, 000 spins instead
of 36, and the probability distri bution
for the block spins generated from this
lattice cannot be calculated because
there are far too many confgurations.
As it turns out, however, there is a meth
od for extracting useful information
from a small set of block spins. It is a
method for observing the behavior of
the system over a large region wi thout
ever dealing expli ci tly with the confgu
rations of all the spi ns in that region.
Each block spin represents nine spins
in the original lattice. The complete set
of block spins, however, can also be re
garded as a spin system in its own right,
with properties that can be investigated
by the same methods that are applied to
the original model. It can be assumed
that there are couplings between the
block spins, which depend on the tem
perature and which determine in t urn
the probability of each possible spin
confguration. An initial guess might be
that the couplings between block spins
are the same ones specifed in the origi
nal lattice of Ising spins, namely a near
est-neighbor interaction with a strength
PROBABILITY DISTRIBUTION for a system of block spins is found by adding up the prob
abilities for all the confgurations of the original lattice that contribute to each confguration
of the block spins. The calculation is shown for a system of six spins on a triangular lattice.
Two blocks of three spins each are formed from the lattice, and each block is replaced by a sin
gle spin whose orientation is determined by majority rule. The six spins have 64possible confg
urations, which are assigned to columns in such a way that all the confgurations in each col
umn give rise to the same block-spin confguration. For example, all the confgurations in the
column at the far left have at least two spins in each block pointing up, so that they are rep
resented by two up block spins. The coupling strength in the original lattice is set equal to .5,
which yields the nearest-neighbor probabilities shown at the top of the page. From this set of
numbers a probability is calculated for every confguration of the original lattice; then all the
probabilities in each column are added up to give the probability of the corresponding block
spin confguration. The block-spin probabilities are not the same as those specifed for the orig
inal lattice, which implies that the coupling strength is also dferent, as is the temperature.
given by the parameter K, the recipro
cal of the temperature.
This guess can easily be checked, be
cause the probability distribution for
the confgurations of at least a small
part of the block-spin system is already
known; it was computed from the con
fgurations of the original lattice in the
course of defning the block spins. Sur
pri si ngly, thi s hypothesi s i s generally
wrong: the block spins do not have the
same couplings as the spins in the orig
inal model. Assuming that only adja
cent sites interact and that they have a
coupling strength equal to K gives the
wrong set of probabilities for the confg
urations of the block spins.
THI RD-STAGE
BLOCK SPI NS
FOURTH
STAGE
BLOCK
SPI NS
Ising spins j ust below the Curie temper
ature will seldom reveal that the model
is slightly magnetized. At this tempera
ture there is only a small excess of one
spin direction over the other, and the
many small-scale fuctuations obscure
the overall bias_ After several applica
tions of the block-spin transformation,
however, the smaller fuctuations disap
pear and the long-range magnetization
becomes obvious.
Much of the physical meaning of the
block-spin transformation is to be found
in the way the couplings between spins
change. The rules for deriving the new
couplings from the old ones at each
f= .99f ORI GI NAL LATIICE
FI RSTSTAGE BLOCK SPI NS
SECOND-STAGE BLOCK SPI NS
THI RD-STAGE
BLOCK SPI NS
FOURTH
STAGE
BLOCK
SPI NS
to be shown, and after the fourth stge there are only 36 spins left,
each one repreenting more than 6,000 site in the original lattice. In
the frst stage any fuctuations whose scale of length h smaller than
three lattice unit are eliminated by the averaging procedure. The sec
ond stage remove the fuctuations between three and nine lattice
units, the thi stage those between nine and 27nnit, and so on. When
the initial temperature habove 1
-
,the spins become more nearly ran-
dom in appearance with each iteration and large-scale fuctuations
disappear; when the temperature is below 1
-
, the spins become more
nearly uniform and what fuctuations remain are small in scale. When
the starting temperature is exactly equal to 1
-
, large-scale fuctua
tions remain at all stage. Because the block-spin transformation
leave the large-scale structure of the lattice unchanged at the Curie
temperature, a system at that temperature is said to be at a fxed point.
OV
1979 SCIENTIFIC AMERICAN, INC
stage are often complicated, but the ef
fect of the change can be i l l ustrated by a
quite simple example. Although the as
sumptions are not realistic, I shall dis
cuss a model in which no coupl ings with
a range longer than the original, nearest
neighbor interaction are introd uced.
The only change in the coupling is an
adj ustment in the value of whi ch is
equivalent to a shift in the temperature.
Moreover, this adj ustment in K will
have a simple form: at each stage in the
procedure the coupl ing strength in the
new lattice wi l l be set equal to the
sq uare of the coupl ing in the old lattice.
I f the new coupl ing is denoted K' , it is
given by the equation K' K2.
Suppose in some i ni ti al state K is
eq ual to 1 / 2 (which meals that the tem
perature has been given an initial value
of 2 in the arbitrary units employed
here). In the thinned-out lattice formed
as a product of the block-spin transfor
mation K will be replaced by K' , with a
value of ( 1 /2) 2, or 1 / 4. Repeating the
transformation yi elds successive values
of 1 / 1 6, 1 / 25 6 and so on, in a series that
K15 (K' 4)'
K1 4 (K, ,) '
K' 3 ( K, ,) '
K" (K, , )'
K
"
(K
l O
) '
K, o (Kg)'
J
Kg (K.)'
2
K. (K,)'
0
u
L
K, (K.)'
2
J
K. (Ks)'
Ks (K4)' .
0
K4 (K3)'
K3 (K,)'
K, (K, ) '
K, (Ko)'
Ko
rapidly approaches zero. With each iter
ation the spin system is converted into a
new system that not only has a thinner
lattice but also has weaker coupl ings
between the spins. Since K is equal to
1 / Jthe temperature increases with each
iteration and the lattice approaches the
li mi t of infnite temperature and ran
dom spins.
I f the initial coupl ing strength is set
equal to 2 (so that the temperature has
a value of 1 / 2), the coupling increases at
each stage in the cal culation. After the
frst block-spin transformation the cou
pling strength is 4, then 1 6, then 25 6; ul
timately the strength becomes infnite.
At the same time, of course, the temper
ature fall s and the system approaches
the state of zero temperature, in which
al l the spins are aligned.
I t shoul d be emphasized that what is
being observed is not the evolution of
any single spin system as the tempera
ture changes. Nothing is being heated or
cooled. Instead a new spin system is be
ing created at each stage, a system dis
tinguished by a diferent set of cou-
.C 1
. l l l l
1 0 8 7 C 4
COUPLI NG STRENGTH (K 1 /T)
Z 1 . C
TEMPERATURE (T 1 1K)
l l l l l
. .ZC .Z . 1 7 . ! . !
CHANGE IN THE COUPLING BETWEEN SPINS is part of the renormalization-group
transformation. The adjustment that must be made to the coupling strength with each itera
tion can take many forms, but a simple example is presented here: If the strength of the cou
pling in the original lattice is given by the number K, then in the new lattice the coupling
strength is equal to K-.Any initial value of Kgreater than 1 must approach infnity when Kis
squared repeatedly; any value less than 1 must approach zero. The special value K 1 re
mains unchanged no matter how many times the transformation is repeated. Because the tem
perature can be defned (in appropriate units) as the reciprocal of the coupling strength, the
renormalization-group transformation can be seen as establishing a correspondence between
the original lattice and a new, thinned-out lattice that will generally have a diferent coupling
strength and a diferent temperature. It is only at the fxed point, which corresponds to the Cu
rie temperature, that the coupling and the temperature remain invariant with 'a value of 1 .
pl ings between spins. The large-scale or
long-range behavior of the new lattice
is equivalent to the behavior that woul d
be observed in the original l atti ce at a
diferent temperature.
There is one initial value of K that
does not diverge either to infnity or to
zero, namely the value K 1 . Since 1 2 is
equal to !, K' remains equal to K no
matter how many times the transforma
tion is repeated. When K is equal to
1 , the system is said to be at a fxed
point, where continued application of
the renormalization-group transforma
tion leaves all essential properties of the
lattice unchanged. Actually the values
K 0 and K infnity also represent
fxed points, since zero squared is still
zero and infnity squared is still infnity.
Zero and infnity, however, are consid
ered trivial fxed points, whereas the val
ue K 1 corresponds to the critical
point.
I n this di scussion of the block-spin
technique all the efects of the transfor
mation have been expressed through a
single parameter: the nearest-neighbor
coupling strength K. Actually many oth
er parameters are introduced by the
transformation, each one corresponding
to a longer-range coupling. All the pos
sible combinations of these parame
ters can be represented geometrically by
constructing an imaginary multi dimen
sional space in which distance measured
along each dimension corresponds to
variation in one of the parameters. Ev
ery initial state of the spjn system and
every block-spin transformation of it
can be represented by a point on a sur
face somewhere in this parameter space.
T
he macroscopic properties of a ther
modynamic system near the critical
point are determined by the temper
ature. To be more precise, properties
such as the spontaneous magnetization,
the susceptibility and the correlation
length are functions of the amount by
which the temperature of the system de
parts from the critical temperature, 1
For this reason it is convenient to defne
the temperature in such a way that all
EVOLUTION OF A SPIN SYSTEM in response to repeated renormalization-group transfor
mations can be described as tbe motion of a point on a surface constructed in an imaginary,
multidimensional space: tbe parameter space. Tbe form of tbe surface is defned by all tbe con
plings between block spins, but only tbe nearest-neigbbor coupling, K, is considered bere. Tbe
surface bas two peaks and two sinkboles, wbicb are connected to a saddle point. Tbe trajectory
follo
w
ed by tbe point tbat represents tbe state of tbe system is determined entirely by tbe slope
of tbe surface. An initial value of K sligbtly greater tban 1 corresponds to an initial position
sligbtly to one side of tbe ridgeline tbat connects tbe peaks. After several block-spin transfor
mations tbe point rolls down tbe bill, passes near tbe saddle point and veers of into one of tbe
sinkboles, wbere Ktends toward infnity. An initial value of Ksligbtly less tban 1 leads to a simi
lar trajectory on tbe otber side of tbe ridgeline and terminates in tbe otber sinkbole, wbere K
approacbes zero. Wben Kis equal to exactly 1, tbe point remains permanently on tbe ridgeline,
approacbing equilibrium at tbe saddl e point. Botb of tbe sinkboles are fxed points (since tbe
values of K 0 and K infnity do not cbange witb furtber renormalization-group transforma
tions), but tbey are considered trivial fxed points. Tbe saddle defnes tbe critical fxed point.
critical points are equivalent. A suitable
quantity is the reduced temperature, t,
defned as the diference between the ac
tual temperature and the critical tem
perature, divided by the critical temper
ature; thus tis equal to 1 1/1. On an
ordinary temperature scale such as the
Kelvin scale the critical temperatures of
diferent systems fall at diferent values,
but all critical points hav'e the same re
duced temperature, namely zero.
All critical properties are proportion
al to the absolute value of the reduced
temperature raised to some power. The
problem of describing critical phenome
na is to determine what that power is, or
in other words to determine the values
of the critical exponents. For example,
the magnetization, A, of a spin system is
given by the proportionality A ,t , ,
where (the Greek letter beta) is a crit
ical exponent and where the vertical
lines designate the absolute value of t.
The magnetic susceptibility i s propor
tional to 1 /,t , , where (the Greek let
ter gamma) i s another exponent. The
correlation length i s associated with a
third exponent, V (the Greek letter nu),
in a relati on of the same form: the length
is proportional to 1 /1 t , .
The earliest attempts to formulate
a mathematical description of critical
phenomena were theories of a kind that
are now called mean-feld theories. The
frst of these was introduced in 1 873
by J . D. van der Waals as an explana
tion of phase changes in fuids. A theo
ry of magnetic phase transitions was pro
posed in 1 907 by Pierre Weiss. In 1 93 7
L. D. Landau of the Academy of Sci
ences of the U. S. S. R. proposed a more
general formulation of mean-feld theo
ry, thereby providing a framework in
which many physical systems could be
discussed. In all these theories' the state
of any selected particle is determined by
the average properties of the material as
a whole, properties such as the net mag
netization. In efect all particles in the
system contri bute equally to the force at
every site, which is equivalent to assum
ing that the forces have infnite range.
Mean-feld theories are qualitatively
successful. They account for important
1
1979 SCIENTIFIC AMERICAN, INC
K_ * 1.1
K_ * 1.01
K_ * 1.001
K_ * 1.0001
K_ * 1
SLOPE OF THE PARAMETER SURFACE in tbe vicinity of the
critical fed point determines tbe macroscopic properties of the Ising
model. If trajectories are plotted for many initial values of Knear the
critical value (wbicb in tbis case is K 1), it is the slope at the saddle
point tbat determines bow quickly tbe trajectories veer of toward the
trivial fed points at K 0 and K infnity. If tbe surface is com para-
tively fat (lc)tI,a trajectory witb an initial value of Ksucb as K 1. 01
passes close to tbe saddle point. Wben tbe surface i s more steeply
curved (right), tbe corresponding trajectory bends more abruptly
toward tbe sinkbole. Because tbe temperature is tbe reciprocal of K
tbe slope near tbe fed point reveals bow tbe properties of tbe sys
tem change as the temperature departs from the critical temperature.
features of the phase diagrams of fui ds
and ferromagnets, the most notable of
these features being the existence of a
critical point. The quantitative predic
tions, however, are less satisfactory: the
theories give the wrong values for the
critical exponents. For , the exponent
that governs the spontaneous magneti
zation, mean-feld theory implies a val
ue of 1 12; in other words, the magneti
zation varies as the square root of the
reduced temperature. The exponent as
sociated with the susceptibility, , is as
signed a value of 1, so that the suscep
tibility is proportional to 1 /1 1 1 . The ex
ponent for the correlation length, v, is
1 12, so that this quantity is proportional
to 1 / v,
The exponents calculated from mean
feld theory suggest a plausible form for
each of these functions. The magnetiza
tion has two possible values ( +vand
-V) at all temperatures below the crit
ical point, and then it vanishes above the
critical temperature. Both the suscep
tibility and the correlation length ap
proach infnity as l nears zero from ei
ther above or bel ow. The actual values
of the mean-feld exponents, however,
are known to be wrong.
For the two-dimensional Ising model
the critical exponents are known exactly
from Onsager' s solution. The correct
values are 1 / 8, 7/ 4 and v I ,
which difer signifcantly from the pre
di ctions of mean-feld theory and imply
that the system has rather di ferent be
havior. For example, the magnetization
is proportional not to the square root of
the reduced temperature 1 but to the
eighth root of l. Similarly, the suscepti-
!
bility is given by the reciprocal not of l
but of l raised to the I . 75th power,
which makes the divergence near the
critical point steeper and more abrupt.
The reason for the quantitative fail ure
of mean-feld theories is not hard to
identify. The infnite range assigned to
the forces is not even a good approxima
tion to the truth. Not all spins make
equal contri butions; the nearest neigh
bors are more important by far than any
other spins. The same objection can be
expressed another way: the theories fail
to take any notice of fuctuations in spin
orientation or in fuid density.
I.n a renormalization-group calcula
tion the critical exponents are deter
mined from the slope of the parameter
surface in the vi cinity of the fxed point.
A slope is simply a graphic represen
tation of a rate of change; the slope
near the fxed point determines the rate
at which the properties of the system
change as the temperature (or the cou
pling strength) i s varied over some nar
row range near the critical temperature.
Descri bing the change in the system as a
function of temperature is also the role
of the critical exponents, and so it is rea
sonable that there is a connection be
tween the exponents and the slope.
Renormalization-group calculations
for the two-dimensional Ising system
have been carried out by several work
ers. In 1 973 Niemeijer and van Leeuwen
employed a block-spin method to study
the properties of a system of Ising spins
constructed on a triangular lattice. I
have applied a somewhat di ferent re
normalization-group technique, called
spin decimation, to a square lattice. In
spin decimation, i nstead of assembling
blocks of a few spins each, every other
spin in the lattice is held fxed while a
probability distri bution is computed for
the remaining spins. These calculations
were much more elaborate than the
model calculation described here; in my
own work, for example, 2 1 7 couplings
between spins were incl uded. The crit
ical exponents derived from the calcula
tion agree with Onsager's values to with
in about . 2 percent.
D
ecause an exact sol ution is known for
the two-dimensional Ising model,
the application of the renormalization
group to it is something of an academic
exercise. For a system of Ising spins in a
three-dimensional lattice, however, no
exact solution is known. A method has
been devised, by Cyril Domb of Univer
sity College London and many others,
for fnding approximate values of the
exponents in the three- dimensional case.
First the properties of the system at high
temperature are determined with great
precision, then these properties are ex
trapolated to the critical temperature.
The best results obtained so far by this
method give values fer the exponents of
. J J , 1 . 25 and v . 63.
Although extrapolation from a high
temperature solution leads to good ap
proximations for the critical exponents,
it provides l ittle intuitive understand
ing of how the system behaves near the
critical point. A renormalization-group
calculation gives essentially the same
values for the exponents, but it also
explains important universal features of
critical behavior.
1979 SCIENTIFIC AMERICAN, INC
Two remarkable facts about the ex
ponents in the three-dimensional Ising
model should not be overlooked. The
frst is simply that the val ues are difer
ent from those for the two-dimensional
model. In mean-feld theories the dimen
sional ity of space does not enter the cal
culations and so the critical exponents
have the same values in any space. The
second surprise is that the exponents are
not integers or ratios of small integers,
as they are in mean-feld theories. They
may even be irrational numbers.
If it is surprising that the spatial di
mensionality infuences the critical ex
ponents, it is equally remarkable that
certain other properties of the model
have no efect at all. An example of such
an irrelevant parameter is the structure
of the lattice. In the two-dimensional
Ising model it makes no diference
whether the lattice is rectilinear, as in
my own work, or triangular, as in the
model employed by Niemeijer and van
Leeuwen; the critical exponents are the
same. By extension, in a real ferromag
net the great variety of crystal structures
all yield identical critical behavior.
There is an intuitive j ustifcation for
the irrelevance of the lattice structure
and of other microscopic properties. A
change in the form of the lattice has a
large efect on events at the scale of the
lattice spacing, but the efect diminishes
as the scale of interest increases. In a
renormalization-group calculation the
fuct uations at the scale of the lattice
spacing are averaged out after the frst
few iterations, and so models with many
diferent lattices have the same critical
behavior. Through the renormalization
group the appearance of the same crit
ical exponents in many systems is seen
to result from the topography of the sur
face in parameter space. Each lattice
structure corresponds to a diferent po
sition in parameter space, but at the crit
ical temperature every lattice is rep
resented by a point somewhere along
the ridgeline. After repeated renormal
ization-group transformations all these
systems converge on the same fxed
point, namely the saddle point.
The idea that certain variables are ir
relevant to critical phenomena can be
extended to systems other than ferro
magnets. A fui d near i ts critical point,
for example, has the same properties as
the three-di mensional Ising model of a
ferromagnet. In order for this identity to
be understood some correspondence
must be establ ished between the macro
scopic properties of the f ui d and those
of the magnet. The magnetization,
which is the number of up spins minus
the number of down spins, can be identi
fed with the density diference in the
fui d: the density of the li qui d phase mi
nus the density of the vapor phase. J ust
as the magnetization vanishes at the Cu
ri e temperature, so the density difer
ence falls to zero at the critical point of
the fui d. These q uantities-the magnet i-
zation and the density diference-are
called the order parameters of their re
spective systems. The susceptibility of
the magnet, which is the change in mag
netization for a given small change in
the applied magnetic feld, is analogous
to the compressibility of the fui d: the
change in density that results from a giv
en small change in pressure. Like the
s usceptibility, the compressi bility be
comes infnite at the critical point. The
critical behavior of the fui d and that of
the three-dimensional Ising model are
identical in that they have the same sur
face i n parameter space. The two sys
tems have diferent initial positions on
the surface, but they converge on the
same saddle point and hence have the
same critical exponents.
T
he similarity observed in the criti
cal behavior of fuids and of ferro
magnets is an instance of a more general
hypothesis called critical-point univer-
PREDI CTI ONS OF
MEAN-FI ELD THEORI ES
EXACT SOLUTI ON
0
REDUCED TEMPERATURE (|)
} = J
0
REDUCED TEMPERATURE (|)
P Y2
.
'8
0
REDUCED TEMPERATURE (|)
0
REDUCED TEMPERATURE (|)
P 1
o 0 +
REDUCED TEMPERATURE (|) REDUCED TEMPERATURE (|)
CRITICAL EXPONENTS express the dependence of macroscopic properties on the extent
to which the temperature of the system departs from the critical temperature. The temperature
is most conveniently given in the form of the reduced temperature, l, defned by the equation
l ~ I I/I.All macroscopic properties are then proportional to the absolute value of traised
to some power; the power is the critical exponent for that property. The exponents and power
laws in the graphs at the left are those predicted by mean-feld theories, which ignore all fuctu
ations. The exponents in the graphs at the right are derived from an exact solution of the two
dimensional Ising model reported in 1944 by Lars Onsager of Yale University. The exponents
show how the properties of the system change as the temperature or the coupling strength is
changed; that is the same information conveyed by the slope of the surface in parameter space
near the critical fed point. The exponents can be determined from the slope, and calculations
by the author and others for the two-dimensional Ising model give values close to Onsager!s.
J
1979 SCIENTIFIC AMERICAN, INC
sality. According to the hypothesis, only
two quantities determine the critical be
havior of most systems: the dimension
ality of space and the dimensionality of
the order parameter. These quantities
are labeled respectively dand H. All sys
tems that have the same values of dand
H are thought to have the same surface
in parameter space and the same critical
exponents. They are said to be mem
bers of the same universality class.
The dimensionality of space is seldom
difcult to determine, but the dimen
sionality of the order parameter re
quires more careful consideration. In
magnetic systems, where the order pa
rameter is the magnetization, N is the
number of components needed to defne
the spin vector. The vector of an Ising
spin can be oriented only along a single
axis, and so it has only one component;
for the Ising model H is equal to 1. A
spin vector that is allowed to point any
where in a plane has two components,
which are customarily drawn along the
two axes that defne a plane. Similarly,
UNIVERSALITY CLSS THEORETICAL MODEL
d = 2 n = \
Ising model in
to di mensions
n = 2
X7model i n two
dimensions
n = 3
Heisenberg model
in to dimensions
d > 2 n = "Spherical" model
d = 3 n = 0
Selfavoiding
random walk
n = J
Ising model in
three dimensions
n * z
X7model in
three di mensi ons
z
Quantum chromo
dynami cs
a vector that can point anywhere in
three-dimensional space has three com
ponents, so that H equals 3 .
For the three-dimensional Ising mod
el d equals 3 and H equals I. Ordinary
fuids belong to the same universality
class. The space in which the fuid exists
clearly has three dimensions. The order
parameter-the diference in density be
tween the liquid and the vapor phases
is a quantity that has only a magnitude
and hence only one component; it can
be expressed as a single number, j ust as
the value of an Ising spin can be.
Several other physical systems are
members of this class. A mixture of two
liquids such as oil and water exhibits
critical behavior near the temperature
where the component fuids become
completely miscible in each other, a
temperature called the consolute point.
At temperatures below the consolute
point the mixture separates into two
phases, and the order parameter is de
fned as the concentration diference be
tween the two phases, another quantity
PHYSI CAL SYSTEM ORDER PARAMETER
Adsorbed films Surace density
Helium-4 films
Amplitude of
superluid phase
Magnetization
None
Conformation of long- Density of
chain polymers chain ends
Uni axial ferromagnt Magnetization
Fluid near a critical Density difference
point between phases
Mixture of l i quids Concentration
near consolute point diference
Al loy near order- Concentration
disorder transition diference
Planar ferromagnet Magnetization
Helium 4 near super Ampl itude of
fluid transition superfl ui d phase
Isotropic ferromagnet Magnetization
None
Quarks bound i n
protons, neutrons, etc.
UNIVERSALITY HYPOTHESIS states that diverse physical systems behave identically near
their critical points. In most cases the only factors that determine the critical properties are the
dimensionality of space, d,and the dimensionality of the order parameter, II. For magnetic sys
tems the order parameter is the magnetization, and its dimensionality is the number of compo
nent needed to describe the spin vector. Most systems with the same values of d and II are
members of the same universality class and share the same critical exponents. For example,
ferromagnets that resemble the three-dimensional Ising model, fuids, mixtures of liquids and
cerain alloys are all members of the class with d Jand h ~ 1; graphs of their properties near
B critical point should all have the same form. The interpretation of some values of dand II is
less obvious, and values such as II 2 can be defned mathematically but correspomto no
known physical system. The XImodel and the Heisenberg model are similar to the Ising mod
el but describe ferromagnets whose spin vectors have two and three components respectively.
!+
that can be expressed as a single num
ber. Alloys such as brass have a transi
tion between an ordered phase, where
the two metals occupy alternate sites in
a regular lattice, and a disordered phase,
where their distribution is less uniform.
The order parameter in this system is
again a concentration diference, so that
H equals I. All these systems are expect
ed to have the same critical exponents
as the three-dimensional Ising model.
So are some real ferromagnets, those
that are easily magnetized only along a
single axis. The available experimental
evidence confrms these predictions.
The universality hypothesis would be
trivial if the critical exponents had the
values of integers or simple fractions
such as I /2. Many physical laws share
such exponents, and there is no compel
ling reason for postulating a connection
between them. Gravitation and electro
magnetism both have an inverse-square
law (an exponent of 2),but that coin
cidence does not demonstrate that the
two forces are identical. The correspon
dence of exponents does seem remark
able, however, when the values are not
round numbers but fractions such as . 63 .
The convergence of many systems on
these values cannot be coincidental. It is
evidence that all the details of physical
, structure distinguishing a fuid from a
magnet are less important than the geo
metric plOperties expressed by the val
ues of dand H.
The two-dimensional Ising model
(d 2, H I ) typifes a class of systems
that are confned to two-dimensional
space. One example is a thin flm of liq
uid; another is a gas adsorbed on a solid
surface. An ordinary ferromagnet falls
into the class with d 3 and H 3, that
is, the lattice is three-dimensional and
each spin has three components, so that
it can point in any direction. When the
spins are constrained to lie in a plane,
the class is reduced to d 3 and H 2.
In this same class are the superfuid
transition of liquid helium 4 and the
superconducting transitions of vari ous
metals.
O
ther universality classes have val
ues of dand H whose interpretation
is somewhat less obvious. The case of
d 4 is of interest in the physics of ele
mentary particles, where one of the four
spatial dimensions corresponds to the
axis of time. In a theoretical lattice of
spins callee the spherical model, where
an individual spin can have any magni
tude and only the total of all the spins i s
constrained, N is efectively infnite. A
selfavoiding random walk through a
lattice of points, or in other words a ran
dom walk that never occupies the same
lattice site more than once, describes the
folding up in space of a long-chain poly
mer; Pierre Gilles de Gennes of the Col
lege de France has shown that this prob
l em belongs to a universality class with H
equal to zero, In theoretical models H
1979 SCIENTIFIC AMERICAN, INC
c8f
tc 0 m
bf
t t
t
oturc g((Quc m
cnnt Ou
tc t tbct8
andu8rt ct.
A=l
sa Aaanew
ltna8 c
o atha
aab t
ha Ounct.
1lo tbtOu c
t|da0u8taram
Otsc natu
snc dcat t uc
ncn 0
Ia rsts a
am|a auct mtl
m ss
t
c8aO aat
c u8 O tc h
sctdinio
at ml' sat mtc8
ot t8u88lnm r|o
`c t8ot c
cul oc c m
0Oc olc o o8ngc8
tbo 8 o
caadeocu u
bsobc a Ot
8X btm mllc s
|
Ycs, start my subscription with a Ircc currcnt issuc.
l
cncosc a $I 0chcck $I ?Iorcign) Ior si bi-monthy m-coor issucs.
I
Pcasc scnd Gardcn to:
g lamc
|
l
Addrcss
Edition wanted:
DChicago Horticultural Society
DHorticultural Society of New York
DLos Angeles Arboretum
Clip and mail to:
DNew York Botanical Garden/
Queens Botanical Garden
DNo regional news, please
Subscription Department, Gardcn magazine, Botanical Garden, Bronx, N. Y. I 0458
1979 SCIENTIFIC AMERICAN, INC
If you have ever taken a lux
ury sports car through a tight
turn, you know the feeling.
It's the sense of supreme pre
cision with which this trim,
compact camera proclaims
its Nikon heritage. A feeling
that is borne out by the pro
fessional quality pictures
the Nikon FE delivers with
automatic ease. And one
that, unlike other fine things
in life, is readily affordable.
WlD lDCKODL, you can
simply focus and shoot . . .
and rely on its Nikon elec
tronics to give you sharp,
magnificently exposed
photographs, automatically.
Or, switch to manual opera
tion and enjoy complete
creative control over every
exposure, more easily than
you ever thought possible.
ZO0CB, this is a camera
that makes no compromise
in its supreme Nikon quality.
Stroke the advance lever,
and feel the smoothness of
precision gearing machined
to microscopic tolerances.
Press the exposure button,
and hear the shutter respond
with hushed precision. Look
through the bright, silver
coated viewfinder, and see
your picture snap into sharp
focus with a fingertip touch.
DOH, lOO, that the world's
gretest photographic sys
tem stands behind your
Nikon FE. Add the dynamic
firepower of motor drive, at
up to . shots a second.
Banish darkness with the in
genious automatic thyristor
fash. Explore new perspec
tives through more than
Nikkor lenses, the same
superb optics chosen by
most professionals for their
sharpness and color fidelity.
OtIDC@UtSIt
1DC MBOD M
For those who prefer only
manual exposure control, the
Nikon FM offers the reliable
guidance of one-step elec
tronic metering. It's as com
pact and precisely responsive
as the FE and costs even less.
At your Nikon dealer.
h| kon|nc. GardcnC|,hcwYork11.
| nCnada.h|konCnada, | nc.
_h| kon|nc.J/
pedeace
a seasee|pemct|ea.
ne8|keaL.
1979 SCIENTIFIC AMERICAN, INC
can even take on the value Z, although
the physical meaning of a negative num
ber of vector components is not clear.
The only values of d and H that have
a straightforward physical meaning are
integer values. This is parti cularly clear
in the case of d, since a space wi th a non
integer number of dimensi ons is hard
even to imagin. In renormalization
group calculat ions, however, d and ll
appear in equati ons where they can be
allowed to vary conti nuousl y over some
range. It is even possi ble to draw a graph
in which the values of cri ti cal exponents
are plotted as continuous functions of d
and ll . The exponents have well-defned
val ues not only for integer di mensions
but also for all fractional dimensions be
tween the integers. Such a graph shows
that the exponents approach the values.
given by mean-feld theory as the num
ber of spatial dimensions approaches 4.
When d is equal to exactly 4, and at al l
higher val ues of d, the mean-fel d values
are exact. Thi s observation has gi ven
rise to an important method of perform
ing renormal ization-group cal cul ations.
The di mensionality of space is ex
pressed as being equal to 4 L, where L
( the Greek letter epsi lon) is a number
that is assumed to be small. The cri tical
exponents can then be calculated as the
sum of an i nfnite series of terms incl ud
ing progressively hi gher powers of L. If L
is less than I , a high power of L wi ll have
a small value, and reasonable accuracy
can be obtained by neglect ing all but the
frst few terms in the infnite series.
This calculation method, which is
called the epsilon expansion, was devel
oped by Michael E. Fi sher of Cornell
University and me. It is a general meth
od for solving all the problems to which
mean-feld theory can be applied, and it
represents the nat ural successor to Lan
dau' s theory. Indced, it supplies answers
in the form of corrections to the values
.given by mean-feld theory. The block
spin method i s the more transparent
techni que, but the epsi l on expansion is
the more powerful one.
It is not enti rely surpri sing that the
critical exponents should converge on
the mean-feld val ues as the number of
spatial dimensions increases. The fun
damental assumpti on of mean-feld the
ories is that the force at each lattice site
is infuenced by condi ti ons at many oth
er si tes. The number of nearest-neighbor
sites increases along wi th the number of
spatial dimensions. In a one-di mension
al latti ce each si te has j ust two nearest
nei ghbors, in a two-di mensional lattice
four, in a three- di mensional lattice six
and in a four-di mensional lattice eight.
Hence as the dimensionality increases,
the physical situation begins to resemble
more closely the underlying hypothesis
of mean-feld theory. It remains. a mys
tery, however, why d 4 should mark a
sharp boundary above which the mean
feld exponents are exact.
In this article I have disc ussed main-
CONTOURS OF (EXPONENT ASSOCI ATED WI TH MAGNETIZATION)
"SPHERI CAL" MODEL - g
'2 n = *
HEI SENBERG
MODELS
AYMODELS
I SI NG MODELS
NONPHYSI CAL
REGI ON OF
NEGATIVE
EXPONENTS
n = 7
n = 6
n = 5
n = 4
n = 3
n = 2
n = 1
n = O
n = -1
a a
n =
-
2
o = ! o = 2 d = 3 o = 4 d > 4
DI MENSI ONALI TY OF SPACE
CONTOURS OF j (EXPONENT ASSOCI ATED WITH SUSCEPTI BI LITY)
"""ERICAL" MOLC-
f fj
HEI SENBERG
MODELS
j
r
/
+
AY MODELS 9
I SI NG MODELS
n =
n = 7
n = 6
n = 5
n = 4
n = 3
n = 2
n = 1
n = O
n = -1
======= .. -.. n = -2
d = 1 d = 2 d = 3 d = 4 o > 4
DI MENSI ONALI TY OF SPACE
D
U
B
D
U
L
D
L
U
U
L
2
L
m
2
U
C
D
b
|
b
D
b
L
L
U
3
|
U
L
2
L
m
2
U
C
V ARIA TlON OF CRITICAL EXPONENTS with the dimensionality of space (d) and of the
order parameter (u) suggests that physical systems in diferent universality classes should have
diferent critical properties. The exponents can be calculated as continuous functions of J
and R, but only systems with an integral number of dimensions are physically possible. In a
space with four or more dimensions all the critical exponents take on the values predicted by
mean-feld theories. The graphs were prepared by Michael E. Fisher of Cornell University.
11
1979 SCIENTIFIC AMERICAN, INC
Iy the applications of the renormaliza
tion group to critical phenomena. The
technique is not confned to those prob
lems, however, and indeed it did not be
gin with them.
The procedure called renormalization
was invented in the 1 940' s as part of
the development of quantum electrody
namics, the modern theory of interac-
UP SPI N
DOWN SPI N
PRESENCE
OF AN ATOM
ABSENCE
OF AN ATOM
ZI NC ATOM (
COPPER ATOM
/
/ /
tions between electrically charged par
ticles and the electromagnetic feld. The
difculty encountered in the formula
ti on of the theory can be understood as
one of mUltiple scales of length. For
some time it had been apparent that
the charge of the electron predicted by
quantum-mechanical theories was inf
nite, a prediction that was in seri ous
/
\
/
UNI AXI AL FERROMAGNET
FLUID NEAR CRI TI CAL POINT
BRASS NEAR ORDER-DI SORDER TRANSI TI ON
PRESENCE _
OF A PARTI CLE
OR AN ANTI PARTI CLE
ABSENCE
OF A PARTI CLE
OR AN ANTI PARTI CLE
QUANTUM FI ELD THEORY
LATTICE SYSTEM can be interpreted as a model not only of a ferromagnet but also of other
physical systems that have fuctuations on many scales. The Ising model describes a nniaxial
ferromagnet, one with a preferred axis of magnetization. It can also be applied to a fuid near
its critical point, where each lattice site either is occupied by an atom or is vacant, so that the
fuctuations become variations in density. An alloy such as brass has a similar structure, where
each site is occupied by one kind of metal or the other. In all these systems the fuctuations are
thermal; in the quantnm feld theories that describe the interactions of elementary particles
there are quantum fuctnations of the vacuum, which allow particles and antiparticles to ap
pear spontaneously. A simple quantnm feld theory can be formulated on a lattice by specify
ing that the particles and antiparticles can be created and annihilated only at the lattice sites.
l b
confict with the measured charge. The
renormalized the'ory of electrodynam
ics does not abolish the infnity; on the
contrary, the electron is defned as a
point particle whose "bare" charge is in
fnite. In quantum electrodynamics the
bare charge has the efect, however, of
inducing a charge of opposite polarity
in the surrounding vacuum, which can
cels most of the infnty, leaving only
the small net charge that is observed in
ordinary experiments.
One can imagine a probe particle that
could measure the electron' s charge at
arbitrarily close range. At long range
it would fnd the familiar fnite value,
which is the diference between the bare
charge and the induced charge. As
the layers of shielding were penetrated
the measured charge would increase,
and as the range was reduced to zero
the charge would become infnite. The
renormalization procedure provides a
means for subtracting the infnite shiel d
i ng charge from the infnite bare charge
so that a fnite diference results.
In the 1 950' s it was pointed out by
several workers, among them Murray
Gell- Mann and Francis E. Low, that the
renormalization procedure adopted for
quantum electrodynamics is not unique.
They proposed a more general formula
tion, which is the original version of the
renormalization group. In their applica
tion of the method to quantum electro
dynamics a mathematical expression is
constructed that gives the magnitude of
the charge at some defnite distance
from the electron. Then the form of the
expression is examined as the distance at
which the measurement is made is al
lowed to approach its limiting value of
zero. The arbitrariness of the procedure
is in the choice of the initial distance.
Any value can be selected without
changing the ultimate results, so that
there is an infnite set of eq uivalent re
normalization procedures.
A "group" in mathematics is a set of
transformations that meets a special re
quirement: the product of any two trans
formations must also be a member of
the set. For example, rotations are trans
formations that make up a group, since
the product of any two rotations is also a
rotation. What this means in the case of
the renormalization group is that the
procedure can be iterated indefnitely,
since applying the procedure twice i s
equivalent to applyi ng the product of
two transformations. Actually the re
normalization group is properly called a
semigroup because the inverse of the
transformation is not defned. The rea
son for this can be seen plainly in the
block-spin technique applied to the two
dimensional Ising model. A block of
nine spins can be condensed into a single
average spin, but the original spin con
fguration cannot be recovered from the
average because essential information
has been lost.
The version of the renormalization
1979 SCIENTIFIC AMERICAN, INC
group outlined in this article di fers in
several respects from the one intro
duced by Gell - Mann and Low. The ear
lier version of the technique is useful
only for understanding problems that
can be solvd by one of the traditional
methods of physics: by fnding some ap
proximate expression for the behavior
of a system and then cal cul ating better
approximations as a series of pertur
bations departing from the original ex
pression. Moreover, in the original for
mulation only one quantity is allowed
to vary; in the example given above it is
the charge of the electron. As a conse
quence the surface in parameter space
is not a multidimensional landscape but
a mere l i ne. The modern version of the
renormal ization group, which was in
trod uced by me in 1 97 1 , gives access to
a much broader spectrum of physical
problems. What is equally important, it
gives a physical meaning to the renor
mal ization procedure, which otherwise
seems purely formal.