P H YS ICAL REVIEW B VOLUME 24, NUMBER 1 1 JULY 1981

Duality of the block transformation and decimation for quantum spin systems

J. Solyom'
Department of Physics, University oflflinois at Urbana-Champaign, Urbana, Illinois 6180(
(Received 27 May 1980)

The zero-temperature renormalization-group transformations for quantum spin systems are

analyzed. The block transformation and the decimation-type transformation used in the study
of the one-dimensional Ising model in transverse field are extended to the quantum version of
the Potts model and Ashkin-Teller model. It is shown that for these self-dual models the two
kinds of renormalization-group transformations are dual to each other and therefore give the
same result for the critical behavior. The duality persists even if the higher-lying states are tak-
en into account in a perturbational way.

I ~ INTRODUCTION lem. The critical exponents can also be calculated

The renormalization-group (RG) transformations
have proved to be very powerful in the description of
critical phenomena. ' The momentum shell integra-
tion method, when combined with the large-order
perturbational calculation2 can produce good values
for the critical exponents of three-dimensional sys-
tems. 3 In the real-space RG transforrnations there is
some arbitrariness in the choice of the mapping and
the results depend strongly on the mapping. The
usual transformations can be classified into two broad
categories. In the block transformation4 a cluster of
spins is mapped onto a single block spin according to
an ad hoc rule. The coupling between the new block
spins is obtained from the couplings of the individual
spins between the neighboring blocks, weighted,
however, with the weight with which the individual
spins appear in the block spin state. The decimation
transformation' is an alternative approach. There
one eliminates a fraction of the spins by considering
the effective couplings these spins mediate between
the remaining spins. These transformations were ori-
ginally invented for classical systems and the RG
transformation is performed in a way that the parti-
tion function or free energy, from which the critical
behavior is derived, should remain invariant during
the RG mapping.
The extensions of the RG transformations to quan-
tum systems ' have been extensively used recently
both to describe critical phenomena and to under-
stand the properties of quantum field theories.
Quantum effects are usually irrelevant for the critical
behavior of systems near their phase transition point,
nevertheless quantum models are often used to cal-
culate critical properties, since d-dimensional classical
statistical mechanical problems can be mapped onto
d —1 dimensional quantum mechanical problems.
The ground-state energy and first excited state energy
of the quantum problem are related to the partition
function and coherence length of the classical prob-
from the behavior of the quantum equivalent,
Since in the quantum problems one is interested in
the ground-state energy and low-lying excited state
energies, the number of degrees of freedom should
be thinned in the quantum RG transformation in a
way that these states should be well approximated.
This is achieved by keeping the low-lying states in
each step of an iterative procedure and neglecting
some higher-lying states. This can be done for the
quantum systems in several different ways. The
method introduced by Jafarey et al. ' is based on split-
ting the system into blocks. The eigenvalue problem
of the finite block is solved and as many lowest levels
are retained as it is necessary to map these states to
the quantum states of a single block spin. The cou-
pling between the blocks is obtained again from the
couplings between the individual spins taking into ac-
count the wave function of the block state.
An alternative approach to the quantum RG treat-
ment of the Ising model and lattice gauge theories
has been proposed by Fradkin and Raby. ' They de-
cimate the number of lattice sites by fixing the quan-
tum states on a fraction of the sites, keeping that
state of the intermediate spins which gives the lowest
energy for the fixed configuration of the selected
spins and then mapping this state to a new state
where only the selected sites have spins.
The two RG transformations seem quite different.
The first one is similar to the classical block transfor-
rnation, the second one is more like a decimation
transformation. We will show in this paper that the
two transformations are in fact very closely related.
The decimation type RG transformation on quantum
spin systems leads to the same result as a block
transformation on the dual model.
The setup of the paper is as follows. A general
" description of block transformation and decimation
for quantum spin systems is given in Sec. II. The
quantum version of the Potts model (of which the Is-
ing model in transverse field is a particular case) is

24 230
24 DUALITY OF THE BLOCK TRANSFORMATION AND. . . 231

studied in Sec. III using both RG transformations. cal mechanical problem in higher dimensions. In the
The duality relation between the two transformations quantum RG transformation the number of degrees
is discussed in Sec. IV. These results are given for a of freedom, the number of states is decreased by
scale factor b =2. In Sec. V the considerations are mapping the chain with N sites to a chain with N/b
extended for arbitrary scale factor. A similar rela- sites in a way that the q" states of the new system
tionship between the two RG transformations for the should possibly coincide with the q lowest states of
Ashkin-Teller model is shown to exist in Sec. VI. the original chain.
The effect of the higher-order perturbational. correc- In the block transformation this mapping is "
tions is considered in Sec. VII. It is shown that the achieved by grouping the sites into cells (each having
duality persists even if new couplings are introduced b sites) and mapping the lowest-lying states of the
by these corrections. Finally, Sec. VIII contains a cells onto equivalent new states. The sites will be in-
discussion of the results. dexed by a cell index I (I = 1, 2, . . . , N/b ) and a fur-
ther index a(u=1, 2, . . . , b). The Hamiltonian is
split into intracell and intercell parts
II. GENERAL FORMALISM
Hintra + Hinter (2.2)
In this paper we will be concerned with one- where the intracell part is
dimensional quantum systems on a lattice. Assuming
N/b
a nearest-neighbor interaction T;;+1 and a single site
term U;, the total Hamiltonian of the system has the
H;„„,= gH„„(I) (2.3)
I 1

form with

H=XT, „,+XU, , (2. 1) H„„(I)= X T(

a
b-1

1
(, + X a
b

1
U.
, (2.4)

where periodic boundary condition has been imposed, while the intercell part is
though this is not important in the further calcula- N/b

tions. H, „„,= XH, „„,(I, I+1) (2.5)

I 1
If the system can be in q states at each site, the to-
tal number of states is q . We are interested in the with
ground state and low-lying excited states, either be-
H;„„,(I, I +1) = Tis i+i i (2.6)
cause the phase transition occurs at T =0, or because
these quantitites of the quantum mechancial problem Solving the eigenvalue problem of a single cell
are the analogs of the relevant quantities of a statisti- first, one finds qb states. They have the form

(2.7)

where ls;) ~
is the ith state at site I, n Keepin. g the first site of each cell is selected to be kept while the
q lowest-lying states, they can be identifed as the q other sites are to be eliminated. We separate the
states of a renormalized entity p, '. Hamiltonian into two parts, a fixed spin (fs) and in-
termediate (im) part as
Ipp)i P=l 2 (2.8)
H =Hf, +H; (2.9)
The new Hamiltonian acting in the space of p, states
should have the same form as the Hamiltonian in Eq. where Hf, contains the single site terms on the select-
(2. 1) acting in the space of s states, only the cou- ed sites,
plings can have renormalized values. N/b
The new single site term is obtained from the ener- H„= XHr, (I) (2.10)
gy spectrum of the low-lying cell states. The coupling
between the neighboring cells is calculated by requir- with
ing that the matrix elements of the new Hamiltonian = U( i
Hr, (l) (2. 11)
between the p, states should be the same as the ma-
trix elements of the original Hamiltonian between the while H; contains the single site terms on the inter-
corresponding cell states in the s-state representation. mediate spins and the coupling terms
The decimation transformation' starts from a dif- N/b
ferent splitting of the Hamiltonian (2. 1). Using the H~ = $H; (I, I+1) (2.12)
same convention as above for indexing the sites, the I~1
232 J. SOLYOM 24

with q possible configurations. For each such con-

b—1
b
figuration the states on the intermediate sites are
Him(iii+1)
a
$ 1
Tl, a;I, n +1+ Ib;I, +1, 1 + $
a~2
Ul chosen in such a way that the energy be minimal,
This is equivalent to finding the lowest-lying eigen-
state of Hl l+1 with fixed states s; and s;'l+/, on the
(2. 13) 1, 1 1

end sites t, 1 and I +1, 1. Denoting this state by

Fixing the states on the N/b selected sites gives 4 I, 1+1 ( Sii Sii+I 1

IIIII+1(&I I
si i1 1) = lsii
J
1)I, I &II+I (si isi I) Isii+I 1) I+1, 1 (2. 14)

where
b
(2. 15)
S. a~2
'l, a

is a linear combination of the states of the intermediate sites. The states of the system which are kept in the RG
transformation are of the form

Isi ) I, I jri, 2(si si ) Isi ) 2, 1 jul I+1(si si I) Isi 1) I+1. 1

" (2. 16)

This state will then be mapped onto the state with H„ii given in Eq. (2.4), while for the decimation
transformation they are eigenstates of the Hamiltoni-
lj '1 1)1IP'2 1)2 'IP'I~i 1)I+1' (2.17) an of the intermediate spins

The Hamiltonian acting in the space of the p, states N/b

should have the same form as the original Hamilton- Hii = H; = $ H; (I, I +1) (2. 19)
ian, the new couplings should be calculated from the
requirement that the matrix elements of the renor- with H; given in Eq. (2. 13). The rest of the Hamil-
rnalized Hamiltonian between the states given in Eq. tonian
(2. 17) should be the same as the matrix elements of
N/b N/b
the original Hamiltonian between the states given in
Eq. (2. 16).
V =H;nicr= QHlnicr(1, 1+1)= $ Tl pi+I, I, (2.
, 20)
l 1 l 1

From the formulation of the problem it is clear

and
that the block transformation is conveniently done in N/b N/b
such a representation where the s and p, states are V =H„= QH„, (I) = $ U, , (2.21)
eigenstates of the single site term, the decimation is
conveniently done in a representation where the
respectively, are treated as perturbations. If ~III;) and
nearest-neighbor coupling is diagonal.
ilij ) are eigens tates of Hp with energies E; and E, ,
In the RG transformations presented until now the
respectively, such that they are the products of the
higher-lying states are completely neglected. The ma-
low-lying ce11 states, while P is an eigenstate with
trix elements are calculated between states which are
products of low-lying eigenstates of individual cells.
energy E,
such that at least one of the cells is in a
higher-lying state, the RG transformation should be
Hirsch and Mazenko' have shown that a systematic
done by comparing the matrix elements of the renor-
improvement can be achieved by taking into account
malized Hamiltonian to the matrix elements calculat-
the higher-lying states in a perturbational way. Using
ed in second order in V:
the same mapping of the s states to the p, states as
before, the requirement is not simply that the matrix '
elements of the Hamiltonian should be unchanged, (III, ( Hii+ V+ —, /Villi. )
but the shift due to virtual excitation of the higher-
lying states is taken into account. The states we are
working with are eigenstates of a truncated Hamil-
i a
+Ej a
tonian. In the case of the block transformation they
are eigenstates of the intracell part (2.22)
N/b
H p = H;„„,= $. H„p ( I) (2. 18) In one case the perturbation is a nearest-neighbor
l 1 coupling between the end sites of neighboring ce1ls,
24 DUALITY OF THE BLOCK TRANSFORMATION AND. . . 233

awhile in the other case the single site term on select- ponents of the Potts spin, and 0 is a diagonal matrix
ed sites serves as 'a perturbation.
The two transformations can in general lead to com-
pletely different approximation schemes. We will see
M2
in the next sections that for self-dual models the two
277 l
RG transformations lead to equivalent results. Block co=exp -=
(3.3)
transformation is the same as decimation in the dual q
model and therefore the critical exponents calculated
in the two ways are equal. q-1

while Hf;, 14 is the "transverse field" which rotates the

spins,

Hr„id= —h X $ M;", (3.4)

i 1k 1
III. RENORMALIZATION-GROUP
where
TRANSFORMATIONS FOR THE QUANTUM
VERSION OF THE POTTS MODEL 010 ~
0

M=
001 0
(3.5)
The one-dimensional Ising model in transverse
field has been used extensively as a test of various 0 0 0
quantum RG transformations, since the exact solu-
tion of this model is known. The Potts model'9 be- "
ing a simple generalization of the Ising model, we A. Block transformation
will consider now the quantum RG transformations
on this model. The preceding paper20 contains the results of the
The one-dimensional Hamiltonian version of the block transformation. Here I only quote the results.
two-dimensional classical Potts model has been dis- Starting with the Hamiltonian given in Eqs. (3.1)
cussed in the preceding paper. The Hamiltonian —(3.4), new conplings are generated, which corre-
contains two terms spond to the simultaneous flip of two neighboring
spins. The strength of these new couplings is renor-
H Hp «+Hf' 14 (3.1) malized in such a way that a well-defined relationship
is maintained and in fact only a single new coupling
where Hp, «, is the usual Potts coupling between the is needed in terms of which all other couplings can be
neighboring spins, expressed. Using now the notations of Ref. 20, the
new couplings are denoted by A. 1, P~, and A3 and they
satisfy the relations
. .,. = ——g1kX ()„"(),e;,",
N N q-1
H, „,= —) Xg. . (3.2) A. 2=x'. 1, 3=x2A. A, 1 (3.6)
i 1

The recursion relations for the renormalized cou-

where s; = 1, 2, . . . , q; q is the number of com- plings are as follows

= — —
Xicell
1+ q —1 a' 2
q
—2+2b2 [)ii (1+a )'b'+2(q 2) li2(1+ a ) ab + (q 2)'Z, a']
) 2cell
=
—1 —2+2b [2ki(1+a)b +X2[(1+a)b(q —3+b2) +2(q —2)ab]
1+ q a2 i/2
i~
q
+ (q —2) Z3a (q —3+ b') ) (3.8)

X3 ii
=
—2+2b2 2
[4', b' +4k, b(q —3 + b') + k, (q —3 + b')'] (3.9)
q
with
f
1/2
—qh + 2 —q 2 1
a = 3+ qh + q
(3.10)
q
—1 Xi 2q
li.
2q
X3
q2

b= q
-h+
q )ii —(q -3)&3
+ -h+-
q lii —(q -3))i3
+ 2(q
- 2) i 1/2

(3.11)
2A2 2 2q 2 2q
234 J. SOLYOM 24

and

q~cc[1 = ~1' —~1 (3.12)

where
—2 — —2 —I 1/2

Et = —(q —.2) h —q h. 3 qh —q X3 + q
(3.13)
2q 2q q

Xt+(q —3)X3 —(q —3) x3

1

—4 2(q —2) „p
1/2

gI q h- q h xt
(3.14)
2q 2 2q q

These recursion relations have been analyzed in where

Ref, 20. Here we want to compare them with the N/2 N/2 q-1
results of the decimation transformation.

(3.16)
B. Decimation transformation and
N/2 q-1
We will outline the decimation for the b =2 case,
V= —h g1/(QMI. (3.17)
I 1
when only the odd sites are retained after the de-
cimation. According to the prescriptions given in Fixing the states at the (/, 1) sites, the eigenfunctions
Sec. II, the Hamiltonian is split as of Hp are easily obtained. If two neighboring fixed
spins are in the same state i, the lowest-energy con-
H =Hp+ V (3.15) figuration of the intermediate spin is

xt(ii) = +cli —1&+ Ii &+cIi +1& +

1
[1+(q —1) c']/' (cl1) +cl2&+ +cIq&) (3.18)

where
/

—2 —2 1/21
= ' —X+ q
h+ ——
q
+(q —1)h' (3.19)
C
(q -1)h h. h

and the energy is

—2 — —2 1/2

Et (ii ) = —h. — P. — h + (q —1) h'

q q
2
h
2
(3.20)

If the two end spins are in different states, the lowest-energy configuration is

xt(/& j) = (q 1
2+2d&) tn
(I1)+I2)+ +II —1&+dII&+II'+1&+ "+Ij —1&+dlj &+Ij+1&+" +Iq)),
with
(3.21)

(—it —— 'h, ——(q —4) h

'(q —4) h + [ [ — ]'+2(q —2) h' }'
= '
d 2 2
) (3.22)
2h
and the energy is

E, (i & j) = ——X ——, (q —2) h —( [-; X —, (q —4) h ]'+2(q —2) h~ —I' ' (3.23)

When one performs now the mapping as discussed in Sec. II, the new Potts coupling between the renormalized
spins is obtained from the energy difference of the configurations when the neighboring fixed spins are in identi-
cal or different states:

x„„=E,(i w j) —Et(ii) (3.24)

The renormalized field is given by the matrix element between states, where one spin is different. It turns out
that the matrix element will depend on the configuration of the neighbors. We have to introduce three different
 DUALITY OF THE BLOCK TRANSFORMATION AND. . . 235

fields defined by the matrix elements:

—hl = (V(IH Ill(& = &ii~'I&llil), l &i (3.25)

h2=—&jk(IHI(ij ). l & j, i &k, &k j (3.26)
—h3=(jli[H)ikj), i &k, i Ai, jWk, j A( (3.27)
The difference in the renormalization comes clearly from the fact, that in calculating these matrix elements,
identical neighbors appear twice in Eq. (3.25), once only in Eq. (3.26), and there are no identical neighbors in Eq.
(3.27). For a self-consistent renormalization we have to introduce these couplings from the very beginning.
Xl(ii) and Xl(i W j)
still have the same form as in Eqs. (3.18) and (3.21), but now

—g+ —2 —2 &/2

e= 1 q
h3+ (l. —q h3 + (q —1) hl (3.28)
(q -1)h, 2 2

—2
1

—2 &/2

El(ii) = —X —q
~,
h3 — X —q h3 +(q —1)hl (3.29)
2 2

d = [—(l + —[hl —(q —3) h3] + ([ —(l. + —[hl —(q —3) h3] ]2+2(q —2) h22)' 2] (3.3O)
2h2

El(l & j) = —,X ——,' [hl+(q —3)h3] —(( 2 h, + 2 [hl —(q —3)h3]] +2(q —2)h22)'
' 2
(3.31)

The renormalized value of X is still given by the energy difference as in Eq. (3.24), but with the new energy ex-
pressions. The renormalization of the three fields is obtained by calculating the matrix elements (3.25) — (3.27)
between the renormalized states. We get

= 1
+2(q —2)h2(1+c)cd+(q —2)2h3c ]
hl ll
1+ q
—1 c22 q
—2+2d22 [hl(1+c) d (3.32)

h 2cell [2hl(1 + c) d2+ h, [(1+c) d(q —3 + d') +2(q —2) cd]

[1 + ( q —1)c2] 'i'( q —2 + 2 d2) 3(2
+ (q —2) h3c (q —3 + d') I, (3.33)

[4hld'+4h2d(q —3+d ) +h3(q —3+d )'] (3.34)

(q 2 + 2d2)2

The three fields are not really independent of each IU. DUALITY RELATIONS IN THE POTTS MODEL
other. Since in the unrenormalized model
h~ = h2 = h3, in each step of the iteration The Hamiltonian of the quantum version of the
Potts model has been given in terms of the matrices
h3cell/hlcell (h2celljhlcell) (3.35) 0; and M;. These matrices commute if they belong
Comparing now these recursion relations with to different sites, while on the same site they satisfy
those obtained in the block transformation [see Eqs. the following algebra:
(3.7) —(3.14)], it is seen immediately that the substi-
tution
n,kn, l = n,k+l, (4. 1)

g ~ qh, qh &
~ A&, qh2 ~ X2, qh3 ~ X3 (3.36)
kMI Mk+l (4.2)
generates the results of the block transformation Mkg l k/O IMk (4.3)
from the decimation transformation. We will show in
the next section that these relations are the conse- Let us introduce the dual lattice and define the
quence of the self-duality of the Potts model. operators 0&' and M&' on the sites of the dual lat-
236 J. SOLYOM 24

tice, i.e. , on the links of the original lattice: taneously. It is convenient for the further compar-
ison to shift the energies by —h so that
(4.4)
(5.3)
~rk g q-kg k
(4.5) The intracell part of the Hamiltonian will mix the fol-
It is easy to see that these operators satisfy the same lowing q' states:
algebra. For example 1'1'1') l'2'q'& 1'3'(q —1) '& ' ' ' 'q'2'&
I I I I1

(4.6) 12'q'1') —1)'1'), ' ' ' Iq'2'1')

I3 (q
The Harniltonian can be written in terms of these —1 ) ')
2'1'q'& 13 '1'(q ' ' ' q'1'2'& (5.4)
new operators as I I

I2'2'(q —1)'), I2'3'(q —2)') I2'(q —1)'2')

H =Hp„„+Hf;,(d (4.7)
I3'2'(q —2)') ' ' ' ' ' ' Iq'q'3')
where now
q-1
Hp„„= ——X $M, 'k
!V

(4.g) One has to find the lowest energy eigenstate of the

i 1k 1 intracell Hamiltonian in the subspace spanned by
and
these states.
Alternatively in the decimation transformation one
has to find the lowest-energy configuration of two in-
(4.9)
termediate spins. Fixing the two end spins, the inter-
mediate spins can be in q' configurations
We have used the relation cuq = I in deriving this
form of the Hamiltonian,
Since the new operators satisfy the same algebra as
the original ones, comparison of the two forms of the
Hamiltonian leads to the duality relationship": for
any value of )t/q and h the model should behave in the I22). 133). " Iqq) (5.5)

",
~

same way as the model in which X/q and h are inter-

changed. The relations in Eq. (3.36) are the generali-
zations of this duality relationship for the renormal-
ized Hamiltonian.
V. CALCULATION FOR LARGER CELLS
The quantum RG transformations usually do not
give good values for the critical exponents when the
scale factor b =2. One way to improve the results is
to take larger scale factors. ' The analysis becomes
very cumbersome and the lowest-energy states of the
cell problem can be found numerically only. It is,
however, possible to see, without solving the prob-
lem, that the duality of the block transformation and
decimation persists even for b 2. )
Let us look at the case b =3. It is convenient to
use in the block transformation the states defined by
(5. 1)
They are eigenstates of Hf;, 1d,
H„„, 1') = —(q —1) h
l I
1 '),
(5.2)
while Hp, «, will now flip the neighboring spins simul-
I23), "Iq(q —1))
We have to find the lowest-energy configuration of
the two intermediate spins when the end spins are
fixed, e. g. , in the ll) ll) state. One can easily
convince oneself that the eigenvalue matrices in the
two transformations are related by the duality rela-
tions given in Eq. (3.36).
In the same way as for the b =2 case, the next
lowest-lying states of the cell problem in the block
transformation are in an orthogonal subspace which
can be generated starting from the state ll'1'2').
Analogously one can look for the lowest-energy state
of the two intermediate spins in the decim'ation
transformation when the two end spins are fixed in
the states ll) I2). Again the two eigenvalue ma-
trices are related by the duality relations.
So in general one can show by writing down the
eigenvalue matrices and the wave functions of the
lowest-energy configurations, that the two RG
transformations are dual to each other. This is a
consequence of the fact that H; in Eqs. (2. 12) and
(2. 13) and H;„,„, in Eqs. (2.3) and (2.4), of which the
lowest-lying states are considered and Hf, in Eqs.
(2. 10) and (2. 11) and H;„„, in Eqs. (2.5) and (2.6)
which are treated as perturbations, are dual to each
other. It is important to emphasize that this duality
persist even after renormalization, when new cou-
plings are generated.
24 DUALITY OF THE BLOCK TRANSFORMATION AND. . . 237

VI. QUANTUM Z(4) MODEL OR where Hq is the field term in the Hamiltonian.
ASHKIN-TELLER MODEL Equivalently we could use a linear combination of
the states
Let us consider now the RG transformations for
the quantum version of the Ashkin-Teller model. In
the classical Ashkin-Teller model ' there are four 11'& = —,' (ll&+ I»+13&+ 14&)
possible states at each lattice site. The energy of the
system depends on the configuration of the nearest 12'& =-2'(I»+~l»-13&-~14» .
neighbors. It is — —, X& if the neighbors are in the (6.2)
same state, i.e. , for the configurations 111), 122),
'(I» —I»+13& —14&),
13'& = —,

133), and 144). The energy is +2 h. ; for the con- 14'& =-2'(I» -~12& -13&+~14&) .
figurations 113) and 124), whiie for the-configurations
112), 114), 123), and 134) the energy is —, X2. In the
case when X~ = A2, we recover the four-state Potts
model, while A2=0 is the usual clock model. which are eigenstates of Hq,
In the quantum version of the Ashkin-Teller
model spin-flip terms are introduced. The transverse
field which flips the spins can be defined by the rela- Hhll'& =-»ill'&. Hhl2 ) 2~212'&,
tions (6.3)
Hh I » = +~211 —& i12) —&213 ) —)ri 14)
&
Hhl3') =2&i13'). Hhl4'& =2&214'&
Hhl» =-&il» +&212) ~113) ~214&
(6.1} The Ashkin-Teller coupling part of the Hamiltoni-
Hh13 &
=-»I » —&i I» + &213 &
—&F14&
an, H&, in this representation will flip the neighbor-
Hh 14 & = —~ i I
I& —~212 —& i 13 +»14
& & & ing spins, e.g. ,

H. I I'I'& = 4' —'I ——12'4') ——

Il
' '& —— '
14'2'&,
4 4
13'3'&
4
'

H. I2'4'& = —ll'I'&+ —'

'12'4'& '13'3'& —— '14'2'& . — 4

—
4 4 4
(6.4)
H. I3'3'& = ' ll'I'& ——'12'4'&+ — '13'3'& —— '14'2'&,

H. 14'2'& = — 4

l'I'& ——
4
'12'4'& - —
'
I
'13'3'& + —'
4

4
.
4

4
4

4
14'2'&

Similar relations hold for the states 11'2'), 12'1'), l3'4'), and 14'3'), for 11'3'), 12'2'), 13'1'), and 14'4') as well
as for 11'4'), 12'3'), 13'2'), and 14'1').
In the block transformation ft is convenient to use this representation. In the same way as in the Potts model,
new couplings are generated by the renormalization. Accordingly, we will generalize the Ashkin-Teller coupling
part of the Hamiltonian, H„, to have seven couplings:

—
He'll'I') =' ll'1') - —l2'4') - — ' l3'3') - — 14'2') ' '
4 4 4 4

Hhl2'4') = ——'I + — ' ' —— '13'3'& ——

Il
''& 12'4'& 14'2'&
4 4 4 4 (6.5)

H„13'3') = —— ll'1') —— 12'4') + — 13'3') —— 14'2')

4 4 4 4

H. =-— '11'I ——
14'2'& ' 12'4') —— '13'3') + — 14'2'),
'&
4 4 4 4
238 J. SOLYOM 24

and

0.11'2'& = —'11'2'& —— 12'1'& - —'13'4'& ——' 14'3'&,

'
4 4 4 4

Hil2'1'& = — ' —
'
11'2'& + —'
4
——' 12'1'&
4
13'4'& 14'3'& ,

—
4
(6.6)
Hil3'4'& = ' 11'2') ——
' +—
' ——' 14'3')
12'1'& 13'4'& ,
4 4

—
4 4

Hil4'3'& = '11'2') ——'12'1'& ——'13'4'& + —

'14'3'& .
4 4 4 4

T"e reiations f' or the set of states 11'4') 14'1'), 13'2'), 12'3') are the same as in Fq. (6.6), whffe

4
— 4 4
—
4
14'4'

Hil2'2') = — '12'2'&- —
'11'3'& + — '13'1'& —— '14'4'&
4 4 4

Hil3'1'& = —
'11'3'& ——'12'2'&+ —
4
'13'1'& —— '14'4'&,
4

Hil4'4'& —
= '11'3'& —— '12'2'&- —'13'1'& + —'14'4'&

Ap can be set equal to zero, it does not play any role The corresponding wave functions are
in the renormalization of the other couplings.
Solving first the ce11 problem with two sites, the (11'1') + ~ t 12'4')
four lowest-lying levels of the cell are two nondegen- 1+2a,z+azz»»
'"
erate and one doubly degenerate levels. A nonde-
generate level is at Ei, which is the solution of + az13'3') + a|14'2') ) (6. 11)
—Ei —4h) with
4 4 4
—Ej +4h2 A. 4 (—El —4h1) 4+ —, ~i ~z
ai= —
( El +4hz —h.4/4)
4 4 4 &z+
=0 z
&i&z
A, g
—Ei+4hi (6. 12)
4 4 4 —4hi)kz+-
( Ei
A, 4
—Ei+4h2 a2=
4 4 4 ( Ei +4ht ) kl + —
—AzXz

(6.8)
a doubly degenerate level is at 2+2b', '" (b 11'2') + b12'1')
„,
+13'4') +14'3') ) (6. 13)
A. i + A, 3
2= 2— (b 11 '4') + b14'1')
8 2+2hz '»
As+ A6
2' 1/2 + 13'2') +12'3') ),
2hi+ + (6.9)
t with

and another nondegenerate level is at 4 A. i

—3 A.
'2hi+
A. 2+ A. 4 ~, + ~, 8
E3 =2h2— '2
8 2 &/2

—X4 ~2 ' '2 + 2h)+ + ~s+

+—
r
1/2
A2 As 4
2h2+ (6. 10)
8 2 (6.14)
24 DUALITY OF THE BLOCK TRANSFORMATION AND. . . 239

states of the renormalized spin is chosen as:

I1') ll y(1) I2')cell

, „, (cll'3'&+cl3'1'&+12'2'&+14'4'&)
(2+2c')'/' ~

(6.15) e"'- 14'&"" (6.17)

The renormalized values of the fields are obtained
with
from the energy spectrum of the renormalized states:
c= —'2h2+ A. 2
—X4
8
hl, = —, (E3 —El)
ll
'
5
(6. 18)
l/21 h2cell = 4 (E2
1
—El ) — (E3 —E2)1

—4
+—
I

4
A. 2 A.
+ 2h2+ (6.16)
8 2 The renormalized X couplings can be calculated
from the matrix elements between states differing by
The mapping of these four states of the cell to the two spin flips. We get

~1cell
2+2b2
[)1/(1+a, ) b +2 qx(1 +a )1b(a 1+a2) +X3(a/ +a2) ]
1+2a12 +a22

~2cell [X2(1+a2) c +4X4(1+a2)cal+4X4al ]

1+2al +a2 2+2c
~3cell
2 +2b2 2 +2c2
[kl(b +c)2+2xq(b +c)(1 +bc) + X3(1+bc)2]
(6. 19)
~4cell [4X2b2+4X4b (1 + b2) + X4(1+ b2)']
2+2b
"scell =
1
[Xl(1+al) b(b + c) + Xq[(1+al) b(1+ bc) + (b + c) (al + a2) ]
(1+2a/2 + a22)'/2(2+2b )(2+2c )'/2
+x3(a/+a2)(1 +bc) I
= [2 X2(1 + a2) cb+ Xq [ (1 + a 2) c (1 + b ) + 4a ] + 2 X4a (1 + b ) ]
(1+2a,' +a')'~'(2+2b')(2+2c2)'/2
) 6celi 1 b 1
/

In the physical model, before renormalization there Fixing now every other spin on the chain, the
are only two couplings, xl = x3 = x5 and P 2 = x4 = A6, eigenvalue problem for the fixed configuration is
and as it is easy to see, the generated new couplings easily solved. If the two end spins are in the same
are not independent of each other. There are in fact state, say in 1 ), the lowest-energy configuration of

=,
I

two new couplings, since the intermediate spin is

2
( XScell~Xlcell)
X3cell/ Xlcell
(6.20)
X, (11)
1+2d/2 +d22, „,(11) + d, I2)
/14ce/1/ X2cell ( X6cel// X2cell I

Let us look now at the decimation transformation.

+d I3) +d, I4))
Here it is more convenient to work in that represen- (6.22)
tation, in which the Ashkin-Teller coupling is diago- with
nal. The renormalization will lead to new spin-flip [—El(11) —Xl ]h3+hlh2
terms, more precisely the spin-flip amplitude will dl=
depend on the two neighboring spins. %e will intro-
[-El(11) + X2 —h4]h2+2hlh3
duce six spin-flip terms defined by the matrix ele- (6.23)
ments [—El (11) —h. l ] h3 + h h21
d2= —
—hl = (121IH4 I111) = &221IH41112) . [ El (11) + Xl ] h1 + h2h3
—h2 = (131IH/, I111) = (331IH/g I113) and El(11) is the lowest-energy solution of
—E, —Xl —hl
—h3= (131IH4I121) = (241IH/, I132)
—hl —El +f2 —h3
-h4 = &141IH4 I»» = &341IH4 I»», (6.21)
—h2 —h3 —El+Pl —h3 =0
-h3 = &241IH4 I»» = &231IH4 I»» . —hl —h4 —h3 —El+~2
h4 = (231IH/, —
I112) = (241IH/, I122) (6.24)
240 J. SOLYOM 24

If the two end spins are in the {12&, {14&, {23&, or with
{34& configurations, the lowest-energy configuration 1 —(), +h, —h, )
{
of the intermediate spin is different. For the {12&

=,
2h5
state, e.g. ,
+ [ —(ht+ht —h4)'+4hg ]' ~], (6.29)
Xt(12)
2+2e' , /, (e{1& +e {2& + I3& +14&)
'" and the energy is
(6.25)
E)(13) = 1
—hg —h4) —[ —
„(4+her —hg)
1
with t (h. g

e = 1 .
( —h. t+ —(ht
1 1
—h3) + { [ —
]
z, + — (h) —A3)] 2
1
+4A,']'/' . (6.30)
h5+ h6
2 2

+(h5+h6)' {' ') The renormalized Ashkin-Teller couplings are sim-

ply given by the energy differences of the various
(6.26)
configurations:
and the energy is

E, (12) = —ht —h3)

~
(&g h. i,ea = Ei (13) —Et (11)
(6.31)
[~ (g, +h, h3) +(h5+h6)&]i/& (6 27)

Similarly, when the two end spins are in the {13 & h. g„„——[E, (12) —E, (11)]—[E, (13) —E, (12)]
configuration, we get

2 +2y2
', „,(f{»+{2&+f{3&+14&) .
t/2
The spin-flip amplitudes can be obtained by calculat-
ing the matrix elements of Eq. (6.21) between the re-
(6.28) normalized states. We get

a= [h)(1 +dt) e'+2hg(1 +d))e(dt +dg) +A3(d) +dt) ]

(I +2d",t + d,')
ht t (2+2e')
t

hgqqa
= [hg(1 + dt)~f +4A6(1 +dg) fdt +4A4dt ]
(I +2di~t +dp~~ 2+2't
A3„a=
(2+2et) (2+2f~
& &
[ht(e +f) +2A5(e +f)(1+ef) +A3(1+ f)e]

~, + e') + A4(1 + e')']

1
A4 [4A)e' +4A6e ( I (6.32)
IJ
(2+2e')'
»,
g g

1
hq„a= {h~(1 +d~) (ee+ f)+h [(51+d~)e(1+ef) +(d +tdq)( +ef)]
(I +2dt~ +dt~ )'/~(2+2e~)(2+2ft)'/~
/» /

+h3(dt+dt)(I+ef) )

A6 = {2hg(l +dg)ef +A6[(1 +dp)f(1 +e~) +4d~e]+2A4d)(l +e ) ]

~~

(I+2dt t +dp & '/t/& 2+2e~ &

2+2f & t/&

Comparing now these recursion relations with those models is equivalent to the block transformation in
obtained in the block transformation, Eqs. (6. 18) and the dual model.
(6.19), we see that after the substitution

4h;, i =1, 2 VII. PERTURBATIONAL CORRECTIONS TO

(6.33) THE RECURSION RELATIONS
4h,. A, j=1 2, . . . , 6 A consistent way to improve the results of the
the two transformations lead to identical recursion re- block transformation has been suggested by Hirsch
lations. In the block transformation new A. couplings and Mazenko. '" The intercell Hamiltonian is treated
are introduced, while in the decimation new spin-flip as a perturbation and the higher-order corrections are
terms entered, but in a dual manner, So the decima- calculated in a consistent perturbational way. In
tion transformation in these general four state second order, e.g. , the renormalized Hamiltonian is
24 DUALITY OF THE BLOCK TRANSFORMATION AND. . . 241

obtained to match the second-order matrix elements and

given in Eq. (2.22). N/b
A consistent treatment of the Potts model or V = —h $ rrl"t (7.11)
Ashkin-Teller model is prohibitively difficult due to
the large number of new couplings generated in The eigenstates of Hp are easily obtained by fixing
higher orders. We will show here for the Ising model the spins on the selected sites and solving the cell
that the new couplings are again dual to each other. problem with the fixed end spins. For a scale factor
We write the Hamiltonian of the Ising chain in b = 2, the wave functions with both end spins up are
transverse field in the form
N N
H= ~
h XlT
2
err cr+~ (7.1)
i i
1 1
= IT»i(TT&IT& ~

where cr" and cr' are the Pauli operators. Hirsch and
(7. 12)
Mazenko' have shown that in second order a new =
coupling of the form
e~(TT)
, , „, IT&( -cIT& + Il&)IT&
~ ~i-1~i
2 X
~i+1
2
(7.2) = T»z(
I T T) I T & ~

is generated. Note that in Ref. 14 both the Hamil-

tonian and the new coupling are written with a dif- where
ferent choice of the coordinate system.
c = —[(X +h')' ' —X] (7.13)
The recursion relations obtained by Hirsch and h
Mazenko' can be written in the form
and the energies are

a')' E)(TT) = —(lt +h )' =(X +h )'~'

+— 8(l(1 —ya&)&
1

Ep(TT)
h„g = —(El —Eo) 1 1
E) Eo (7. 14)
(7.3) The wave functions when both end spins point
down can be obtained by flipping all spins.
(1+a) 2&1
—a
(7.4) When the two end spins are in different states, the
2(1+ a') 1+ a' wave functions are

(1 —a')' el(Tl) = IT&(IT&+Il&)Il&=IT&&1(Tl) Il& ~

(7.5)
8(1+a')' E
(7.i5)
e~(Tj) = IT)(IT& —Il&)ll&=IT&x&(Tj)Ij) .
with
1
1/2

0 —— —2h
2 A.
'
+4h (7.6) with energies
2,
1/2 Et(Tj) = —h. Ep(Tj) =h (7.i6)
Ep=— A.
+4h (7.7) The intermediate spins are eliminated by taking the
2
lowest-energy state for each configuration, i.e. , keep-
F. i = —— (7.8) ing Pi(TT) and Pq(T j) and mapping these states onto
2 the TT) and ITj) configurations of the renormalized
spins. The first-order renormalized value of A. is ob-
tained from the energy difference of the TT) and
In the decimation transformation the Hamiltonian is I

split as ITl) configurations:

H=Hp+ V (7.9) lt. . = Ei( l) —El(T

u T T) (7.17)

with The renormalization of the transverse field can be

"X g ~;.—
calculated from the matrix element between two con-
~;.„
N/b b N/b b-1
H hgl~l a~2
0 X ~(". 2 a~1
figurations differing by one spin flip on a selected
lt 1 site. This gives
N/b
~l, b ~I+1, 1 (7.10)
(1+c)' (7.18)
I 1 2(1+ ')
242 J. SOLYOM 24

It is straightforward to calculate the matrix ele-' and leads to an extra renormalization of h'
ments appearing in the second-order correction in Eq. —c 2
Ah„jj = —
1
(2.22). The second-order energy shifts of the can- E +c2
(7.24)
1
figurations, when the fixed spin orientations are
or 1j1,
lead to a second-order
correction to the coupling between the renormalized
Collecting the various contributions, finally the recur-
sion relations are
spins:

(1 —c')'
.
x. ii = Ei(1l) —Ei(11)
,
„, (1-")'
1
(7. 19)
4(1+ c') Ei (1j) Eg(11) (7.2S)
4(1+ c ) Ei(l j) Ei(11)
In second order there is also a possibility for the
simultaneous flip of two neighboring spins, a process (1+c)' (7.26)
which is not present in the original Hamiltonian. The 2(1+ c') 1+ c'
matrix element of the second-order term of Eq.
(2.22) between the states , (1 —c)' 1
(7.27)
I
I'& xi(I'1)11& xi(11)11»i(1jli &
8(1+ c ) E~(1j) E~(11)

and
A comparison of these relations with those given in
~ ~
1I &x, (I j) j&x,
1 (j j) I j&x, (jl) IJ ) Eqs. (7.3) —(7.8) shows again that the two transfor-
mations lead to identical results if the h h, /2 inter-
1S change is made.
2(l —c)' 1 1
(7.20)
It is furthermore apparent that the new coupling
generated in the block transformation, Eq. (7.2) and
8(1+ c')' Et(11) E~(1j) the one generated in the decimation transformation
are dual to each other. For the special case of the Is-
whereas the matrix element between the states
ing model the duality relations in Eqs. (4.4) and (4.5)
" I
~
) xi(11)11)xi(1j) 1 j) xi( ji) 1 j ) ~
can be written in the usual form:
'z x
and (7.28)
j(i
" II &xi(l j) I j»i(j1)11&xi(1i)li& " lx z z
l+1 (7.29)
1S
and therefore
, (1 —c')' 1
+ 1 (7.21) Q ~ rZ ~ rX~ rz g ~Z~Z ~X~X (7.30)
8(l + c')' E (11) E&(1j)
This proves that the higher-order corrections do
Thus the sign of the two-spin flip process depends on not destroy the duality of the two transformations.
whether the two spins are parallel or antiparallel.
The corresponding term in the Hamiltonian can be
written as UIII. DISCUSSION
z x x
~ ~lz ~I+1~I x P' 3'
(7.22)
~I+1 ~ ~l ~l+&
In this paper we compared two types of quantum
where to second order in h RG transformations. In the block transformation"
the low-lying levels of independent cells are mapped
onto new spin states, the coupling between the cells,
8(1+c')' Et(11) + Et(jj) '

a nearest-neighbor coupling between the two adjacent

Since in a consistent RG calculation this generated
new coupling has to be introduced from the outset,
its effect on the other couplings should also be con-
sidered. This new coupling contributes to the matrix
elements between the states
" I & x ( 1)11)x ( 1i) li)
and
end spins, is treated in a perturbation way, In the de-
cimation transformation' the spin states of selected
sites are mapped onto the states of new spins, by tak-
ing the lowest-energy configuration for the intermedi-
ate spins. The single site term of the Hamiltonian
acting on the selected sites is used now as a perturba-
" tion.
%e have shown that the two RG transformations
lead to equivalent results when applied to the one-

" II)xi(1'j)lj»i(jj)lj& " dimensional quantum versions of the Ising model,

Potts model or general Z(4) model. We have seen
24 DUALITY OF THE BLOCK TRANSFORMATION
that the results of the decimation transformation are
identical to those obtained by the block transforma-
tion in the dual model. This is true even if several
couplings are introduced, as in the Potts model or
Z(4) model calculation, and also in higher orders of
perturbation theory, where further new couplings are
generated.
It was apparent in the first applications of the
quantum RG transformation to the quantum Ising
model that the transformations do not conserve the
self-duality of the model. By treating the Ising cou-
pling and the transverse field on equal footing,
Fernadez-Pacheco" was able to find an RG transfor-
mation which conserves self-duality and therefore
gives the critical coupling exactly.
Except for the critical exponent v, the other ex-
ponents are not given exactly. When applied to the
Potts model, ' the critical coupling is again obtained
exactly, the critical exponents, however, are not exact
and become worse as the number of components in-
Permanent address: Central Research Institute for Physics,
P.O, Box 49, H-1525 Budapest, Hungary.
'See, e.g. , K. . G. Wilson and J. Kogut, Phys. Rep. C 12, 75
(1974); G. Toulouse and P. Pfeuty, Introduction au groupe
de renormalisation et a ses applications (Presses Universi-
taires de Grenoble, Grenoble, 1975); S.-K. Ma, Modern
Theory of Critical Phenomena (Benjamin, Reading, 1976).
2E, Brezin, J. C. LeGuillou, and J. Zinn-Justin, Phys. Rev.
D 15, 1544 (1977).
J. C. LeGuillou and J. Zinn-Justin, Phys. Rev. Lett. 39, 95
(1977).
4Th. Niemeijer and J. M. J. van Leeuwen, Phys, Rev. Lett.
31, 1411 (1973); Physica 71, 17 (1974).
5L. P. Kadanoff and A. Houghton, Phys. Rev. B 11, 377
(1975) ~ '9R, B. Potts, Proc. Cambridge Philos. Soc. 48, 106 (1952),
K. G. Wilson, Rev. Mod. Phys, 47, 773 (1975).
7S. Jafarey, R. Pearson, D. J. Scalapino, and B. Stoeckly
(unpublished) . H. A. Kramers and G. H. Wannier, Phys. Rev. 60, 252
S. D. Drell, M. Weinstein, and S. Yankielowicz, Phys, Rev.
D 16, 1769 (1977); S. D. Drell, B. Svetitsky, and M.
Weinstein, ibid. 17, 523 (1978).
9R, Jullien, J. N. Fields, and S. Doniach, Phys. Rev. Lett.
AND. . . 243
creases. There is no indication of the crossover from
second-order to first-order transition around q =4.
The other RG transformations, the usual block
transformation and its dual, the decimation transfor-
mation have the merit, that the second-order to
first-order crossover is reproduced. ' Due to the gen-
eration of new couplings, these transformations can
give a more realistic description of the behavior of
the Potts model.
We have not looked in this paper at the solutions
of the recursion relations for the Z(4) model. Our
aim was just to establish the duality of the two
transformations. We will return to the solution of
the equations in a subsequent publication.
ACKNOWLEDGMENT
I am grateful to Dr. E. Fradkin for useful discus-
sions.
38, 1500 (1977); Phys. Rev. B 16, 4889 (1977).
'OZ, Friedman, Phys. Rev. Lett. 36, 1326 (1976).
"K. Subbarao, Phys. Rev. Lett. 37, 1712 {1976).
G. Um, Phys. Rev. B 15, 2736 (1977); 17, 3670 (1978).
' R. Jullien, P, Pfeuty, J. N. Fields, and S. Doniach, Phys.
Rev. B 18, 3568 {1978);R. Jullien and P. Pfeuty, ibid, 19,
4646 (1979); K. Penson, R. Jullien, and P. Pfeuty, ibid.
19, 4653 (1979),
' J. E. Hirsch and G. F. Mazenko, Phys. Rev. B 19, 2656
(1979); J. E. Hirsch, ibid. 20, 3907 (1979).
' J. N. Fields, Phys. Rev. B 19, 2637 (1979).
' E. Fradkin and S. Raby, Phys. Rev. D 20, 2566 (1979),
M. Suzuki, Prog. Theor. Phys. 56, 1454 (1978).
P. Pfeuty, Ann. Phys. 57, 79 (1970).
J. Solyom and P, Pfeuty, Phys. Rev. B 24, 218 (1981)
(preceding paper).
(1941); G. H. Wannier, Rev. Mod, Phys. 17, 50 (1945),
J. Ashkin and E. Teller, Phys. Rev. 64, 178 (1943).
A. Fernandez-Pacheco, Phys. Rev. D 19, 3173 (1979).