VOLUME 40, NUMBER 25 PHYSICAL REVIEW LETTERS 19 JuNE 1978
where
Q, = -y(5x'+ 1), Q, = 2x(3x'+ 8y'+ 1), Q, = -4y(11x'+ 2y'+ 1), g, = 8x(5x'+ 12y'+ 1) . (12)
Higher k's repeat in this case also. All P ~ 's, k -4, can be reduced to multiples of P& ), P&', P&'), P&').
The TS solution previously known corresponds to the restrictions P&') = —12P&'), P&') = P&" = 0. The
complete solution we will now be able to generate from 5= 2 will contain five arbitrary parameters in-
cluding the mass. %hen the NUT parameter is excluded, we will have a four-parameter asymptotically
flat metric, As an example, we give the solution corresponding to P&') = —4P&'), P&') = P&3) =0:
4ix(x' —1) —4Py (x' —y')
(x + 1)'(x' —1) —2i Py (x + 1) (x' —2x + y') —P'(x' — y')' (13)
The complete 5= 2 solution plus further details will be published elsewhere. ' This work was sup-
ported by the National Science Foundation under Grant No. PHYV6-12246.
W. Kinnersley, J. Math. Phys. (N. Y.) 18, 1529 (1977).
W. Kinnersley and D. M. Chitre, J. Math. Phys. (N. Y.) 18, 1588 (1977).
W. Kinnersley and D. M. Chitre, to be published.
D. M. Chitre, to be published.
5W. Kinnersley and D. M. Chitre, to be published.
Yang-Lee Edge Singularity and p3 Field Theory
Michael E. Fisher
Baker Laboratory, Cornell University, Ithaca, Net York 14853
(Received 20 April 1978)
The edge of the gap in the distribution of Yang-Lee zeros at B=iHO(T) on the imaginary
magnetic field axis in ferromagnets above Y', is essentially a critical point. In terms of
-
the edge exponents 6 and q, the density of zeros obeys 8 &H") [H" —Ho(T)], with o.
=1/6=(d —2+@)/(d2 —q). Classical behavior (a=~) occurs for d&d =6. The appropriate
field-theoretic renormalization. group entails a my coupling and, with e = 6 —d —0, yields
q = —e/9 for all n & ~. This correlates well with refined series estimates for d =2 and
d=8 and with exact results for d= 1 (q= —1),
Consider the magnetization, M(H, T), of a fer- netization as a function of real field exhibits a
romagnet at fixed temperature T. According to first-order transition with a jump 2M, (T) o"- g.(0;
'
Yang and Lee, the analytic behavior of M (FI, T) T).
as a function of the magnetic field H can be under- On the other hand, for T &T„ there is agap of
stood by studying the asymptotic distribution of width 2H, (T) in the distribution of zeros, and
the zeros of the partition function in the complex M(FI, T) is analytic for )Im(H)~ &H, (T). The edges
magnetic-field plane (H', H") = (Re[H], Im[H]). of this gap, at H = + iH, (T), must be branch points
Very generally, the distribution of zeros in the of the function M(H, T); Kortman and Griffiths'
thermodynamic limit is expected to concentrate have pointed out the interest in determining the
on curvilinear loci in the complex field plane; in- nature of these branch points, which we term the
deed, for a variety of models of a ferromagnet, ' ~ Yang-I ee edge singularities. Since these are the
including Ising models' and spherical models, it signularities closest to the real axis, they play a
is known rigorously that the zeros concentrate on- dominant role in determining the observable be-
ly on the imaginary axis, H =i&". In that case havior of M for realII and T. Indeed they should
the magnetization for all real and complex H can enter into the asymptotic equation of state near
be expressed as an integral over g(II"; T), the the critical point, although it transpires that none
asymptotic density of zeros. Below the critical of the equations proposed for d &4 in the current
temperature T„one has g(0;T) &0 and the mag- literature contain the correct singularities t
1610 1978 The American Physical Society
VOLUME 40, NUMBER 25 PHYSICAL REVIEW LETTERS 19 JUNz 1978
More concretely, suppose the density of zeros
varies as
8(H )-(H" H-(T)) 2 2
where II -Ho+; then the magnetization exhibits
a branch point of the form ~ -h' with
m =M -M(iH„T) and 0 =H iH,-(T). (2) rZ
The value of the exponent cr, its universality, its 0 0
dependence on the dimensionality, d, and the sym-
metry number, n, of the ferromagnet, and its /I
/
relationship, if any, to critical exponents describ- /
/
ing the real, directly observable singularities,
/
are the topics of this Letter. /
It will be argued that the edge singularities are 2
closely analogous to ordinary critical points and
that corresponding scaling laws and exponent re-
lations, for example,
—
0= 1 d 2+/
d+2 —g'
are applicable. This can be checked explicitly
for the@ =1 Ising model where o = —2, for all T Z S 4 6
&T, =O; however, a consideration of mean-field
or Landau theory indicates that the appropriate FIG. 1. Variation of the Yang-Lee edge exponents 0.
critical point is that associated with a y', rather and g with dimensionality d. (i) The heavy black dots
than with the usual y4 theory. This fact leads to and solid lines correspond to exact and renormaliza-
tion-group results [see Eq. (4)]; (ii) the light dotted
a crossover dimensionality d" = 6, above which
the classical mean-field value o =+ —, ' applies. A and broken (dashed) curves represent the approxima-
tions q=0 and g= —-e, with e = 6-d, inserted in Eq.
f ield-theoretic renor malization-group treatment (3); (iii) the dashed-dotted curve results from the
is then possible and yields two-point Pade approximant g= —5e/(45- 4e), based
1 1 1 on Eq. (4) and q(l) = —1; (iv) the solid bars for d = 2
a d o (4) and d = 3 represent new numerical estimates derived
to first order in e = 6 -d () 0); in second order from high-temperature-series analysis for a range
of lattices; (v) the more widely spaced error limits
the correction factor for g is 1+ac, with a = &.
indicate the estimates of Kortman and Griffiths for
More generally the values of the exponents at the square and tetrahedral lattices (Ref. 5).
the edge singularity are seen to be independent of
the symmetry number, or number of components,
z, of the original order parameter, s, provided be expected on heuristic grounds. In addition, it
that n &~. The limit n- ~ corresponds to the may be checked explicitly for (i) the nearest-
spherical model where, in fact, o=+& applies for neighbor Ising chain, and, using matrix or inte-
all d. 4 (The nonuniformity of the n- ~ limit for gral kernel methods, ' for (ii) one-dimensional
d &6 is puzzling and not yet understood. ) The models of general n (&~); and, for alla, in
variation of o and g with d according to this anal- (iii) spherical models (n =~),~ and (iv) mean-
ysis is shown in Fig. 1. The solid bars for d =2 field models. ' Kortman and Griffiths' have also
and 3 correspond to new high-temperature-ser- checked the numerical constancy of o to within
ies-expansion estimates for Ising models which +0.05 for T not too close to T„ for the square
extend and refine previous results of Kortman and tetl ahedral Ising lattices. Finally, for all pg
and Griffiths' for the square (d = 2) and tetrahe- &~, universality follows within the renormaliza-
dral (d = 3) lattices (shown by []). The numerical tion group, at least for small &=6-d, from the
estimates evidently accord well with the analyti- relative stability of the fixed point describing the
cal expectations. edge singularity.
To develop the arguments, consider first the At the critical point in zero field, however, the
universality of o for T &T, : This is certainly to gap in the distribution of zeros vanished and one
1611
VOLUME 40, NUMBER 25 PHYSICAL REVIEW LETTERS 19 JUNE 1978
must expect v to take on a different value, say standard reduced Hamiltonian density'
a'. Indeed (2) then yields M -H' so that 0'= 1/
KR =HsR —2rsR 2 —~e(VsR 2
-usR (6)
5, where 5' denotes the standard critical ex- )
ponent for the critical isotherm (with 6'= 15 for in which, in contrast to the usual analysis, "the
d=2, n=1, and 6'= 3 for d&4, all n). That v'ev -
real parameter r = C (T T,) is to be held fixed
may likewise be checked in all the cases (i) to and Poslfi've.
(iv) above. For the classical, fluctuationless theory, set
The idea that the edge singularity should be re- e =0 and, to find a divergent susceptibility, put
garded merely as a critical point occurring at a sR ~s„+m, . A pure imaginary shift, m, = —,i(r/
complex or imaginary magnetic field is natural 3u)'I', eliminates the quadratic term in (6) and
if, following Kortman and Griffiths, ' one notes generates a cubic term -ms R' with w = 2iu(r/
that (2) implies that the susceptibility y= BM/BH 3u)"'. With m = {s,)-m, and H, (T) =u(r/3u)'I',
diverges as 1/h' '
when h =H —iH, —0, provided the equation of state is then seen to be h — = (8
that o (1 (which seems generally true). By ana- -iH, ) ~ 3@m . The residual fourth-order term,
lytic continuation of the pair correlation function, us, plays no role asymptotically.
' ash-0 Thus X di-
G(R;H, T), it follows that the correlation length verges likeh ' and so 0= —, '.
also diverges, say as $ -1/h "c, and that G(R) de-
cays slowly at the edge singularity, say with ex- field-theoretic calculation, "
These considerations demonstrate that, for a
in Fourier space,
ponent g. Standard heuristic scaling arguments' it suffices to consider
then suggest
X=hso 2f +eq )s
(r— sq
G(R;H, T) ={s, ~
s, ) -(s, )'
~g)(Rh c)/R
f
—K qf &e &ss qs l& &i, (7)
as h, I/R -0. Indeed for (i) the linear-chain Ising with z=-0 and e =
—1. However, one must expect zv
model, this is precisely confirmed with exponents
q = —1, v, = ~, and scaling function D(w) =Ae
s / point. "
to be purely imaginary at the appropriate fixed
Of course, additional fourth-order terms
se'. Likewise d-dependent, hyperscaling argu- are required for stabilization, but these turn out
ments' lead to the exponent relation (3) and to v, to be irrelevant for d) 6 and small ~. It is now
= 2/(d+2 -rf), both of which are verified by the straightforward to use perturbation theory in ze
values quoted. to construct a momentum-shell integration re-
Now it is known that hyperscaling relations like normalization group" with a spatial rescaling q
(3) generally fall at borderline or crossover di- ~ q'/b. A shift and spin rescaling, s„~m6
mensionality, d, and become inequalities [with + $ qt are required to maintain z' = r = 0 and e
the second "=" in (3) replaced by "("]for d &d".'
IQ
=e — = 1. (Note that one more condition is imposed
Furthermore, one may determine d" heuristical- than at a normal critical point. ) In zeroth order
ly' by inserting the classical or mean-field val- one finds c'ccb~"+ 1 so that
11 =O(te') which
ues and solving for d. Putting g= 0, on general leads to the recursion relations h'~b " ' "h and
'
grounds, and' o =-, (see also below) in (3) yields ge't=b@'se with e =6-d. This confirms d =6 as
=6. This is confirmed by renormalization- the borderline dimensionality.
group analysis. A first-order calculation, involving graphs with
To set up a field-theoretic renormalization one, two, and three third-order vertices, yields
group it is essential to understand the classical r1 ~ 6K,zo' and a fixed point h* = —,'K~A'Ev*, and re*
or Landau phenomenological theory and to recog- =i ~'e/(54K, )"' (where qA is the cutoff momentum
nize that the desired critical behavior, namely a and K, is the area of a unit sphere at d = 6). These
divergent susceptibility, must be sought at a, com- results imply g ~ —e/9 [see (4)]; the hyperscaling
plex value of the magnetic field. However, intro- relation (3) is checked directly by the eigenvalue
duction of any magnetic field, real or complex, y, = 1/v, ~4 —ve-= — ', (d+2 —q).
breaks the original O(n) symmetry, which means An alternative renormalization-group proce-
that the singular behavior (at least for n(~) dure imposes h' =h = 0 and allows r to vary in (7).
should be the same as for n=1 or Ising-like sys- This leads to completely equivalent physical re-
tems. Integral kernel methods enable one to sults although some surprisingly singular renor-
check this directly' when d = 1. Accordingly, we malization-group flows are encountered. It is
may take scalar spins, s„, and consider the also possible" to adapt Amit's dimensional regu-
1612
VOLUME 40, NUMBER 25 PHYSIGAI, REVIEW I, KTTERS 19 JvNs 1978
larization calculation" by relaxing his Potts- 'C. Yang and T. D. Lee, Phys. Rev. 87, 404
¹
model constraint [and putting o. , =P, =P, =Q», e0 (1952).
in his Eqs. (6. 1)-(6.3) et seq. ]. This confirms (4) ~T. D. Lee and C. N. Yang, Phys. Rev. 87, 410
(1952).
and yields the second-order term quoted.
Finally, as demonstrated by Kortman and Grif-
(¹Y.
3R. B. Qriffiths, J. Math. Phys. ) 10, 1559
(1969); T. Asano, Phys. Rev. Lett. 24, 1409 (1970);
fiths, ' the exponent o may be estimated numer- M. Suzuki and M. E. Fisher, J. Math. Phys. (N. Y.) 12,
ically for Ising models by studying the high-tem- 235 (1971); F. Dunlop and C. M. Newman, Commun.
perature series for X with purely imaginary field Math. Phys. 44, 223 (1975).
and varying temperature. However, if the tem- D. A. Kurtze and M. E. Fisher, to be published.
5P. J. Kortman and R. B. QriEiths, Phys. Rev. Lett.
perature independence of o is accepted, more ac-
27, 1439 (1971); see also P. J. Kortman, Ph. D. thesis,
curate results can be found by studying the high- Carnegie-Mellon University, 1971 (unpublished).
temperature limit of the series. If one sets z 6D. A. Kurtze and M. E. Fisher, to be published.
=tanh(J/ksT)[tanh(FI/ksT)]', where J' is the near- YSee, e.g. , M. E. Fisher, Rep. Prog. Phys. 30, 615
est-neighbor coupling, one finds" that the expan- (1967).
sions then reduce to those for the monomer-di- 8Note that standard demonstrations of the Bucking-
mer problem' on the same lattice at dimer activ- ham-Gunton inequality for 6 and q [J. D. Gunton and
ity z. It follows that the Yang-Lee zeros for the M. J. Buckirgham, Phys. Rev. Lett. 20, 143 (1967);
and M. E. Fisher, Phys, Rev. 180, 594 (1969)1 Sail
dimer problem must lie on the negative z axis"
because the positivity and monotonicity of the analyt-
and the dimer density, p(s), must exhibit a branch
ically continued correlation functions cannot be as-
point of the form p, +A(z -z, )'. Long series are sumed,
available for the monomer-dimer problem'+" on IAs observed by Q. Toulouse, Nuovo Cimento B 23,
various lattices. From ratio analysis' of expan- 234 (1974).
sion coefficients for the square (17 terms), tri- For notation, etc. , see, e.g. , M. E. Fisher, Rev.
angular (14 terms), tetrahedral (16 terms), sc Mod. Phys. 46, 597 (1974).
(15 terms), bcc (13 terms), and fcc (10 terms) "This is a major point of difference from other e ex-
pansions about d = 6 such as Q. Mack, in Lecture Notes
lattices the following estimates have been ob-
jn Physics, edited by W. Ruhl and A. Vancura (Springer-
tained: Verlag, Berlin, 1973), Vol. 11, p. 300; S.-k. Ma and
Q. F. Mazenko, Phys. Rev. Lett. 33, 1383 (1974);
a= —0. 155+0.010 for d =2, A. B. Harris et al. , Phys. Rev. Lett. 35, 327 (1975);
o = 0.098 + 0.012 for d = 3. R. G. Priest and T. C. Lubensky, Phys. Rev. B 13,
4159 (1976); D. J. Amit, J. Phys. A 9, 1441 (1976);
A. B. Harris et al. , Phys. Rev. Lett. 36, 415 (1976);
Within their quoted uncertainties of + 0.050, A. J. McKane, D. J. Wallace, and R. K. P. Zia, Phys.
these compare well with the Kortman-Griffiths Lett. 65B, 171 (1976). However, Dr. E, Brbzin (pri-
estimates' of 0=- ' '
—, (d=2), +-, (d=3); see Fig. l. vate communication) has pointed out that cp~ models
Independent estimates for the square and triangu- with imaginary coupling constant have also been con-
lar lattices differ by about 0.006; likewise, in sidered in connection with Reggeon field theories
three dimensions lattices of larger coordination [see, e.g. , H. D. I. Abarbanel et al. , Phys. Rep. 21C,
121 (1975)]; the results obtained, however, c~~ot be
number suggest values 0.006-0.012 higher. These used directly here since they involve an anisotropic
differences, however, are within the extrapola- propagator (I am indebted to Dr. E. Brezin for this
tion uncertainties; so there are no serious grounds remark).
for doubting that v is truly lattice independent. ' See E. Brbzin, Ref, 11.
Pade-approximant analyses of various logarith- '3This has been observed independently by Q. A.
mic derivative series' yield consistent results. Baker, Jr., and P. Moussa, J. Appl. Phys. 49, 1360
The support of the National Science Foundation, (1978).
' See, e.g. D. S. Gaunt, Phys. Rev. 179, 174 (1969).
in part through the Materials Science Center at ,
'5This is already known rigorously from O. J. Heil-
Cornell University, is gratefully acknowledged. mann and E. H. Lich, Phys. Rev. Lett. 24, 1412 (1970).
Dr. Helen Au-Yang and Mr. Douglas A. Kurtze ~6I am grateful to Dr. M. F. Sykes of King's
College,
kindly provided numerical assistance. London, for sending me his recent data.
1613
