UCSD/PTH 95-20
The Shape of the Renormalized Trajectory
in the
Two-dimensional O(N) Non-linear Sigma Model
Wolfgang Bock⋄ and Julius Kuti∗
Department of Physics 0319
University of California at San Diego
9500 Gilman Drive, La Jolla, CA 92093-0319
Abstract. The renormalized trajectory in the multi-dimensional coupling parameter space
of the two-dimensional O(3) non-linear sigma model is determined numerically under
δ-function block spin transformations using two different Monte Carlo renormalization group
techniques. The renormalized trajectory is compared with the straight line of the fixed point
trajectory (fixed point action) which leaves the asymptotically free ultraviolet fixed point
of the critical surface in the orthogonal direction. Our results show that the renormalized
trajectory breaks away from the fixed point trajectory in a range of the correlation length
around ξ ≈ 3-7, flowing into the high temperature fixed point at ξ = 0. The analytic large N
calculation of the renormalized trajectory is also presented in the coupling parameter space
of the most general bilinear Hamiltonians. The renormalized trajectory in the large N approximation exhibits a similar shape as in the N = 3 case, with the sharp break occurring at
a smaller correlation length of ξ ≈ 2-3.
Date: October 15, 1995
⋄ bock@rccp.tsukuba.ac.jp,
∗ kuti@axpjk.ucsd.edu
address after Oct. 1, 1995: Dept. of Physics, Univ. of Tsukuba, 305 Ibaraki
2
1. Introduction
One of the most important issues in lattice field theory is the removal of cutoff effects in
physical quantities. Two different approaches have recently received considerable attention,
with the promise of efficient reductions in cutoff contaminations. The first approach, which
combines the original Symanzik improvement program [1, 2] with tadpole improvement in
the lattice coupling constant [3], has been successfully tested in the spectroscopy of heavy
quark-antiquark bound states [4]. The second approach [5], which attempts to approximate
the renormalized trajectory (RT) within the context of Wilson’s renormalization group program [6], shows considerable promises for lattice QCD applications [7, 8]. In this work, we
determine the RT in the two-dimensional O(3) non-linear sigma model and compare it with the
fixed point approximation, as proposed in the second approach [5]. The precise connection
between Symanzik’s improvement program and a truncated approximation to the asymptotically free fixed point in Wilson’s renormalization group approach remains an interesting and
yet unresolved issue.
It has been known for a long time that, in contrast with the standard lattice action, each
point on the RT defines a perfect lattice action [5] which is free of cutoff effects at any finite
correlation length [6]. A sketch of the RT in the phase diagram of asymptotically free field
theories is shown in Fig. 1 as plotted in the infinite dimensional space of couplings Ki . The
inverse coupling 1/K1 of the standard lattice action is singled out to label the horizontal axis.
The critical manifold is given in Fig. 1 by the K1 = ∞ plane and the asymptotically free fixed
point on the critical surface is designated by UVFP. The continuum theory at finite lattice correlation lengths is defined by the RT which flows along the unstable direction from the UVFP
to the high temperature fixed point (HTFP) with vanishing correlation length at Ki = 0 [6]. The
determination of the perfect lattice action has not been considered feasible in the past, mainly
because a practical approximation to the RT in the infinite dimensional coupling parameter
space was thought to require the uncontrolled truncation of a large number of terms in the
blocked Hamiltonian.
Considerable progress has been made recently by Hasenfratz and Niedermayer [5] who realized that the fixed point lattice action of the UVFP can be determined from a classical saddle
point problem in asymptotically free field theories. The fixed point trajectory (FPT), which is
defined as the straight line that originates from the UVFP and is perpendicular to the critical
surface, has been suggested to be a good approximation to the RT at sufficiently large correlation lengths [5] (dashed line for K1 < ∞ in Fig. 1.) It was also demonstrated that the fixed
point action (FPA) can be rendered very short-ranged by optimizing the block-spin renormalization group transformations with the expectation that an approximate FPT which is based
on a truncated FPA, with the very small long range couplings neglected, will exhibit almost no
cut-off dependence at large and moderate correlation lengths.
The performance of the approximate FPT has first been tested by Hasenfratz and Niedermayer in the 2d non-linear O(3) sigma model which is known to be asymptotically free [5].
They carried out a pilot study using a truncated FPT with 24 different couplings. Several tests
showed that the residual cutoff effects were not visible even down to a correlation length of
three suggesting that truncation effects are negligible. This indirectly implies that the FPT
3
K3 /K 1 ....
FPT
UVFP
RT
K2 /K 1
HTFP
0
1/K 1
0
1/ ξ
Figure 1 . Phase diagram of asymptotically free field theories. The inverse coupling 1/K1 of the standard action, or equivalently the inverse correlation length, labels the horizontal axis, and the ratios Ki /K1 label the remaining (infinitely many)
axes. The abbreviations RT, FPT, UVFP, and HTFP stand for renormalized trajectory, fixed point trajectory, ultraviolet fixed point, and high temperature fixed point,
respectively.
runs close to the RT in an extended range of lattice correlation lengths.
For a better understanding of the FPT approximation, it is important to determine the crossover region in the correlation length where the RT eventually has to break away from the FPT
to flow into the HTFP as depicted in Fig. 1. In this work, we determine numerically the position of the RT in a finite dimensional subspace of the infinite dimensional coupling parameter
space and compare it with the FPT. A more detailed account of our investigation will appear
elsewhere [9].
2. The non-linear O(3) sigma model and the fixed point action
The path integral of the O(N) non-linear sigma model in the continuum is defined by
Z
(2.1)
Z =
D[φ(x)] exp [−βHcont ] ,
Z
1
d2 x∂µ φ(x)∂µ φ(x) ,
(2.2)
Hcont =
2
where β = 1/T is the inverse temperature, φ(x) denotes the N-component scalar field of unit
R
length, and D[φ(x)] is the O(N) invariant measure. If we integrate out the momenta between
the cutoff Λ and Λ/τ, we find the two-loop renormalization group result
dT
N −2 2 N −2 3
T + O(T 4 ) ,
=
T +
d ln τ
2π
(2π )2
(2.3)
4
which implies that the model is asymptotically free for N ≥ 3 and the UVFP is located at T = 0
(β = ∞.)
The lattice Hamiltonian H is not unique. The only constraints come from the symmetries
of the system and the fact that it ought to reduce in the classical continuum limit to the Hamiltonian Hcont of Eq. (2.2). With these constraints, the lattice Hamiltonian can be parametrized
as
H = H2 + H4 + H6 + . . . ,
1 XX
H2 = −
ρ(r ) (1 − φx φx+r ) ,
2 r x
X
H4 =
c(x, y, z, w) (1 − φx φy ) (1 − φz φw ),
(2.4)
(2.5)
(2.6)
x,y,z,w
where H2 includes all terms that are bilinear in the field variable φx , and H4 , . . . denote all
the other terms which are quartic or higher order in the field variables. The summations in
Eqs. (2.5) and (2.6) are over two-dimensional lattice vectors r , x, y, z, and w. The standard
lattice action is given by the first two terms in Eq. (2.5) with r1 = (1, 0), (0, 1), and K1 =
βρ(r1 ). Physical quantities are measured in units of the lattice spacing a which is set to one
for notational convenience.
Let us now consider the following δ-function block spin transformation,
Z
′
′
′ ′
exp −β H (φ ; ρ , c , . . . ) =
Dφ P (φ′ , φ) exp [−βH (φ; ρ, c, . . . )] ,
!
P
Y
′ φx
δ φ′x ′ − Px∈x
P (φ′ , φ) =
,
k x∈x ′ φx k
x′
(2.7)
(2.8)
where a blocked lattice site x ′ is assigned to a 2 × 2 cell of sites x on the unblocked lattice.
We put primes in Eqs. (2.7) and (2.8) to label the quantities on the blocked lattice. For details
on how the FPA can be computed in asymptotically free field theories we refer the reader to
Ref. [5]. The authors of Ref. [5] showed that the fixed point couplings ρ ∗ (r ) agree with the
ones obtained earlier for the non-interacting model with Gaussian block spin transformation
[10],
Z +π
d2 p
∗
ρ (r ) =
ρ̃ ∗ (p) eipr ,
(2.9)
2
−π (2π )
−1
2
+∞
Y
X
sin2 (pi /2)
1
.
(2.10)
ρ̃ ∗ (p) =
2
2
2
n1 ,n2 =−∞ (p1 + 2π n1 ) + (p2 + 2π n2 ) i=1 (pi /2 + ni π )
We have determined the couplings for several lattice vectors r by evaluating Eqs. (2.9) and
(2.10) numerically. The values of the fourteen largest couplings are given in Table 1. The interactions are very short-ranged, since the couplings decrease rapidly when the distance |r |
between two interacting spins grows. The couplings of the type ρ(r ), r = (n, 0) fall exponentially with n, |ρ((n, 0))| ∼ exp[−1.45n] [5]. The authors of [5] also demonstrated that the
couplings can be rendered even more short-ranged by using a soft block spin transformation.
Quartic and higher order fixed point couplings can be also determined from their saddle point
equation.
5
r
r1
r2
r3
r4
r5
r6
r7
= (1, 0)
= (2, 0)
= (1, 1)
= (3, 0)
= (4, 0)
= (3, 1)
= (2, 2)
ρ ∗ (r )
−3.42839
+0.74981
+0.32486
−0.16871
+0.03877
−0.01976
−0.01658
r
r8
r9
r10
r11
r12
r13
r14
= (2, 1)
= (5, 0)
= (4, 1)
= (3, 2)
= (5, 1)
= (6, 0)
= (4, 2)
ρ ∗ (r )
+0.01060
−0.00909
+0.00828
+0.00502
−0.00274
+0.00217
−0.00100
Table 1 . The fourteen largest couplings ρ ∗ (r ).
3. The renormalized trajectory in the large N limit
In the large N limit, Hirsch and Shenker derived the following recursion relation for the blocked
two-point function in momentum space [11],
P
G(p ′ /2 + π l) s 2 (p ′ /2 + π l)
G′ (p ′ ) = 4 Pl
,
(3.1)
′
2
′
q′ ,l G(q /2 + π l) s (q /2 + π l)
L2
s 2 (q′ ) =
2
Y
µ=1
sin2 qµ′
sin2 (qµ′ /2)
,
(3.2)
where l is a vector with components equal to 0 or 1, and L2 designates the number of points
on the lattice before blocking. The momentum space propagator
G(p) =
T
ρ̃(p) + m2
(3.3)
is the saddle point solution of the large N expansion (in the large N limit, β remains finite
when N is factored out from the bilinear Hamiltonian.) For a given β and ρ(p), the mass gap
m is determined from the gap equation
1 X
G(p) = 1 .
(3.4)
L2 p
Eq. (3.1) can now be iterated to determine the blocked propagators after repeated block-spin
transformations. In the subspace of bilinear Hamiltonians, we can obtain the couplings βρ(r ),
β′ ρ ′ (r ′ ), etc. from the inverse propagators by Fourier transformation, i.e., for r , r ′ ≠ 0,
4 X
1 X 1
1
′ ′
eipr ,
β′ ρ ′ (r ′ ) = 2
eip r , . . .
(3.5)
βρ(r ) = 2
′
′
L p G(p)
L p′ G (p )
We have iterated Eq. (3.1) numerically, using two different expressions for the unblocked
propagator. The first choice originates from the bilinear Hamiltonian that includes the four
largest couplings in Table 1,
X
X
2(1 − cos(2pµ ))
2(1 − cos pµ ) + α2
ρ̃(p) = α1
µ
+
µ
X
2(1 − cos(3pµ )) , (3.6)
α3 2(1 − cos(p1 + p2 )) + 2(1 − cos(p1 − p2 )) + α4
µ
6
Figure 2 . Flow lines of ρ(r2 )/ρ(r1 ) (a) and ρ(r3 )/ρ(r1 ) (b) in the large N limit.
The flow lines represented by the crosses start from an initial propagator that is
given by Eq. (3.6). The lines represented by squares, full, and open circles start from
the fixed point propagator (cf. Eq. (2.10)). The fixed point ratios ρ ∗ (r2 )/ρ ∗ (r1 ) and
ρ ∗ (r3 )/ρ ∗ (r1 ) are marked by horizontal dashed lines.
with coupling parameters α1 , α2 , α3 and α4 . The second choice is the fixed point propagator resulting from Eq. (2.10). In the first case, we can start the flow lines, which after several
blocking steps will approach the RT, at different positions in the four-dimensional coupling
parameter space by varying the five coupling parameters α1 , . . . , α4 and β. In the second case,
the flow lines start on the FPT and it is interesting to see how the blocked couplings break away
from the FPT after a few block spin transformations while tracing the exact RT. The correlap
tion length ξ on the unblocked lattice is given by ξ = (α1 + 4α2 + 2α3 + 9α4 )/m2 for the
p
first choice, and by ξ = 1/m2 for the second choice (using the fact that ρ̃(p) in Eq. (2.10) behaves as ρ̃(p) ≈ p 2 for small p 2 values.) The parameter m2 has been determined numerically
from the gap equation (3.4).
As an example, we have displayed in Fig. 2 the flow lines of the ratios ρ(r2 )/ρ(r1 ) and
ρ(r3 )/ρ(r1 ) as a function of ln(1/ξ). Since the correlation length is reduced by a factor of
two after every block spin transformation, two consecutive points on a renormalization flow
line are always separated by an amount of ln 2 on the horizontal axis. The flow patterns of
three-link and other couplings look very similar and we refer the reader for more details to
Ref. [9]. The UVFP is located in this plot at ln(1/ξ) = −∞ and the HTFP at ln(1/ξ) = +∞.
The ratios ρ ∗ (r2 )/ρ ∗ (r1 ) and ρ ∗ (r3 )/ρ ∗ (r1 ) of the fixed point couplings are represented by
horizontal dashed lines. The flow lines which start from the four-coupling Hamiltonian are
represented by crosses. Fig. 2 shows that those flow lines that start at large correlation lengths
are attracted to the FPT within a few block spin steps indicating that the FPT is indeed a very
good approximation to the RT at large correlation lengths. At small correlation lengths, however, the flow lines approach a curve (RT) which is substantially different from the FPT. The RT
will eventually flow into the HTFP where the two ratios ρ(r2 )/ρ(r1 ) and ρ(r3 )/ρ(r1 ) vanish, as
can be shown by the high temperature expansion [9]. Fig. 2 shows that the sharp break from
7
the FPT occurs at a correlation length of ξ ≈ 2-3. The three flow lines that start on the FPT at
three different values of β are represented by squares, full, and open circles. These flow lines
trace the RT quite accurately, since they start at a correlation length larger than three.
4. Numerical simulation of the O(3) model
The renormalization group flows in the previous section were restricted to the subspace of bilinear Hamiltonians in the large N limit. In contrast, the Monte Carlo renormalization group
method will allow us to study the exact flows of selected couplings in the O(3) model. Although interaction terms which do not fit on the finite lattice are truncated, their effects are
expected to be negligible. The only practical limitation of the method is the signal to noise
ratio in measuring very small blocked couplings.
The 1-cluster algorithm [12] has been used to generate unblocked spin configurations on
a 256 × 256 lattice using a Hamiltonian that includes the first four couplings of Table 1. The
spin configurations have been blocked four times down to a lattice of size 16 × 16 using the
block spin transformation of Eq. (2.7). We have implemented two different methods to infer
the couplings from the blocked spin configurations, a microcanonical demon method [13], and
a canonical one [14]. The microcanonical method has been used earlier to determine blocked
couplings in the two-dimensional O(3) sigma model [15, 16], but for a block spin transformation that differs somewhat from the one we are using in this paper. The microcanonical
method performs very well from a computational viewpoint, with the disadvantage that the
blocked couplings have a finite volume bias [13]. In contrast, the canonical method is exact, but suffers from larger autocorrelation effects which make the method very cumbersome
from a computational point of view. For more details about the simulation and a comparison
of the two techniques, we refer the reader to Ref. [9].
To illustrate our O(3) results, we have displayed in Fig. 3 the renormalization group flows
of the ratios ρ(r2 )/ρ(r1 ) and ρ(r3 )/ρ(r1 ) as a function of ln(1/ξ). The correlation length on
the unblocked lattice has been determined from the propagator in coordinate space by using
the improved estimator technique of ref. [17]. Most of the renormalization group flow lines
exhibited in Fig. 3 have been obtained with the microcanonical method (crosses.) We have also
included renormalization group flow lines (circles) which were generated by the canonical demon method. For one of the canonical flow lines, the initial couplings were chosen to coincide
with the starting point of a microcanonical flow line. Fig. 3 shows that the two flow lines with
the same initial couplings coincide within the error bars, and the flow pattern is very similar
to the one seen in Fig. 2. The only difference is that the sharp break from the FPT occurs now
at a somewhat larger correlation length. The grey lines in Fig. 3 mark the approximate position of the RT, as determined from the flow lines. For comparison, we have also included the
RT in the large N limit (heavy lines) (cf. Fig. 2.)
8
Figure 3 . Flow lines of ρ(r2 )/ρ(r1 ) and ρ(r3 )/ρ(r1 ) for N = 3. All flow lines start
from the four-coupling Hamiltonian. The lines which are represented by crosses
(circles) have been generated with the microcanonical (canonical) method, respectively. The grey lines indicate the approximate position of the RT, while the solid
lines depict the projections of the large N RT onto the two planes. The fixed point
ratios ρ ∗ (r2 )/ρ ∗ (r1 ) and ρ ∗ (r3 )/ρ ∗ (r1 ) are marked by horizontal dashed lines. Error bars are omitted when they are smaller than the symbol sizes.
5. Conclusion
We have demonstrated, by explicitly computing the RT in the O(3) model, that the FPT provides a good approximation to the RT in the region of the coupling parameter space where the
correlation length is large. A significant break occurs, however, in a range of the correlation
length around ξ ≈ 3-7 where the RT sharply departs from the FPT and flows into the HTFP.
The RT in the large N limit exhibits a similar behavior with the sharp break shifted to somewhat smaller correlation length. Although we have only shown the projection of the exact RT
in the K1 -K2 and K1 -K3 planes, several other projections of the flows were determined as well
[9].
Both ends of the RT can be determined without the utilization of the Monte Carlo renormalization group. The RT in the large correlation length regime, near the UVFP, is well approximated by the FPT, while the RT in the regime of small correlation lengths, near the HTFP,
can be obtained from a high temperature expansion. It is expected that both approximations
will break down in the crossover region where the Monte Carlo renormalization group technique may remain the only useful tool. It is not clear whether this regime will be important
in practical applications.
The studies we presented here will be useful to extend to non-Abelian lattice gauge theories
in four dimensions [7].
Note added. After the completion of this work, M. Okawa pointed out some earlier results [18]
where a similar large N approximation was applied to block spin transformations that differ
somewhat from the ones used in our paper.
9
Acknowledgements
This work was supported by the DOE under grant DE-FG03-91ER40546. One of us (W. B.) would
like to acknowledge discussions with W. Bietenholz and U. Wiese.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
K. Symanzik, Nucl. Phys. B226 (1983) 187; B226 (1983) 205.
M. Lüscher and P. Weisz, Comm. Math. Phys. 97 (1985) 59; Phys. Lett. B158 (1985) 250.
G.P. Lepage and P.B. Mackenzie, Phys. Rev. Lett. 61 (1988) 26635.
M. Alford, W. Dimm, G.P. Lepage, G. Hockney, and P.B. Mackenzie, FERMILAB-PUB-95-199-T (heplat/9507010).
P. Hasenfratz and F. Niedermayer, Nucl. Phys. B414 (1994) 785.
K. Wilson and J. Kogut, Phys. Rep. C12 (1974) 75.
T. DeGrand, A. Hasenfratz, P. Hasenfratz, and F. Niedermayer, BUTP-95-14 (hep-lat/9506030), BUTP-95-15
(hep-lat/9506031), BUTP-95-30 (hep-lat/9508024).
W. Bietenholz and U.-J. Wiese, MIT-CTP-2475 (hep-lat/9510026).
W. Bock and J. Kuti, in preparation.
T.L. Bell and K.G. Wilson, Phys. Rev. D11 (1975) 3431.
J.E. Hirsch and S.H. Shenker, Phys. Rev. B27 (1983) 1736.
U. Wolff, Phys. Rev. Lett. 62 (1989) 361; Nucl. Phys. B334 (1990) 581.
M. Creutz, Phys. Rev. Lett. 50 (1983) 1411.
M. Hasenbusch, K. Pinn, and C. Wieczerkowski, Phys. Lett. B338 (1994) 308.
M. Creutz, A. Gocksch, M. Ogilvie, and M. Okawa, Phys. Rev. Lett. 53 (1984) 875.
M. Okawa, Phys. Rev. Lett. 54 (1985) 963.
F. Niedermayer, Phys. Lett. B237 (1990) 473.
A. Guha, M. Okawa, and J.B. Zuber, Nucl. Phys. B240 [FS12] (1984) 566.