
2
larity of the density of states in 1D harmonic chains at
zero energy (chiral symmetry point) [9]. Further works
extended the analysis to localization properties and to
quasi-1D systems. It was found that an N-channel quasi-
1D system of chiral class has Ntopological phases. At
transitions between these phases, the density of states ex-
hibits the Dyson singularity [10, 11], and the localization
length diverges [12–20]. Critical points of these tran-
sitions have infinite-randomness character, with critical
wave functions showing very strong fluctuations [15, 21].
For 2D chiral-class systems, most of the past research
focussed on properties of the metallic phase. The Gade-
Wegner sigma model was re-derived and analyzed in
many works [22–25]. A particular attention was paid
to the asymptotic infrared behavior, with is of infinite-
randomness character, exhibiting a very strong diver-
gence of the density of states and a “freezing” of the mul-
tifractality spectrum [6, 26–28]. On the numerical side,
most papers showed critical properties of the metallic
phase that are characterized by non-universal exponents
for various observables (such as multifractality, density
of states, localization length at finite energy) [29–34]
and are essentially different from those expected in the
infinite-randomness infrared limit. This is not surprising:
the Gade-Wegner flow towards the line of infrared fixed
points is logarithmically slow, so that in a typical situ-
ation the infrared limiting behavior can likely be out of
reach on any realistic length scale. In several works [35–
37], evidence of the asymptotic behavior of the lowest
Lyapunov exponents in the quasi-1D geometry has been
reported.
Apart from realizations in disordered electronic sys-
tems, the interest to models in the chiral classes is due
to their relation to models of Dirac fermions coupled to
fluctuating gauge fileds that are discussed in the con-
text of quantum chromodynamics (QCD) [38]. It was
proposed that ATs in such models may be connected to
QCD phase transitions; see Ref. [39] for a recent review.
It is also worth mentioning that chiral-class models can
be experimentally realized in microwave setups based on
coupled resonators [40]. Recently, MITs in 3D chiral-class
systems were studied in Refs. [41, 42].
Field theory of 2D chiral AT. In the fermionic replica
formalism, the sigma-model manifolds for classes AIII,
BDI, and CII are U(n), U(2n)/Sp(2n), and U(n)/O(n),
respectively. In the analytical and numerical analysis be-
low, we focus on the class AIII. The Gade-Wegner sigma-
model action has the form [4, 5]
S[Q] = −Zd2rhσ
8πTr(U−1∇U)2+κ
8π(TrU−1∇U)2i.
Here U(r)∈U(n)(with the replica limit n→0to be
taken at the end of the calculation), σis the conductivity
in units of e2/πh; the second term (known as “the Gade
term”) couples only to the U(1) degree of freedom and is
specific for chiral classes. To describe the transition, one
has to include also vortices, with a fugacity y[8]. The
RG equations for three couplings σ,κ, and yread
∂K/∂ ln L= 1/4−2Ky2,(1)
∂y/∂ ln L= (2 −K)y , (2)
∂σ/∂ ln L=−σy2,(3)
where K= (σ+κ)/4. Equations (1) and (2) form a
closed system, with a fixed point at K= 2 and y=1
4.
In the three-dimensional parameter space (σ,κ,y), this
corresponds to a critical line of MITs, σ+κ= 8,y=1
4.
Along this line, there is a flow according to Eq. (3) to-
wards the ultimate fixed point σ= 0,κ= 8, and y=1
4.
This flow is, however, very slow: σ(L)=σ0L−1/16. There-
fore, while in the strict infrared limit the transition is
described by the ultimate fixed point, on realistic scales
one expects to see a transition described by some point
on the critical line. This is expected to lead to an ap-
parent non-universality of some of critical properties, as
discussed below.
The RG flow that follows from Eqs. (1)–(3) is illus-
trated in Fig. 1. The overall flow is three-dimensional
and is thus difficult to display. What is shown is the pro-
jection of the flow on the σ–κplane, with all RG trajec-
tories having an initial value of the fugacity y0=1
4. The
fixed points of the flow are as follows. First, there is an
infrared-stable line of fixed points describing the metal-
lic phase, with σbeing an arbitrary constant, κ→ ∞,
y→0. Second, there is an infrared-stable fixed point
describing the insulating phase: σ, κ →0and y→ ∞.
Finally, there is a fixed point σ= 0,κ= 8,y=1
4, de-
scribing the MIT. It has one unstable direction, so that
there is a two-dimensional critical surface with a flow to-
wards this point. A cross-section of this surface with the
plane y=1
4is the critical line σ+κ= 8 shown in Fig. 1.
Linearizing the RG equations (1) and (2) near the
transition point, we get the critical exponent of the lo-
calization length ν= 1.54 and the irrelevant exponent
yirr = 0.77. In addition, there is a very slow flow towards
the fixed point along the critical line described by Eq. (3);
it yields an exponent y0
irr = 1/16 '0.06. The fact that
the ultimate fixed point of the transition is at σ= 0
implies very strong fluctuations of critical eigenfunctions
in the infrared limit (with freezing of the multifractal
spectrum). This is expected on physical grounds: we
know that eigenstates in the metallic phase possess this
property, and it would be surprising if eigenstates at the
transition would be “less localized” than in the metal.
Let us reiterate that the RG equations are only con-
trollable at y1. Since the fixed point of the tran-
sition is at y=1
4, that is not parametrically small, all
quantitative conclusions about the transition should be
taken with caution. A plausible assumption is that the
obtained flow is qualitatively correct but numbers de-
scribing the transition may differ substantially. It is thus
crucially important to explore the transition numerically,