For many statistical-mechanics models,
by using model-tailored mappings to graph-theoretical problems and
applying sophisticated problem-dependent optimization algorithms 1, one
can obtain results which allow to understand the behavior numerically
much better in comparison to applying standard approaches like Monte
Carlo simulations.
Here, as an example for such an approach,
we study the two-dimensional Ising spin glass, which is a
prototypical model for complex systems with quenched disorder.
In particular we are interested in the behavior
near the phase transition point $T_c=0$, where many properties of this
model are not well understood.
We review how, by using a mapping to the
minimum-weight perfect matching problem and by applying fast
matching algorithms from computer science, one can study the model
at and close to $T=0$ in perfect equilibrium for large system sizes, easily up
to $N=512^2$ spins.
By using this approach, recently much progress has been obtained for
the system with Gaussian disorder by
studying domain-wall excitations 2 (see figure),
droplet excitations 3, energy
barriers 4 and the SLE properties of domain walls 5. All results
show that the behavior of the Gaussian model is well described by the
famous droplet ansatz, and characterized by one single exponent
$þeta\approx-0.29$, which is related to the correlation-length
exponent via $\nu=1/þeta$, as confirmed recently by Monte
Carlo simulations 6.
For the model with bimodal $J$ distribution, the situation is less
clear, in particular it is currently heavily discussed
whether the correlation length diverges
algebraically or exponentially when approaching $T_c$. Many
contradicting results from finite-temperature approaches exist. Here,
again the droplet approach 3 is applied, taking advantage of working
exactly at $T=0$. For large systems, the
droplet energy shows a power-law behavior $EL^þeta$
with $þeta=-0.234(6)$, corresponding to an algebraic
correlation-length divergence with $\nu=-1/=4.1(1)$.
This behavior is qualitatively similar to
the Gaussian system, but the value of $þeta$ is different, hence
non-universal. Furthermore, the disorder-averaged
spin-spin correlation exponent $\eta$ defined via
$S_i S_i+l \rangle^2_J l^-\eta$ is determined
here via the probability
to have a non-zero-energy droplet, and $\eta=0.223(7)>0$ is
found.\\
1) AKH and H. Rieger, Optimization Algorithms in Physics,
(Wiley-VCH, Berlin 2001)\\
2) AKH and A.P. Young, Phys. Rev. B 64, 180404 (2001)\\
3) AKH and M.A. Moore, Phys. Rev. Lett. 90, 127201 (2003)\\
4) C. Amoruso, AKH, and M.A. Moore, Phys. Rev. B 73, 184405
(2006)\\
5) C. Amoruso, AKH, M.B. Hastings and M.A. Moore,
Phys. Rev. Lett. 97, 267202 (2006)\\
6) J. Houdayer and AKH, Phys. Rev. B 70, 014418 (2004)