Double Flip Move for Ising Models with Mixed Boundary Conditions

Lexing YingarXiv 2022 · OpenReview

Abstract

This note introduces the double flip move for accelerating the Swendsen-Wang algorithm for Ising models with mixed boundary conditions below the critical temperature. The double flip move consists of a geometric flip of the spin lattice followed by a spin value flip. Both the symmetric and approximately symmetric models are considered. We prove the detailed balance of the double flip move and demonstrate its empirical efficiency in mixing.

The Challenge of Mixed Boundary Conditions

The Ising model is a fundamental tool in statistical mechanics for simulating collective magnetic behavior through interacting 'spins' (valued +1 or -1) on a lattice. Sampling these configurations typically relies on Markov Chain Monte Carlo (MCMC) methods, most notably the Swendsen-Wang (SW) algorithm.

However, a significant sampling bottleneck arises in models with mixed boundary conditions. For instance, in a square lattice where the left and right boundaries are fixed to +1 and the top and bottom to -1, the system develops two distinct, high-probability macroscopic profiles. Below the critical temperature, these states are separated by a massive energy barrier. Standard local algorithms like SW struggle to navigate this barrier because transitioning between states requires the simultaneous breaking of a vast number of aligned bonds—an extremely rare event. Consequently, simulations often become 'stuck' in a single state, leading to poor mixing and biased statistical estimates.

Macroscopic states for square (a) and disk (b) domains with mixed boundary conditions.

The Double Flip Mechanism

To overcome this stagnation, the Double Flip Move is proposed as a global update to be used in conjunction with the Swendsen-Wang algorithm. The move is a composite operation:

  1. Geometric Flip: A reflection of the lattice across a symmetry axis (e.g., a diagonal).
  2. Spin-Value Flip: Negating the value of every interior spin.

The core insight is that while either a geometric reflection or a global spin flip alone would drastically alter the system's energy (leading to rejection in a Metropolis framework), performing both together preserves the relative alignment of adjacent spins. If two spins $s_i$ and $s_j$ were aligned, their reflected and negated counterparts will also be aligned, effectively preserving the configuration's energy in symmetric cases.

Exact vs. Approximate Symmetry

In perfectly symmetric domains, such as a square lattice, the double flip is a deterministic, energy-preserving move. This allows the algorithm to 'teleport' between dominant macroscopic states with a 100% acceptance rate.

In many practical applications, such as rectangular grids or irregular triangular meshes on a disk, symmetry is only approximate. For these cases, the author employs a Metropolized Move:

  • Heuristic Matching: A distance-based matching procedure (akin to a greedy optimal transport) maps reflected vertices back onto the original lattice points.
  • Metropolis-Hastings Acceptance: Because energy is not perfectly preserved, the move is treated as a proposal and accepted or rejected based on the probability ratio of the new and old configurations. This ensures the algorithm maintains mathematical rigor by satisfying the detailed balance condition.

Empirical Results and Mixing Efficiency

The effectiveness of the Swendsen-Wang with Double Flip (SWDF) algorithm was evaluated against the standard SW algorithm across various synthetic configurations, including square lattices, $100 \times 99$ rectangular grids, and randomly generated triangular meshes on a disk.

The results demonstrate a stark contrast in performance:

  • Mixing Efficiency: In all test cases, the standard SW algorithm failed to transition between the two dominant macroscopic states even once over 10,000 iterations.
  • Transition Frequency: The SWDF algorithm successfully explored the full state space. In the square lattice, it achieved approximately 100 transitions. In the complex triangular disk models, where the double flip was proposed with a probability of $\eta = 1/3$, it achieved between 35 and 48 transitions over 3,000 trials.
  • Stability: Average spin value plots confirm that SWDF oscillates between the +1 and -1 dominant regimes, while the standard SW remains trapped in its initial configuration.

Conclusion and Future Directions

The double flip move provides a robust solution for sampling Ising models with mixed boundary conditions by leveraging underlying geometric symmetries. While highly effective, its performance depends on the degree of symmetry present in the domain. Future work may explore more sophisticated optimal transport algorithms for vertex matching in irregular grids or extending the matching logic from vertices to edges. Proving polynomial mixing times for these symmetric cases remains an open theoretical challenge.

Summary updated

Abstract

This note introduces the double flip move for accelerating the Swendsen-Wang algorithm for Ising models with mixed boundary conditions below the critical temperature. The double flip move consists of a geometric flip of the spin lattice followed by a spin value flip. Both the symmetric and approximately symmetric models are considered. We prove the detailed balance of the double flip move and demonstrate its empirical efficiency in mixing.
BibTeX
@misc{ying2022double,
  author = {Lexing Ying},
  title = {{Double Flip Move for Ising Models with Mixed Boundary Conditions}},
  year = {2022},
  eprint = {2205.07226},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.07226},
}
Figure