1: \begin{abstract}

2: 

3: The problem of error correction for Gallager's low-density

4: parity-check codes is famously equivalent to that of computing

5: marginal Boltzmann probabilities for an Ising-like model with

6: multispin interactions in a non-uniform magnetic field. Since

7: the graph of interactions is locally a tree, the solution is

8: very well approximated by a generalized mean-field

9: (Bethe--Peierls) approximation. Belief propagation (BP) and

10: similar iterative algorithms are an efficient way to perform

11: the calculation, but they sometimes fail to converge, or

12: converge to non-codewords, giving rise to a non-negligible

13: residual error probability (error floor). On the other hand,

14: provably-convergent algorithms are far too complex to be

15: implemented in a real decoder. In this work we consider the

16: application of the \emph{probability-damping} technique, which

17: can be regarded either as a variant of BP, from which it

18: retains the property of low complexity, or as an approximation

19: of a provably-convergent algorithm, from which it is expected

20: to inherit better convergence properties. We investigate the

21: algorithm behaviour on a real instance of Gallager code, and

22: compare the results with state-of-the-art algorithms.

23: 

24: \end{abstract}