Continuous Credit Networks and Layer 2 Blockchains: Monotonicity and Sampling

To improve transaction rates, many cryptocurrencies have implemented so-called ''Layer-2'' transaction protocols, where payments are routed across networks of private payment channels. However, for a given transaction, not every network state provides a feasible route to perform the payment; in this case, the transaction must be put on the public ledger. The payment channel network thus multiplies the transaction rate of the overall system; the less frequently it fails, the higher the multiplier. We build on earlier work on credit networks and show that this network liquidity problem is connected to the combinatorics of graphical matroids. Earlier work could only analyze the (unnatural) scenario where transactions had discrete sizes. Superficially, it might seem like the continuous case would be harder to examine. However, removing this assumption lets us make progress in two important directions. First, we give a partial answer to the ``monotonicity conjecture'' that previous work left open. This conjecture asks that the network's performance not degrade as capacity on any edge increases. And second, we construct here a network state sampling procedure with much faster asymptotic performance than off-the-shelf Markov chains ($O(\vert E\vert β(\vert E\vert))$, where $β(x)$ is the complexity of solving a linear program on $x$ constraints.) We obtain our results by mapping the underlying graphs to convex bodies and then showing that the liquidity and sampling problems reduce to bounding and computing the volumes of these bodies. The transformation relies crucially on the combinatorial properties of the underlying graphic matroid, as do the proofs of monotonicity and fast sampling.