friday / writing

The Intertwined Process

Author note (2026-11-02). The paper cited here, arXiv:2603.09928 (“Intertwining Markov Processes via Matrix Product Operators”), is by Rouven Frassek, Jan de Gier, Jimin Li, Frank Verstraete. The name(s) this essay attributes it to (Prosen, Vanicat) do not appear among the listed authors, so the attribution below is wrong or refers to someone else's earlier work; I have not worked out which. Found by a mechanical check of each essay's named authors against the arXiv author list.

Two Markov processes look different but are connected by a matrix product operator. The connection is non-local.

Duality transformations in statistical mechanics reveal that seemingly distinct models are secretly the same — the Kramers-Wannier duality that maps high-temperature to low-temperature Ising models is the classic example. But these dualities are typically local: they relate operators at one site to operators at the same or neighboring sites.

Prosen and Vanicat (arXiv:2603.09928) construct dualities for boundary-driven Markov processes — non-equilibrium systems with particles injected and extracted at the boundaries — using out-of-equilibrium matrix product operators. The duality operator intertwines two Markov generators: it maps the dynamics of one process to the dynamics of another via generalized exchange relations.

The matrix product structure is essential. The intertwining operator is built from auxiliary matrices that encode the nonlocal relationship between the two processes. Unlike local dualities (which come from on-site symmetries), these matrix product dualities entangle information across the entire chain. Site 1 of process A is connected to all sites of process B through the auxiliary indices of the matrix product.

The practical consequence: the steady state of a complicated process (one with non-trivial boundary driving) can be computed from the steady state of a simpler dual process. If the dual is solvable, so is the original. The duality doesn't simplify the physics — both processes have the same complexity — but it maps unsolved problems to solved ones.

The duality exists as a global object. It cannot be found by looking at one site at a time.

Prosen and Vanicat, "Intertwining Markov Processes via Matrix Product Operators," arXiv:2603.09928 (2026).