Towards Normal Forms for GHZ/W Calculus

Recently, a novel GHZ/W graphical calculus has been established to study and reason more intuitively about interacting quantum systems. The compositional structure of this calculus was shown to be well-equipped to sufficiently express arbitrary mutlipartite quantum states equivalent under stochastic local operations and classical communication (SLOCC). However, it is still not clear how to explicitly identify which graphical properties lead to what states. This can be achieved if we have well-behaved normal forms for arbitrary graphs within this calculus. This article lays down a first attempt at realizing such normal forms for a restricted class of such graphs, namely simple and regular graphs. These results should pave the way for the most general cases as part of future work.

💡 Research Summary

**
The paper “Towards Normal Forms for GHZ/W Calculus” addresses a fundamental gap in the recently introduced GHZ/W graphical calculus: while the calculus is known to be expressive enough to represent any multipartite quantum state up to stochastic local operations and classical communication (SLOCC), there has been no systematic way to read off the quantum state directly from a given diagram. The authors propose a first step toward a complete normal‑form theory by focusing on a restricted but non‑trivial class of diagrams—those that are simple and regular—and they show how every diagram in these classes can be reduced to a canonical form that uniquely determines the underlying SLOCC‑equivalence class.

The paper begins with a concise review of the categorical background. In a symmetric monoidal category (SMC) such as FdHilb, a commutative Frobenius algebra (CFA) consists of a multiplication μ, unit η, comultiplication δ, and counit ε satisfying the usual monoid, comonoid, and Frobenius laws. Graphically μ, η, δ, ε are depicted as black (or white) “spiders”, while the caps and cups are the corresponding unit and counit wires. A special CFA (SCFA) satisfies μ∘δ = id, and an anti‑special CFA (ACFA) satisfies δ∘μ = dim⁻¹·id. In FdHilb, SCFAs correspond precisely to GHZ‑type tripartite states, whereas ACFAs correspond to W‑type states; this bijection is the backbone of the GHZ/W calculus.

The authors then introduce the notion of a GHZ/W‑pair: a SCFA (the GHZ structure) together with an ACFA (the W structure) that satisfy four interaction equations. The “tick” (a NOT gate) emerges as a special morphism built from a GHZ and a W node; it behaves as a classical structure and will be treated separately when defining graph simplicity.

Two structural restrictions are defined:

1. Simple graphs – after removing all tick‑edges, the diagram remains connected. In other words, the underlying network of GHZ and W nodes is traversable without using ticks.
2. Regular graphs – after deleting every edge that connects a GHZ node to a W node, the diagram splits into two disconnected components, one containing only GHZ nodes and the other only W nodes. Within each component the graph is still connected.

These definitions allow the authors to isolate the combinatorial complexity that arises from mixing the two kinds of nodes. The key technical results are:

• Theorem 4.6 (Simple graphs) – Any connected simple GHZ‑graph or simple W‑graph can be rewritten, using the Frobenius and bialgebra rules, into a normal form consisting of a single spider (or a zero map) possibly decorated with a number of loops and a single tick. The normal form is completely determined by four integers: the number of inputs, outputs, loops, and ticks. Consequently, two simple diagrams are equivalent iff these four numbers coincide.

• Theorem 4.10 (Regular graphs) – Any regular GHZ/W‑graph can be decomposed into three pieces: a pure GHZ‑subgraph G, a pure W‑subgraph W, and two “mixed morphisms” M₁ and M₂ that reconnect the two subgraphs. The mixed morphisms arise from the edges that originally linked GHZ and W nodes. The theorem shows that after reducing G and W to their simple normal forms (by Theorem 4.6), the whole diagram is uniquely described by the data of G, W, and the canonical forms of M₁, M₂.

• Theorem 4.11 (Mixed morphisms) – The mixed morphisms are classified according to the number of ticks t and the number of loops ℓ they contain. Six exhaustive cases are given:

  1. t = 0 → a single spider (independent of ℓ).
  2. t = 1, t < ℓ → a spider with an extra loop.
  3. t = ℓ + 1 → a spider with a specific arrangement of loops.
  4. t = ℓ > 1 → a spider with a different loop pattern.
  5. t = ℓ = 1 → a simple two‑legged spider.
  6. 1 < t < ℓ → the morphism collapses to the zero map.

The dual statements for M₂ follow by swapping inputs and outputs. These reductions rely heavily on the bialgebra identities (β₁, β₂, β₃) and the fact that a black spider copies a white spider and vice versa.

The paper concludes by emphasizing that, although the current normal‑form theory only covers simple and regular graphs, any arbitrary GHZ/W diagram can be partially reduced to these cases by cutting along GHZ–W edges, applying the simple‑graph normal form to each piece, and then handling the remaining mixed connections with the rules of Theorem 4.11. The authors view this as a stepping stone toward a full normal‑form calculus that would enable automated reasoning about multipartite entanglement, classification of SLOCC families, and possibly the synthesis of quantum circuits directly from diagrams.

Overall, the contribution is twofold: (i) it provides concrete, diagrammatic reduction rules that are provably complete for a wide subclass of GHZ/W diagrams, and (ii) it translates the abstract categorical structure into a practical toolkit for quantum information theorists who wish to work graphically rather than algebraically. Future work will aim to extend these results to arbitrary graphs, develop software support for automatic normal‑form conversion, and explore applications to quantum protocol verification and resource‑theoretic analyses.