---
title: Quotient-Augmented Strong Split Tree (QASST)
url: https://www.emergentmind.com/topics/quotient-augmented-strong-split-tree-qasst
type: topic
---

# Quotient-Augmented Strong Split Tree (QASST)

The **Quotient-Augmented Strong Split Tree (QASST)** is a graph-theoretic data structure introduced for the efficient preparation of graph states by exploiting split decomposition and the behavior of local-complement (LC) equivalence classes. In the formulation of "Efficient Preparation of Graph States using the Quotient-Augmented Strong Split Tree" [2603.23892], QASST augments the strong split tree of a connected simple graph with explicit split-node correspondences between quotient graphs, enabling both reconstruction of the original graph and a divide-and-conquer strategy for graph-state preparation. Its principal role is to replace exhaustive optimization over an LC orbit with a structural method that is exact for several distance-hereditary families and extensible, via heuristics, to more general graphs [2603.23892].

## 1. Split decomposition and the strong split tree

Let $G=(V,E)$ be a connected simple graph. A **split** of $G$ is a bipartition $V=U_1 \sqcup U_2$ such that every edge between $U_1$ and $U_2$ induces a complete bipartite subgraph. In the notation given for the framework,
$$
V(G)=U_1 \sqcup U_2,\qquad E(G)\cap (U_1\times U_2)=U_1\times U_2.
$$
A split is **trivial** if $\min(|U_1|,|U_2|)=1$.

Two splits $(U_1,U_2)$ and $(U'_1,U'_2)$ are said to **cross** if each side of one intersects both sides of the other. A nontrivial split is **strong** if it crosses no other nontrivial split; trivial splits are declared strong by convention. A graph with no nontrivial split is called **prime** [2603.23892].

Cunningham’s split decomposition, as summarized in the paper, states that every connected graph admits a unique decomposition into **quotient graphs** obtained by repeatedly collapsing strong splits. The result is organized as a tree $T$, called the **strong split tree**, whose vertices correspond to quotient graphs $Q_1,\dots,Q_k$ and whose edges correspond to the set of strong splits of $G$ [2603.23892].

For a strong split $(U_1,U_2)$, the construction introduces two split-nodes $s_1$ and $s_2$ and forms
$$
Q_i := G[U_i]\cup\{s_i\},
$$
with
$$
E(Q_i)=E(G[U_i]) \cup \{(s_i,v): \exists w\in U_{3-i},\ (v,w)\in E(G)\}.
$$
Collapsing the split replaces the bipartite subgraph between $U_1$ and $U_2$ by the single edge $(s_1,s_2)$, after which the procedure recurses on the quotient graphs until all nontrivial splits have been collapsed [2603.23892].

This decomposition is structurally significant because it converts a graph into a tree-indexed collection of smaller graphs. A plausible implication is that the combinatorial burden of LC-orbit analysis can be shifted from the original graph to its quotient blocks.

## 2. Definition and reconstruction of the quotient-augmented strong split tree

The **quotient-augmented strong split tree** of $G$ is defined as the data
- a tree $T$ whose nodes are labeled by the quotient graphs $\{Q_1,\dots,Q_k\}$, and
- for each edge $e=(Q_i,Q_j)$ in $T$, the distinguished pair of split-nodes $(s_i^j\in Q_i, s_j^i\in Q_j)$.

Accordingly,
$$
\mathrm{QASST}(G) = (T,\{Q_i\},\{s_i^j \leftrightarrow s_j^i\}).
$$
The augmentation consists precisely in retaining, for each tree-edge, the correspondence between the split-nodes on its two incident quotient graphs [2603.23892].

From QASST one can reconstruct $G$ by replacing each tree-edge with all-to-all edges between the neighbors of the associated split-nodes and then deleting the split-nodes. In other words, the tree stores the decomposition topology, while the augmentation stores the information needed to reconstitute the original bipartite interfaces [2603.23892].

This reconstruction perspective is central to the later split–fuse procedure for graph states: the same structural information that permits graph reconstruction also prescribes how separately prepared quotient states can be reassembled. This suggests that QASST is not merely a canonical decomposition, but also an execution plan for state preparation.

## 3. Construction algorithm and computational complexity

The paper gives a recursive procedure `BUILD-QASST(G)`. If $G$ has no nontrivial strong split, the output is a tree with a single node $Q_1:=G$ and no edges. Otherwise, one finds a strong split $(U_1,U_2)$, forms quotient graphs $Q_1,Q_2$ with split-nodes $s_1,s_2$, recursively processes the reduced quotient sides, and then connects the root nodes of the two recursive trees by an edge labeled $(s_1\leftrightarrow s_2)$ [2603.23892].

The procedure depends on the availability of a strong-split decomposition routine. Using Dahlhaus–Charbit et al.’s linear-time split-decomposition algorithm, the paper states that `BUILD-QASST` runs in
$$
O(|V|+|E|).
$$
The complexity claim is therefore linear in the size of the input graph [2603.23892].

Within the paper’s scope, this complexity is important for two reasons. First, it makes the structural preprocessing scalable even when direct enumeration of LC orbits is not. Second, it means that the decomposition overhead does not dominate the cost of subsequent graph-state optimization, especially in the distance-hereditary regime where the quotient structure is particularly simple.

## 4. Invariance under local complement and consequences for LC orbits

For a vertex $v\in V(G)$, the **local complement** at $v$ is defined by
$$
c_v(G) := \text{flip all edges in the induced subgraph } G[N(v)].
$$
Two graphs are **LC-equivalent** if one can be reached from the other by a sequence of local complements; the resulting equivalence classes are the **LC orbits** [2603.23892].

The paper records the theorem attributed to Bouchet ’87 / Connolly et al. ’26:
- strong splits of $G$ are invariant under local complement;
- hence the strong split tree $T$ is an invariant of the LC orbit of $G$;
- each quotient graph $Q_i$ transforms under LC to some $Q_i'\in \mathcal{O}(Q_i)$.

As a consequence,
$$
\mathcal{O}(G)\subseteq \{\text{Reconstruction of } (T,\{R_i\}) : R_i\in \mathcal{O}(Q_i)\ \forall i\}.
$$
This gives a factorized over-approximation of the orbit of $G$ in terms of the orbits of its quotient graphs [2603.23892].

A particularly important specialization occurs for **distance-hereditary (DH)** graphs. The paper states that if $G$ is DH, then every quotient graph $Q_i$ is either a star or a complete graph, citing Cunningham ’82 as equivalent to the DH characterization. In that case,
$$
|\mathcal{O}(Q_i)| = |V(Q_i)|+1,
$$
which yields the bound
$$
|\mathcal{O}(G)| \le \prod_i (|V(Q_i)|+1).
$$
This replaces a potentially intractable global orbit search with a product bound over simple components [2603.23892].

The significance of this result lies in its reduction of LC optimization to local choices on quotient blocks. For DH graphs, those blocks are sufficiently structured that one can characterize orbits analytically rather than by brute force.

## 5. Graph-state preparation and the split–fuse construction

A graph state $|G\rangle$ is prepared as
$$
|G\rangle = \left(\bigotimes_{v\in V}|+\rangle_v\right)\cdot \prod_{(u,v)\in E} CZ_{u,v}.
$$
The paper identifies the standard resource measures as:
- **CZ-gate count**: $|E(G)|$,
- **circuit depth**: $\chi'(G)$, the edge-chromatic index, with $\chi'(G)\le \Delta(G)+1$ [2603.23892].

QASST motivates a **split–fuse** strategy:
1. prepare, in parallel, intermediate graph states $|Q_i'\rangle$ each LC-optimized within its orbit $\mathcal{O}(Q_i)$;
2. convert each $|Q_i'\rangle$ to the exact quotient state $|Q_i\rangle$ via local Cliffords;
3. perform Type-II fusions on all pairs of split-nodes $(s_i^j,s_j^i)$ to reassemble $|G\rangle$ [2603.23892].

The reconstruction proposition stated in the paper is that applying exactly one Type-II fusion per edge of $T$ on the split-node pairs in $\mathrm{QASST}(G)$ reconstructs $|G\rangle$ [2603.23892].

For a distance-hereditary graph with quotient graphs $Q_1,\dots,Q_k$, the notation is:
- $n_i$ = number of original vertices in $Q_i$,
- $d_i$ = degree of $Q_i$ in $T$,
- $k-1$ = number of fusions.

Preparing each $|Q_i'\rangle$ as a star uses $n_i+d_i-1$ CZ gates and depth $n_i+d_i-1$, with LC conversion adding at most one layer of depth. Then the $k-1$ fusions add $k-1$ entangling operations, all in one step. The resulting totals are [2603.23892]:

| Resource | Total |
|---|---|
| CZ gates | $\sum_i(n_i+d_i-1)+(k-1)=n+2k-3$ |
| Depth | $1+\max_i(n_i+d_i)$ |
| Qubits | $n+2(k-1)$ |

Since $k\le n$, the paper concludes that all three quantities scale linearly in $n$, in contrast to direct preparation of $G$, which has $O(|E|)$ gates and $O(\Delta)$ depth [2603.23892].

This construction is the operational core of QASST. It converts a structural decomposition into a preparation schedule whose asymptotic behavior is explicit.

## 6. Distance-hereditary families and extensions to general graphs

For several DH families, the paper states that QASST can be used to characterize LC orbits and identify representatives with reduced controlled-$Z$ count or preparation circuit depth. The explicitly listed families are **complete bipartite** graphs $K_{n,m}$, **clique-stars**, and **complete multipartites**. In these cases one can classify $\mathcal{O}(G)$ via QASST and pick the LC-optimal representative analytically, without brute-force search; the paper refers to closed-form formulas for $|E_{\min}|$ and $\Delta_{\min}$ in Connolly et al. ’26 [2603.23892].

For DH components, split–fuse always uses star qubits and fusions, achieving linear scaling in gates and depth. The paper further notes that for sufficiently large $n$—for example, $n>24$ in $K_{n_1,\dots,n_k}$ families—split–fuse already outperforms direct LC-optimal circuits [2603.23892].

Beyond the DH setting, QASST may contain **prime** quotient graphs. The paper proposes a generalized divide-and-conquer split–fuse strategy in which stars and complete graphs are handled as before, while prime blocks are treated either by known small-$n$ LC-optimal circuits or by a **triangle-enumeration heuristic**:
- enumerate all triangles in $H$;
- if local complement at $v$ reduces $|E(H)|$, perform it and repeat [2603.23892].

The paper reports that this hybrid method—split–fuse combined with the heuristic—shows empirical improvement on dense Erdős–Rényi graphs [2603.23892].

A plausible implication is that QASST functions as a framework rather than a single closed-form algorithm: exact structural simplification where the quotient blocks are controlled, and heuristic reduction where prime blocks remain.

## 7. Theoretical scaling, empirical behavior, and scope

The paper states a **Linear Scaling for DH** theorem: for any distance-hereditary graph on $n$ vertices with $k$ quotient blocks, the split–fuse algorithm prepares its graph state with $O(n)$ CZ gates and $O(n)$ depth, using $O(n)$ temporary qubits [2603.23892]. This gives a formal asymptotic guarantee for the DH class.

The empirical observations reported are:
- in random DH samples up to $n=1000$, split–fuse depth grows $\ll n$, often nearly constant;
- in complete-multipartite LC orbits, split–fuse outperforms direct LC-optimal from $n\approx 24$ onward;
- on dense random graphs, generalized split–fuse plus the triangle heuristic reduces gates by approximately $10$–$30\%$ over naive or heuristic-only methods [2603.23892].

The stated complexity profile is:
- QASST construction in $O(n+m)$,
- fusion scheduling in one additional step,
- LC conversions and heuristic in $O(m^{1.5}+n^3)$ worst-case for prime blocks [2603.23892].

The paper positions these gains against the difficulty of brute-force LC-orbit enumeration and states that the framework avoids the $\#P$-hard enumeration of full LC orbits [2603.23892]. Within that framing, QASST is best understood as a structural invariant and synthesis tool that leverages the preservation of strong splits under local complement, turning graph decomposition into a preparation methodology for graph states.

Source: https://www.emergentmind.com/topics/quotient-augmented-strong-split-tree-qasst