---
title: Pair State Transfer in Quantum Walks
url: https://www.emergentmind.com/topics/pair-state-transfer
type: topic
---

# Pair State Transfer in Quantum Walks

Pair state transfer is a family of state-transfer phenomena in continuous-time quantum walks in which the transported object is not necessarily a single vertex state but a two-vertex superposition, most commonly the antisymmetric state $e_a-e_b$. In the graph-theoretic literature, the central setting uses a graph Laplacian $L$ and unitary evolution $U(t)=\exp(itL)$, with perfect pair state transfer occurring when $U(\tau)(e_a-e_b)=\gamma(e_c-e_d)$ for some $\tau$ and $|\gamma|=1$ [1906.01591]. Closely related strands study adjacency-generated dynamics $U(t)=\exp(-itA)$, signless-Laplacian dynamics, Hermitian or oriented graphs, and generalized $s$-pair states $e_u+s e_v$, where the pair state and plus state arise as the special cases $s=-1$ and $s=1$ [2404.16654]. Across these settings, the theory combines spectral decomposition, strong cospectrality, arithmetic conditions on eigenvalue supports, and graph constructions that either enable or obstruct exact or asymptotic transfer [1906.01591].

## 1. Formal framework and variants

For a graph with adjacency matrix $A$, degree matrix $\Delta$, Laplacian $L=\Delta-A$, and signless Laplacian $Q=\Delta+A$, continuous-time quantum walks are generated by exponentials of one of these Hermitian matrices. In the Laplacian formulation of Chen and Godsil, the pair state associated with vertices $a,b$ is $e_a-e_b$, and perfect pair state transfer (PPST) from $(a,b)$ to $(c,d)$ at time $\tau$ means
$$
U(\tau)(e_a-e_b)=\gamma(e_c-e_d),
$$
with $U(t)=\exp(itL)$ and $|\gamma|=1$ [1906.01591]. The corresponding density-matrix formulation uses
$$
D_{ab}=\frac12(e_a-e_b)(e_a-e_b)^T,
$$
and PPST is equivalent to $U(\tau)D_{ab}U(-\tau)=D_{cd}$ [1906.01591].

The same paper treats the signless-Laplacian analogue by replacing $L$ with $Q$ and $e_a-e_b$ with the plus state $e_a+e_b$ [1906.01591]. More generally, the $s$-pair formalism studies states of the form $e_u+s e_v$ for nonzero $s$, with perfect $s$-pair transfer defined by the same phase-equivalence condition under $A$, $L$, or $Q$ [2404.16654]. In bipartite graphs, plus-state transfer relative to $Q$ is equivalent to pair-state transfer relative to $L$, via a signature matrix $D$ satisfying $DLD=Q$ [1906.01591].

A nearby, older usage speaks of “pairwise” transfer between a chosen source-target vertex pair. In adjacency-generated walks on Hermitian graphs, this includes perfect and pretty good transfer between vertices, and universal state transfer when transfer occurs between every ordered pair [1310.3885]. This suggests that the literature uses two related conventions: one centered on antisymmetric or generalized two-vertex states, and one centered on transfer between specified vertex pairs. The former is the standard meaning of “pair state transfer” in the Laplacian literature [1906.01591].

| Framework | Evolution | State type |
|---|---|---|
| Adjacency | $U(t)=e^{-itA}$ | vertex states, $s$-pair states |
| Laplacian | $U(t)=e^{itL}$ | pair states $e_a-e_b$ |
| Signless Laplacian | $U_Q(t)=e^{itQ}$ | plus states $e_a+e_b$ |
| $q$-Laplacian | $U_{\mathscr L}(t)=e^{it\mathscr L}$ | pair, plus, and $s$-pair states |

## 2. Spectral criteria and arithmetic structure

The spectral decomposition
$$
L=\sum_r \lambda_r E_r,\qquad U(t)=\sum_r e^{it\lambda_r}E_r
$$
is the basic analytic tool for PPST [1906.01591]. The eigenvalue support of $e_a-e_b$ is the set of Laplacian eigenvalues $\lambda_r$ with $E_r(e_a-e_b)\neq 0$, and the decisive symmetry notion is strong cospectrality:
$$
E_r(e_a-e_b)=\pm E_r(e_c-e_d)\quad\text{for every }r.
$$
If PPST occurs, strong cospectrality is necessary [1906.01591].

Chen and Godsil give a necessary and sufficient phase-alignment criterion. Writing $\Lambda^+_{ab,cd}$ and $\Lambda^-_{ab,cd}$ for the eigenvalues on which the projections agree or differ by a sign, PPST occurs at time $\tau$ if and only if the pair states are strongly cospectral and, for some $\lambda_0\in\Lambda^+_{ab,cd}$,
$$
\tau(\lambda_0-\lambda_r)\in 2\pi\mathbb Z\quad (\lambda_r\in\Lambda^+_{ab,cd}),
$$
$$
\tau(\lambda_0-\lambda_r)\in \pi+2\pi\mathbb Z\quad (\lambda_r\in\Lambda^-_{ab,cd}) .
$$
The same paper shows that periodicity of $e_a-e_b$ is equivalent to the ratio condition
$$
\frac{\lambda_r-\lambda_s}{\lambda_k-\lambda_\ell}\in\mathbb Q,
$$
and that the support eigenvalues must be either all integers or all quadratic integers in a common field $\mathbb Q(\sqrt{\Delta})$ [1906.01591].

The adjacency-based PGST theory sharpens this arithmetic picture through relative minimal polynomials $P_+(x)$ and $P_-(x)$ attached to $e_u+e_v$ and $e_u-e_v$ [2010.06779]. In that formulation, PGST occurs if and only if strong cospectrality holds and every integer relation
$$
\sum_i \ell_i \alpha_i+\sum_j m_j \beta_j=0,\qquad \sum_i \ell_i+\sum_j m_j=0,
$$
among roots $\alpha_i$ of $P_+$ and $\beta_j$ of $P_-$ satisfies the parity constraint $\sum_j m_j\equiv 0\pmod 2$ [2010.06779]. The same work gives sufficient conditions via irreducibility and trace-per-degree equality, and obstruction theorems based on odd-degree factors [2010.06779]. For real $s$-pair states, the same strong-cospectrality and integer-or-quadratic-integer dichotomy reappears in the general transfer theorem of [2404.16654].

## 3. Structural phenomena specific to pair states

Pair states exhibit features that do not occur for ordinary vertex transfer. The most striking is transitivity: if PPST occurs at time $\tau$ from $e_a-e_b$ to $e_\alpha-e_\beta$ and simultaneously from $e_b-e_c$ to $e_\beta-e_\gamma$, then PPST also occurs at time $\tau$ from $e_a-e_c$ to $e_\alpha-e_\gamma$ [1906.01591]. The proof uses the density-matrix identity
$$
D_{ac}=D_{ab}+D_{bc}-2D_{ab}D_{bc}-2D_{bc}D_{ab},
$$
which has no vertex-state analogue [1906.01591].

At the same time, pair transfer retains a monogamy property: for any fixed pair $(a,b)$ there is at most one pair $(c,d)$ with PPST to $(a,b)$ [1906.01591]. Another distinctive feature is the existence of fixed pair states. The state $e_a-e_b$ is fixed, meaning $U(t)(e_a-e_b)=\gamma_t(e_a-e_b)$ for all $t$, if and only if $a$ and $b$ are twins, namely $N(a)\setminus\{b\}=N(b)\setminus\{a\}$ [1906.01591]. Fixed pair states cannot exhibit PPST [1906.01591].

For real symmetric adjacency matrices, universal perfect state transfer between all vertex pairs is impossible [1310.3885], but pair-state transfer under Laplacian dynamics is far less rigid. Chen and Godsil report that Laplacian pair PST appears far more frequently than adjacency vertex PST in enumerations up to $8$ vertices [1906.01591]. This suggests that the antisymmetric two-vertex sector has a substantially richer transfer geometry than the one-vertex sector, even though both are governed by comparable spectral constraints.

A recurrent heuristic identifies edge-state transfer in a graph with vertex PST in its line graph. The 2019 paper notes that this correspondence appears for paths and cycles, with matching times, but does not hold generally beyond these classes [1906.01591]. It therefore functions as a useful guide in special families, not as a general equivalence.

## 4. Exact classifications on basic graph families

The path and cycle classifications are the canonical starting point. For Laplacian pair state transfer on cycles, $C_4$ is the only cycle with PPST; it occurs between opposite edges at time $\tau=\pi/2$ [1906.01591]. For paths, $P_n$ has PPST if and only if $n\in\{3,4\}$, with transfer between the two edges of $P_3$ at $\tau=\pi/2$ and between the two end edges of $P_4$ at $\tau=(\sqrt2)\pi/2$ [1906.01591]. By bipartite equivalence, the same list holds for plus-state transfer under the signless Laplacian [1906.01591].

The generalized $s$-pair theory yields a different classification on cycles because it encompasses more than antisymmetric edge states. For $C_n$, perfect $s$-pair transfer occurs only in explicit cases on $C_4$, $C_6$, and $C_8$, including $e_0+s e_1\to e_2+s e_3$ on $C_4$ at $\tau=\pi/2$, $e_0-e_2\to e_3-e_5$ on $C_6$ at $\tau=\pi/2$, and $e_0-e_2\to e_4-e_6$ on $C_8$ at $\tau=\pi/\sqrt2$ [2404.16654]. This does not contradict the Laplacian pair-state classification, because the underlying state space and Hamiltonians differ [2404.16654].

Threshold graphs admit an exact Laplacian characterization. A connected threshold graph has LPST if and only if either it is of the exceptional form $K_2^c\vee K_t$, where LPST may occur without Laplacian vertex transfer, or its block sizes satisfy explicit congruence conditions modulo $4$ [2601.13318]. The same paper proves that, on connected threshold graphs, LPST and Laplacian vertex state transfer are equivalent if and only if the graph is not a join $K_2^c\vee K_t$ with $t\not\equiv 2\pmod 4$ [2601.13318].

For adjacency-based PGST, paths admit a complete classification: PGST on $P_n$ between vertices $a$ and $b$ occurs precisely when $a+b=n+1$ and one of three arithmetic conditions on $n$ holds, namely $n=2^t-1$, $n=p-1$ with $p$ an odd prime, or $n=2^t p-1$ with $p$ odd prime and $a$ a multiple of $2^{t-1}$ [2010.06779]. Cycles admit PGST if and only if their length is a power of two, and then only between antipodal vertices [1607.03598].

## 5. Constructions, reductions, and graph operations

Several later papers convert pair-state transfer into a design principle. For graphs with an involution, the $q$-Laplacian theory reduces antisymmetric pair-state transfer on the full graph to vertex PST on a smaller “half graph” governed by the minus-sector matrix $\mathscr L_-=\mathscr L'-A_\phi$ [2509.20749]. In block form,
$$
M^T U_{\mathscr L}(t) M=
\begin{bmatrix}
U_{\mathscr L_-}(t)&0\\
0&U_{\widetilde{\mathscr L}_+}(t)
\end{bmatrix},
$$
so transfer between $\frac1{\sqrt2}(e_a-e_{\phi(a)})$ and $\frac1{\sqrt2}(e_b-e_{\phi(b)})$ is equivalent to vertex PST between $e_a$ and $e_b$ in the half graph [2509.20749]. This yields infinite families of trees with potentials and unicyclic graphs of maximum degree three that exhibit perfect pair state transfer [2509.20749].

A related locality principle appears for isomorphic branches. If two induced subgraphs $X_1$ and $X_2$ are isomorphic and attached symmetrically to the rest of the graph, then the antisymmetric pair-state subspace spanned by $(e_a-e_{\hat f(a)})/\sqrt2$ evolves exactly as the adjacency walk on one branch:
$$
Q^T U_G(t)Q=
\begin{bmatrix}
U_{X_1}(t)&0\\
0&U_{G/\Pi}(t)
\end{bmatrix}.
$$
Consequently, any graph with high-fidelity vertex transfer can be embedded as isomorphic branches of a larger graph to produce high-fidelity pair-state transfer [2402.07078]. This mechanism is used to exhibit pair transfer on paths, cycles, and augmented graphs [2402.07078].

The generalized-cluster framework of edge-perturbed graphs gives a similar reduction under $A$, $L$, and $Q$. If $C$ is a cluster and $x\perp 1_c$, then
$$
U_{G(H)}(t)\begin{bmatrix}x\\0\end{bmatrix}
=
e^{i\delta t(1_s^T z)}
\begin{bmatrix}
U_H(t)x\\0
\end{bmatrix},
$$
so PST in the local graph $H$ is inherited by the larger graph $G(H)$ [2505.07982]. Using $H=C_4$, the paper constructs an infinite family of non-regular graphs of maximum valency five with perfect pair state transfer under adjacency, Laplacian, and signless Laplacian between the same pair states at the same time [2505.07982].

Product constructions further enlarge the catalogue. For regular graphs, tensor-product and double-cover theorems characterize Laplacian PPST in terms of synchronized PST or pair-PST on the factors, with explicit congruence conditions on spectral supports and transfer phases [2509.18858]. In particular, $K_n\times P_2$, $K_{2n}\times C_4$, and $K_n\ltimes K_n$ furnish families with LP-PST at $t=\pi/2$ [2509.18858].

## 6. Global obstructions, universal regimes, and current directions

The strongest negative results concern graph transforms. For the $Q$-graph of an $r$-regular graph, Laplacian perfect pair state transfer does not occur when $r+1$ is prime or a power of $2$ [2407.14376]. For total graphs $\mathcal T(G)$, if $G$ is $r$-regular with $r>2$ and $r+1$ is not a Laplacian eigenvalue of $G$, then $\mathcal T(G)$ has no Laplacian perfect pair state transfer, although under mild conditions it does exhibit Laplacian pretty good pair state transfer [2602.08684]. These obstructions are derived from explicit block spectral decompositions and support-pairing arguments [2407.14376], [2602.08684].

At the opposite extreme lies universal transfer. For adjacency walks on Hermitian graphs, universal pretty good or perfect transfer between every pair of vertices forces a flat eigenbasis and distinct eigenvalues, and imposes strong restrictions on the switching automorphism group [1310.3885]. Oriented graphs are especially rigid: the oriented edge and the oriented $3$-cycle are the only oriented graphs with universal perfect state transfer between every pair of vertices [2301.01473]. This resolves the oriented universal-PST problem and sharpens the contrast with Laplacian pair-state transfer, where transfer between many antisymmetric states can coexist without universal vertex transfer [2301.01473].

Two broader directions are active. One is generalization of the state space: $s$-pair states, plus states, and real pure states unify pair transfer with other two-vertex superpositions [2404.16654]. The other is implementation. In hypercube architectures, one can determine the unique sub-hypercube containing a chosen pair of vertices as antipodes, switch on only the couplings in that sub-hypercube, and achieve perfect transfer at
$$
t^*=\frac{\pi}{2J},
$$
with amplitude $(-i\sin(Jt))^k$ across a $k$-dimensional active subcube [2011.03586]. This is a physically motivated use of “pair state transfer” in the source-target sense, and it suggests a plausible engineering implication: controllable symmetry reduction is as important in hardware design as it is in the abstract graph-theoretic constructions.

Open problems recur across the literature. These include classification of non-circulant universal PGST graphs [1310.3885], efficient certification of switching equivalence and pair-transfer criteria [1310.3885], extension of LP-PGST results on total graphs to edge-edge and mixed pair states [2602.08684], and broader characterizations of $q$-Laplacian pair transfer beyond involution-based constructions [2509.20749]. The cumulative picture is that pair state transfer is simultaneously more flexible than vertex PST and more algebraically constrained than naive intuition suggests: it thrives on symmetry, strong cospectrality, and controlled arithmetic in the eigenvalue support, but it is sharply limited by parity, integrality, and support-structure obstructions [1906.01591].

Source: https://www.emergentmind.com/topics/pair-state-transfer