---
title: Pretty Good State Transfer
url: https://www.emergentmind.com/topics/pretty-good-state-transfer
type: topic
---

# Pretty Good State Transfer

Pretty good state transfer (PGST) is the approximation-theoretic relaxation of perfect state transfer (PST) in quantum-walk and spin-chain models: instead of requiring exact localization of a state at a target vertex at some finite time, PGST requires that the transfer fidelity can be made arbitrarily close to \(1\). In continuous-time formulations this is expressed by the existence, for every \(\epsilon>0\), of a time \(t\) such that \(\left|\langle b|U(t)|a\rangle\right|>1-\epsilon\) or, equivalently, \(\|U(t)e_a-\gamma e_b\|<\epsilon\) for some unimodular phase \(\gamma\); in discrete-time coined walks the same approximation condition is imposed on powers of the unitary walk operator [1611.09836] [2105.03762]. Across the literature, PGST is treated as the mathematically natural notion once exact transfer is known to be rare, and its occurrence is governed by a combination of spectral symmetry, strong cospectrality, and arithmetic constraints on eigenvalues or eigenphases [1206.0082] [1612.05603].

## 1. Dynamical frameworks and formal definition

In the continuous-time setting most works study a Hamiltonian induced by a graph. For the XY-Hamiltonian on the single-excitation subspace, the Hamiltonian acts exactly like the adjacency matrix \(A\), so the evolution is written as \(U(t)=e^{itA}\) in the path literature; in other conventions the continuous-time walk is written as \(U(t)=e^{-itA}\) or, with vertex potentials, as \(U(t)=e^{itH}\) for \(H=A+Q\) [1611.09836] [1702.07000]. For the Heisenberg (XYZ) Hamiltonian on a graph \(G\), the single-excitation dynamics is governed by \(|E(G)|I-2L(G)\), so the transport problem becomes a Laplacian spectral problem rather than an adjacency one [1608.04722].

The common definition is stable across these conventions. Perfect state transfer from \(a\) to \(b\) requires \(|U(t)_{b,a}|=1\) for some time \(t\), whereas pretty good state transfer requires that for every \(\epsilon>0\) there is a time \(t\) with \(|U(t)_{b,a}|>1-\epsilon\) [1310.3885] [1702.07000]. In state-based formulations, rather than vertex-basis transfer, one asks for \(\|U(t)|v\rangle-\gamma|w\rangle\|<\epsilon\), which is the form used for arbitrary states on paths and for multi-qubit transfer in XX chains [1910.08154] [1405.1296].

Discrete-time theories replace \(e^{itA}\) by a unitary step operator. In weighted coined walks one studies
\[
U=R(2N_t^*N_t-I),
\]
and PGST from \(x\) to \(y\) means that for every \(\epsilon>0\) there exists an integer time \(t\) such that \(\|U^t x-\gamma y\|<\epsilon\) for some unimodular \(\gamma\) [2105.03762]. In Grover walks on graphs, vertex-localized states are realized as \(\Phi_u=N^*e_u\), and transfer is studied between these states rather than directly between vertices [2508.09711].

## 2. Spectral mechanisms: strong cospectrality, parity data, and arithmetic obstructions

A recurring necessary condition for PGST is strong cospectrality. For a real symmetric matrix \(M=\sum_r \theta_r E_r\), vertices \(a,b\) are strongly cospectral when
\[
E_r e_a=\pm E_r e_b
\quad\text{for all }r,
\]
and this condition is necessary for PGST in adjacency, Laplacian, and general weighted settings [1608.04722] [2010.06779]. In the state-based extension on paths, the same condition appears as \(E_j|v^\circ\rangle=(-1)^{j+1}E_j|v\rangle\), so a state and its mirror image are strongly cospectral [1910.08154].

The second ingredient is arithmetic. For paths under the XY model, the complete criterion cited from Banchi–Coutinho–Godsil–Severini requires not only symmetry \(a+b=n+1\) but also a parity constraint on every integer relation
\[
\sum_{\theta_j\in\Theta_a}\ell_j\theta_j=0,\qquad \sum_{\theta_j\in\Theta_a}\ell_j=0,
\]
namely
\[
\sum_{\theta_j\in\Theta_a}\ell_j\sigma_j\equiv 0\pmod 2,
\]
where \(\sigma_j\) records whether the spectral projection agrees or changes sign at the two vertices [1611.09836]. The same structure reappears in the involution-with-potential theory, where the eigenvalues are split into \(+\) and \(-\) symmetry types and Kronecker’s theorem turns PGST into a simultaneous phase-approximation problem [1702.07000].

Several papers recast these conditions in algebraic terms. One approach uses the minimal polynomials \(P_+\) and \(P_-\) of \(A\) relative to \(e_x+e_y\) and \(e_x-e_y\). PGST holds if \(x,y\) are strongly cospectral and every integer dependence among roots of \(P_+\) and \(P_-\) has even total coefficient on the antisymmetric side; irreducibility and trace conditions then provide sufficient criteria, while odd-degree factors with incompatible trace averages provide obstructions [2010.06779]. In discrete-time walks the analogous arithmetic objects are the eigenangles \(\theta_r\) or \(\arccos(\lambda)\), and the parity condition is replaced by congruence conditions modulo \(m\) for \(m\)-strongly cospectral states [2105.03762].

This spectral-arithmetic structure explains why PGST is often characterized by prime, power-of-two, or square-free conditions rather than by graph-theoretic symmetry alone. It also explains a common misconception: symmetry by itself is not enough. The minimal-polynomial analysis explicitly notes that the existence of an involution does not by itself guarantee strong cospectrality; \(K_3\) with an involution swapping two vertices is the standard counterexample [2010.06779].

## 3. Paths and spin chains

Paths are the canonical testing ground for PGST. For the XY-Hamiltonian, the path \(P_n\) has eigenvalues
\[
\theta_j=2\cos\!\left(\frac{\pi j}{n+1}\right),\qquad j=1,\dots,n,
\]
and strong cospectrality on a path is completely rigid:
\[
a\text{ and }b\text{ are strongly cospectral}\iff a+b=n+1.
\]
Thus only mirror-symmetric pairs can support transfer [1611.09836] [1612.05603].

The full classification on paths states that PGST between vertices \(a\) and \(b\) occurs if and only if \(a+b=n+1\), \(n+1\) has at most one odd non-trivial divisor, and, if \(n=2^t r-1\) with \(r\) odd and \(r\neq 1\), then \(a\) is a multiple of \(2^{t-1}\) [1612.05603]. Equivalently, PGST occurs exactly when either \(n=2^t-1\), or \(n=2^t p-1\) with \(p\) an odd prime and \(a\) divisible by \(2^{t-1}\) [1612.05603].

A particularly important subfamily is \(P_{2^t p-1}\) with \(p\) an odd prime. For these paths, PGST occurs between \(a\) and \(2^tp-a\) whenever \(a\) is a multiple of \(2^{t-1}\), giving the first examples of PGST between internal vertices on a path in a regime where endpoint PGST fails [1611.09836]. This family was later absorbed into the complete path characterization [1612.05603].

For Heisenberg chains the arithmetic changes. In an unmodulated Heisenberg chain of \(n\) qubits, PGST between the extremal vertices occurs if and only if \(n\) is a power of \(2\), and when \(n\) is a power of \(2\), PGST occurs between \(j\) and \(n+1-j\) for every \(j\) [1608.04722]. Thus the XY and Heisenberg path theories are related by the same general machinery—strong cospectrality plus Diophantine approximation—but differ in their exact admissible lengths.

The path literature also extends beyond single vertices. In uniformly coupled XX chains, arbitrary multi-qubit states, including entangled states, have PGST if and only if the chain length satisfies the same number-theoretic conditions previously known for single-qubit transfer, namely \(n=p-1\), \(n=2p-1\), or \(n=2^k-1\) [1405.1296]. A later state-based treatment refined this by showing that PGST between a state \(|v\rangle\) and its mirror \(|v^\circ\rangle\) is controlled by the eigenvalue support \(\Theta_{|v\rangle}\); for parity states on \(P_{m-1}\) with \(m=2^tp^s\) or \(m=p^s\), the criterion is expressible by forbidden support blocks \(S_c\) or \(R_c\) [1910.08154].

Finally, potentials fundamentally alter the path problem. In graphs with an involution, a symmetric potential can be chosen so that PGST occurs between symmetric vertices; as a special case, a potential supported only on the two endpoints of a path induces endpoint PGST for paths of any length [1702.07000]. This sharply contrasts with the no-potential classification and with the impossibility of PST for paths of length at least \(4\) under arbitrary potential [1702.07000].

## 4. Beyond paths: trees, circulants, products, and large PGST classes

Outside paths, one of the earliest exact arithmetic classifications is for double stars. No double star graph admits perfect state transfer, but in the symmetric double star \(S_{k,k}\) there is PGST between the two central vertices if and only if \(4k+1\) is not a perfect square [1206.0082]. In the asymmetric family \(S_{2,\ell}\), PGST occurs between the two leaves adjacent to the degree-\(3\) vertex if and only if \(\ell\neq 2\) [1206.0082].

For circulant graphs, the cycle \(C_n\) admits PGST if and only if \(n=2^k\) for some \(k\ge 2\), and when it occurs it is only between antipodal vertices [1607.03598]. More generally, the edge-disjoint union \(C_{2^k}\cup G(2^k,D)\), where \(G(2^k,D)\) is an integral circulant graph with \(1\notin D\), also admits PGST, and so does its complement [1607.03598]. These constructions produce additional non-circulant PGST graphs through Cartesian products and complements [1607.03598].

NEPS of \(P_3\) provide another spectral-arithmetic class. If the basis contains tuples of both even and odd Hamming weights, then the resulting NEPS does not exhibit PST [1604.08858]. Nevertheless, PGST can occur when a zero-sum condition holds for the minimal-weight even or odd part of the basis, with Kronecker approximation used to align the periodic times of one factor with the PST times of the other [1604.08858].

Cartesian products also show that PGST behaves very differently from PST at the level of equivalence classes. For products of paths, there is no bound on the size of a set of vertices that admit PGST between any two vertices of the set: the corners of suitable products form PGST-equivalence classes of arbitrarily large size [2305.14276]. This contrasts with the familiar PST constraint that an equivalence class can have at most two vertices [2305.14276].

At the opposite extreme lies universal transfer. Universal pretty good state transfer means PGST between every pair of vertices. Graphs with universal PGST must have distinct eigenvalues and a flat eigenbasis, their switching automorphism group is abelian and its order divides the number of vertices, and explicit infinite families include prime-length Hermitian cycles \(C_p\), the Cartesian products \(K_2\square C_p\), and a real symmetric Hadamard-based construction [1310.3885]. These examples show that PGST is compatible with universal transport even when universal PST is impossible for real symmetric adjacency matrices [1310.3885].

## 5. Potentials, weighted modifications, and constructive design

A major branch of the literature studies how PGST can be induced rather than merely detected. In graphs with an involution \(\sigma\), if \(\sigma\) fixes at least one vertex or at least one edge, there exists a symmetric potential \(Q\) for which PGST occurs between any pair of symmetric vertices \(u\) and \(\sigma u\) [1702.07000]. In many cases the potential can be chosen to be nonzero only on the target pair, and for paths this implies that every path \(P_N\) can be made to exhibit endpoint PGST by placing the same potential on the endpoints only [1702.07000].

This symmetry-based theory was generalized to asymmetric graphs. Given any graph with a pair of cospectral vertices, a simple graph modification together with a suitable potential yields PGST between those vertices [1804.01645]. The mechanism is algebraic: a transcendental potential on the target pair can force strong cospectrality, make the associated factors \(P_+\) and \(P_-\) irreducible, and activate the trace criterion for PGST [1804.01645].

The design perspective becomes explicit in the isospectral-reduction approach. Isospectral reduction on the target set \(S=\{u,v\}\),
\[
\mathcal{R}_S(H,\lambda)=H_{SS}-H_{S\overline{S}}\left(H_{\overline{S}\overline{S}}-\lambda I\right)^{-1}H_{\overline{S}S},
\]
produces a reduced matrix from which the parity factors \(P_+\) and \(P_-\) can be extracted. By tuning parameters so that \(P_+\) and \(P_-\) are irreducible and satisfy
\[
\frac{\operatorname{Tr}(P_+)}{\deg(P_+)}\neq \frac{\operatorname{Tr}(P_-)}{\deg(P_-)},
\]
one obtains a constructive workflow for designing PGST networks [1908.02046]. The same paper further shows how such networks can be modified to include compact localized states for robust storage and subsequent transfer of qubits [1908.02046].

Weighted path modifications provide another exact design arena. For the symmetrically modified path \(P_N(M,w)\), where two extra vertices are attached at symmetric positions with edge weight \(w\), PGST between the endpoints occurs for transcendental \(w\) when \(M\) is odd, \(N\) is even, and \(\gcd(N+1,M)=1\); several complementary parity and gcd conditions rule it out [2010.06779]. This is representative of a broader theme: once strong cospectrality is present, the deciding factor is often the algebra of \(P_+\) and \(P_-\), not merely the combinatorics of the underlying graph [2010.06779].

## 6. Discrete-time theories, engineered platforms, and current directions

The discrete-time theory of PGST parallels the continuous-time one but replaces eigenvalue relations by angle relations. In a broad class of coined walks, PGST is characterized by a Hermitian adjacency matrix \(H\), \(m\)-strong cospectrality relative to \(H\), and number-theoretic conditions on the angles \(\arccos(\lambda)\) [2105.03762]. For vertex states \(N_t^*e_a\) and \(N_t^*e_b\), the transfer problem reduces to modular constraints on integer relations among these angles, and PGST implies that the support splits into two phase classes differing by \(m/2\) [2105.03762].

Grover walks sharpen this picture further. With discriminant matrix \(P\), the identity
\[
NU^mN^*=T_m(P)
\]
links the walk to Chebyshev polynomials, so PGST becomes a question of whether \(T_m(\mu)\) can be made simultaneously close to \(1\) or \(-1\) on the relevant spectral supports [2508.09711]. For abelian Cayley graphs, PGST between \(u\) and \(v\) requires that \(v-u\) have order \(2\), together with a parity condition on integer relations among \(\arccos\mu_a\); for unitary Cayley graphs the complete characterization is
\[
G_{\mathbb Z_n}\text{ exhibits PGST}\iff n=2^\alpha m,\ \alpha\in\{1,2\},\ m\text{ odd square-free}
\]
[2508.09711]. The same paper notes that on periodic graphs PGST and PST coincide, so the approximate theory becomes genuinely new only beyond the periodic regime [2508.09711].

A closely related development shows that simple weighted Grover coins enable PGST between antipodal vertices on every hypercube \(Q_d\). The construction uses real weighted coins and changes the weight on only one arc per vertex, extending a previously prime-dimension result to all \(d\) [2412.20753]. This provides a discrete-time counterpart to the arithmetic classifications on continuous-time paths and cycles.

In physically motivated spin-chain models, PGST is increasingly treated as a design target rather than a rare emergent property. Site-dependent exchange coefficients obtained by global optimization yield near-perfect transfer in isotropic and anisotropic Heisenberg chains without time-dependent external control, with transfer probabilities above \(0.99\) reported for \(N=30\) examples and chain lengths up to \(N=90\) explored numerically [2101.03194]. Adaptive quantum error correction recasts transfer through 1-D Heisenberg chains as communication over an amplitude-damping-type channel; with a 4-qubit approximate code and channel-adapted recovery, the worst-case fidelity improves to
\[
F_{\min}^2\approx 1-\frac{7}{4}p^2+O(p^3),
\qquad p=1-|f_{r,s}^N(t)|^2,
\]
which enlarges the regime of pretty good transfer and remains effective for weak disorder [1807.04062]. In decorated transmon qubit chains, homogeneous exactly solvable sequences realize PGST at long times through phase alignment of incommensurate eigenvalues, whereas inverse spectral design yields PST and optimized short-time transfer; notably, dimerized SSH-like couplings produce edge states and spectral gaps but do not generally maximize fast transfer fidelity [2501.10580].

Taken together, these results suggest that PGST is best understood not as a single graph property, but as a spectral approximation phenomenon that can arise from unmodulated symmetry, from arithmetic features of cyclotomic or trigonometric spectra, from engineered potentials and weights, or from explicitly optimized hardware models. The unifying structure remains the same: strong cospectrality determines the admissible transfer pairs, and number-theoretic compatibility of eigenphases determines whether arbitrarily accurate phase alignment is possible [1612.05603] [2105.03762].

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