Continuous-Time Chiral Quantum Walk

• The paper introduces continuous-time chiral quantum walks defined by Hermitian Hamiltonians with complex off-diagonal phases that encode chirality through gauge-invariant loop fluxes.
• It demonstrates how breaking time-reversal and spatial-swap symmetries enables asymmetric interference, directional bias, and optimized transport in various network geometries.
• The work highlights design principles for achieving uniform mixing, zero transfer, and enhanced entanglement transfer, providing actionable insights for quantum network routing.

Searching arXiv for recent and foundational papers on continuous-time chiral quantum walks. arxiv_search(query="continuous-time chiral quantum walk", max_results=10) arxiv_search({"query":"continuous-time chiral quantum walk","max_results":10}) Searching arXiv for “continuous-time chiral quantum walk”, “uniform mixing chiral quantum walks”, and related terms. A continuous-time chiral quantum walk is a continuous-time quantum walk generated by a Hermitian graph Hamiltonian with complex off-diagonal phases, so that edge direction is encoded without sacrificing unitarity. In the standard formulation, chirality is carried by gauge-invariant net phases around loops, and these phases break time-reversal symmetry, probability time symmetry, or spatial-swap symmetry in ways that are impossible for real-symmetric adjacency or Laplacian generators. Across graph-theoretic, transport-theoretic, and Floquet settings, continuous-time chiral quantum walks have been used to analyze directional transport, zero transfer, rapid departure from high-degree vertices, uniform mixing on signed complete graphs and Hamming graphs, average uniform mixing on oriented circulants, and long-time asymmetric hydrodynamics on lattices (Levine et al., 6 May 2026).

1. Definition and gauge structure

A unitary-signed graph $X^\sigma$ consists of an underlying simple graph $X=(V,E)$ and a signing $\sigma:E\to U(1)\subset\mathbb C$ such that the Hermitian signed adjacency matrix satisfies

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$

Oriented graphs are the special case $\sigma_{uv}\in\{\pm i\}$. The continuous-time chiral quantum walk is generated by

$$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$

with Born probabilities

$p_v(t)=|\langle v|\psi(t)\rangle|^2.$

Equivalently, one may use the mixing matrix

$$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$

where $\circ$ denotes the Schur product (Levine et al., 6 May 2026).

A broader formulation writes the Hamiltonian on a graph $G(V,E)$ as

$X=(V,E)$0

or, with explicit hopping strengths,

$X=(V,E)$1

with $X=(V,E)$2 and $X=(V,E)$3 (Bottarelli et al., 2023).

The central structural fact is gauge equivalence. If

$X=(V,E)$4

then site-to-site probabilities are unchanged. On trees, all phases can be gauged away, so chirality has no measurable effect on transfer probabilities. On graphs with cycles, the gauge-invariant data are the net phases around independent loops. For planar graphs with $X=(V,E)$5 edges and $X=(V,E)$6 vertices, the number of independent loop phases is

$X=(V,E)$7

This is why chiral effects are intrinsically loop phenomena rather than merely consequences of complex matrix entries (Annoni et al., 2023).

2. Symmetry breaking and interference

For real Hamiltonians in the site basis, transport probabilities obey

$X=(V,E)$8

These identities express time-reversal symmetry and the absence of directional bias. Complex phases on loops generically break these equivalences and thereby enable asymmetric transport (Annoni et al., 2023).

On the triangular chain, chirality is implemented by assigning a phase $X=(V,E)$9 to directed edges. For $\sigma:E\to U(1)\subset\mathbb C$0 and $\sigma:E\to U(1)\subset\mathbb C$1, the Hamiltonian is

$\sigma:E\to U(1)\subset\mathbb C$2

and the net cycle phase per triangle is $\sigma:E\to U(1)\subset\mathbb C$3. In this setting, $\sigma:E\to U(1)\subset\mathbb C$4 and $\sigma:E\to U(1)\subset\mathbb C$5. The paper associates the directional bias with asymmetric interference between amplitudes traversing odd- and even-length paths, controlled by the gauge-invariant cycle phase $\sigma:E\to U(1)\subset\mathbb C$6 (Sağlam et al., 2021).

A related lattice formulation uses nearest-neighbor and next-nearest-neighbor hoppings on an infinite one-dimensional lattice. After gauging away the nearest-neighbor phase, the Hamiltonian is

$\sigma:E\to U(1)\subset\mathbb C$7

with dispersion and group velocity

$\sigma:E\to U(1)\subset\mathbb C$8

For $\sigma:E\to U(1)\subset\mathbb C$9, inversion symmetry is broken, $$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$0, and the causal cone becomes asymmetric about the origin. The third moment and skewness are

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$1

so chirality is reflected directly in odd moments of the position distribution (Bhandari et al., 2021).

These formulations make precise a common misconception. Chirality is not simply “complex hopping”; it is the physically nontrivial part of complex hopping that survives gauge transformations. On trees there is no such invariant content, whereas on loopy graphs the loop flux controls interference, directional bias, and transport asymmetry (Chaves et al., 2022).

3. Transport enhancement and routing

On open chains, chiral phases are gauge-equivalent to a non-chiral Hamiltonian with a phase-imprinted initial state. For a chain Hamiltonian

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$2

a diagonal unitary with $$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$3 removes the phases from the Hamiltonian and transfers them to the initial packet. The resulting local wavenumber shift is

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$4

so the “directionality” of a chiral chain can be interpreted as a momentum-shifted initial state on a standard chain. For a uniform chain,

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$5

and $$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$6 minimizes dispersion while maximizing ballistic speed (Yu et al., 2023).

A non-gauge-trivial example is the Y-junction graph, where three open chains meet in a triangle carrying a gauge-invariant flux. The Hamiltonian is

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$7

with

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$8

and

$$A^\sigma_{uv}=\sigma_{uv}\quad\text{if }(u,v)\in E,\qquad \sigma_{vu}=\overline{\sigma_{uv}}.$$9

The dynamics reduces to three effective open chains with boundary potentials

$\sigma_{uv}\in\{\pm i\}$0

and scattering phase shifts

$\sigma_{uv}\in\{\pm i\}$1

After a single encounter with the junction, the branch amplitudes are

$\sigma_{uv}\in\{\pm i\}$2

Directed complete transport is obtained under the explicit condition

$\sigma_{uv}\in\{\pm i\}$3

which gives $\sigma_{uv}\in\{\pm i\}$4 efficiency toward a selected branch (Yu et al., 2023).

A minimal six-vertex chiral router realizes the same principle on a graph with one triangular loop. With $\sigma_{uv}\in\{\pm i\}$5, the transfer probabilities from input vertex $\sigma_{uv}\in\{\pm i\}$6 to output vertices $\sigma_{uv}\in\{\pm i\}$7 and $\sigma_{uv}\in\{\pm i\}$8 can be tuned by a single flux $\sigma_{uv}\in\{\pm i\}$9. At $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$0, $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$1 yields $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$2 and $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$3, while $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$4 swaps the outputs. At the same time, the router also realizes universal routing of superpositions, with

$$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$5

independent of the input phase $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$6. The same model supports phase estimation; for large $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$7, the quantum Fisher information prefactor obeys $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$8 (Bottarelli et al., 2023).

Chirality also improves entanglement transfer on triangular chains. For $$H^\sigma:=A(X^\sigma),\qquad |\psi(t)\rangle=e^{-itH^\sigma}|\psi(0)\rangle,$$9, edge phase $p_v(t)=|\langle v|\psi(t)\rangle|^2.$0, and Bell injection $p_v(t)=|\langle v|\psi(t)\rangle|^2.$1, the concurrence on the target pair satisfies

$p_v(t)=|\langle v|\psi(t)\rangle|^2.$2

while the site-transfer probability reaches

$p_v(t)=|\langle v|\psi(t)\rangle|^2.$3

For the non-chiral walk on the same graph, the best short-time two-site entanglement transfer occurs at $p_v(t)=|\langle v|\psi(t)\rangle|^2.$4, with

$p_v(t)=|\langle v|\psi(t)\rangle|^2.$5

The long-time regime exhibits pretty good state transfer, with $p_v(t)=|\langle v|\psi(t)\rangle|^2.$6 at $p_v(t)=|\langle v|\psi(t)\rangle|^2.$7 for the chiral walk (Sağlam et al., 2021).

4. Mixing, delocalization, and complete-graph phenomena

Three mixing notions are standard. Instantaneous global uniform mixing at time $p_v(t)=|\langle v|\psi(t)\rangle|^2.$8 is

$p_v(t)=|\langle v|\psi(t)\rangle|^2.$9

Local uniform mixing from vertex $$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$0 is

$$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$1

Average uniform mixing is defined through

$$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$2

and occurs when

$$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$3

These notions separate one-time exact uniformity, vertex-specific uniformity, and long-time Cesàro uniformity (Levine et al., 6 May 2026).

A major recent result is a stopping-rule construction for probabilistic uniform mixing on signed complete graphs. If $$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$4 has order $$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$5 and $$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$6 with $$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$7 simple, then the cone

$$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$8

has uniform mixing in expected $$M_X^\sigma(t)=U_X^\sigma(t)\circ \overline{U_X^\sigma(t)},\qquad U_X^\sigma(t)=e^{-itA(X^\sigma)},$$9 time under adjacency scaling by $\circ$0. Without this scaling, the expected time is

$\circ$1

where

$\circ$2

The stopping rule evolves from an arbitrary non-conical vertex for time $\circ$3, measures the cone vertex, repeats until success with probability $\circ$4, and then evolves from the cone for time $\circ$5 to obtain the uniform distribution. The resulting corollary is that for every $\circ$6 there exists a unitary signing $\circ$7 such that $\circ$8 has probabilistic uniform mixing, in contrast with the non-chiral theorem that $\circ$9 has instantaneous uniform mixing if and only if $G(V,E)$0 (Levine et al., 6 May 2026).

The required signings can be constructed explicitly. For odd $G(V,E)$1, one uses the $G(V,E)$2 circulant

$G(V,E)$3

which is Hermitian and satisfies $G(V,E)$4 with $G(V,E)$5 simple. For even $G(V,E)$6, one uses a block skew-circulant built from roots of unity. These constructions provide a chiral route around complete-graph no-go results (Levine et al., 6 May 2026).

Chiral mixing also appears on Hamming graphs. There exists a signing of $G(V,E)$7 that is switching equivalent to $G(V,E)$8, and $G(V,E)$9 has local uniform mixing at

$X=(V,E)$00

from the cone vertex. By Cartesian closure, for all $X=(V,E)$01 there exists a signing of

$X=(V,E)$02

with instantaneous uniform mixing at the same time

$X=(V,E)$03

The comparison quoted in the same source is

$X=(V,E)$04

so the oriented $X=(V,E)$05 mixes strictly faster than the other listed families (Levine et al., 6 May 2026).

Average uniform mixing behaves even more sharply. If a $X=(V,E)$06-circulant has distinct eigenvalues, then

$X=(V,E)$07

The proof uses the Fourier basis, in which each spectral idempotent $X=(V,E)$08 is rank one and satisfies

$X=(V,E)$09

Infinite families include oriented odd cycles $X=(V,E)$10 with $X=(V,E)$11 odd and transitive tournaments arising from skew-circulants. This is described as a chiral violation of Godsil’s no-go theorem, since in the non-chiral setting the only graph with average uniform mixing is $X=(V,E)$12 (Levine et al., 6 May 2026).

A parallel line of work uses the quantum-classical distance to optimize phases. On complete graphs, maximizing this quantity identifies optimal Hermitian matrices $X=(V,E)$13 whose evolution from a localized state $X=(V,E)$14 is

$X=(V,E)$15

The walk reaches the flat state

$X=(V,E)$16

at

$X=(V,E)$17

which saturates the Mandelstam–Tamm quantum speed limit and yields instantaneous uniform mixing for all $X=(V,E)$18 (Frigerio et al., 2021).

5. Zero transfer, suppression, and swift departure

Zero transfer is the condition

$X=(V,E)$19

so that a connected network forbids amplitude flow between designated vertices by exact destructive interference. One route to zero transfer uses merged path graphs. If $X=(V,E)$20 identical branches of length $X=(V,E)$21 meet at a hub and the initial state is the equal-weight superposition on the $X=(V,E)$22 outer vertices, then the amplitude at the hub is

$X=(V,E)$23

Hence the sufficient and necessary cancellation condition is

$X=(V,E)$24

This produces all-time suppression at the hub without requiring time-dependent control (Sett et al., 2019).

An explicit four-branch example takes accumulated phases $X=(V,E)$25, $X=(V,E)$26, $X=(V,E)$27, and $X=(V,E)$28, whose sum is zero, so the hub probability vanishes for all $X=(V,E)$29. Another basic example is the even cycle $X=(V,E)$30 with a single $X=(V,E)$31-phase on one edge, for which two equal-magnitude paths interfere destructively and $X=(V,E)$32 for all $X=(V,E)$33. The same source emphasizes that breaking time-reversal symmetry is not necessary for zero transfer, because this $X=(V,E)$34 example is bipartite and still exhibits exact suppression (Sett et al., 2019).

A spectral characterization is available for even cycles. Let $X=(V,E)$35 carry phases whose product around the cycle equals $X=(V,E)$36, equivalently

$X=(V,E)$37

Then the Hermitian adjacency satisfies

$X=(V,E)$38

where $X=(V,E)$39 is the Chebyshev polynomial of the first kind, and there is zero transfer between antipodal vertices $X=(V,E)$40 and $X=(V,E)$41 for all $X=(V,E)$42. This family provides a graph-theoretic description of zero transfer in terms of cycle holonomy rather than branch phasor cancellation (Chaves et al., 2022).

A distinct but related phenomenon is swift departure from a high-degree starting vertex. For a cone graph with apex $X=(V,E)$43 over a base $X=(V,E)$44 of order $X=(V,E)$45, if chiral phases on the base satisfy

$X=(V,E)$46

then the reduced adjacency on $X=(V,E)$47 is

$X=(V,E)$48

and the return probability is

$X=(V,E)$49

The first zero occurs at

$X=(V,E)$50

and this saturates the Mandelstam–Tamm bound because $X=(V,E)$51 for any chiral Hamiltonian compatible with a degree-$X=(V,E)$52 starting vertex. Even-degree graphs, cubic graphs with a perfect matching, complete graphs, and complete bipartite graphs admit such phase assignments (Frigerio et al., 2022).

The same paper proves a no-go theorem for Laplacian-type generators. If $X=(V,E)$53 is a cone over $X=(V,E)$54 and $X=(V,E)$55 is the largest degree in $X=(V,E)$56, then for any chiral Laplacian,

$X=(V,E)$57

Thus adjacency-type chirality can cure sedentarity, but Laplacian-type chirality generically cannot when the base is sparse compared with the apex degree (Frigerio et al., 2022).

6. Spectral, hydrodynamic, and Floquet extensions

On the infinite one-dimensional lattice with complex next-nearest-neighbor hopping, long-time dynamics admits a hydrodynamic description. The large-$X=(V,E)$58 probability and current are

$X=(V,E)$59

and the cumulative probability $X=(V,E)$60 and cumulative current $X=(V,E)$61 obey

$X=(V,E)$62

More generally, the scaled cumulative moments satisfy the infinite hierarchy

$X=(V,E)$63

The bulk is ballistic, but the causal structure changes with $X=(V,E)$64 and $X=(V,E)$65. For $X=(V,E)$66, increasing $X=(V,E)$67 past a critical $X=(V,E)$68 produces a transition from one cone to two nested cones; for $X=(V,E)$69, the critical point is $X=(V,E)$70, above which two partially overlapping cones appear because two left-moving maximal fronts become degenerate (Bhandari et al., 2021).

The front scaling is universal but order dependent. Near a $X=(V,E)$71-th order extremal front, the probability density obeys

$X=(V,E)$72

For first-order fronts one gets the Airy exponents

$X=(V,E)$73

while a third-order front gives

$X=(V,E)$74

This identifies a continuous-time chiral quantum walk as a setting in which chirality affects not only transport direction but also edge universality classes (Bhandari et al., 2021).

A topological extension arises in periodically driven systems. For a local chiral drive with half-step operator

$X=(V,E)$75

the two chiral-symmetric timeframes are

$X=(V,E)$76

The connected components of essentially local, essentially gapped half-step operators are classified by five integers,

$X=(V,E)$77

and a half-step is realizable by a continuous local Hamiltonian if and only if

$X=(V,E)$78

This continuizability criterion separates genuinely continuous-time chiral evolutions from half-step operators that require discrete-time or nonlocal structure (Cedzich et al., 2020).

A nonlocal Floquet realization makes the connection explicit. If

$X=(V,E)$79

then the effective Floquet Hamiltonian is

$X=(V,E)$80

Choosing the branch with $X=(V,E)$81 yields the nonlocal continuous-time Hamiltonian

$X=(V,E)$82

which is Hermitian but infinitely long-ranged. The corresponding group velocity is strictly positive,

$X=(V,E)$83

This model was realized on the IBM quantum computer platform, and the reported chiral wave packets were robust against external perturbations but had a finite lifetime due to intrinsic device errors (Bark et al., 2023).

7. Limitations, design principles, and open problems

Several limitations recur across the literature. First, chirality requires gauge-invariant loop structure: on trees all phases are gauge-trivial, so transport probabilities are identical to those of the undirected walk (Annoni et al., 2023). Second, exact phenomena such as zero transfer and perfect phase-cancellation are sensitive to decoherence. In the three-vertex chain used as a decoherence probe, the ideal chiral Hamiltonian

$X=(V,E)$84

gives $X=(V,E)$85 for all $X=(V,E)$86, but the inclusion of Lindblad decoherence lifts the zero and drives the long-time probability toward $X=(V,E)$87 (Sett et al., 2019).

Input purity is another constraint. On the triangular chain, the mixed initial state

$X=(V,E)$88

shows monotonically decreasing fidelity as $X=(V,E)$89 is reduced from $X=(V,E)$90 to $X=(V,E)$91, so entanglement routing is sensitive to coherence in the injection manifold (Sağlam et al., 2021). A similar trade-off appears in routing and transport more broadly: early-time peaks are typically more robust to phase fluctuations, whereas later peaks may be higher but sharper in parameter space (Bottarelli et al., 2023).

From a spectral viewpoint, chiral advantages are not universal. Average uniform mixing on abelian circulants relies on distinct eigenvalues and rank-one Fourier idempotents, but nonabelian Cayley graphs retain repeated eigenvalues by representation theory, which forbids average uniform mixing even with chirality (Levine et al., 6 May 2026). Likewise, Laplacian-type chirality does not generally overcome high-degree localization, whereas adjacency-type chirality can be quantum-speed-limit optimal (Frigerio et al., 2022).

The constructive side is comparatively systematic. For rapid departure or complete-graph mixing, one seeks signings with

$X=(V,E)$92

and $X=(V,E)$93 simple. For average uniform mixing, one targets $X=(V,E)$94-circulants with simple spectrum. For directional routing, one engineers gauge-invariant loop fluxes so that interference phases satisfy explicit scattering constraints such as

$X=(V,E)$95

For topology-dependent phase optimization, the quantum-classical distance has been proposed as a figure of merit because on cycles and switches it correlates with transport-oriented localization, whereas on complete graphs it correlates with mixing-oriented delocalization (Frigerio et al., 2021).

Open problems remain broad and concrete. Recent work asks for the existence of high-diameter trees with probabilistic uniform mixing under suitable chiral signings and stopping rules, the optimization of Las Vegas uniform mixing via stopping-time distributions, and Monte Carlo protocols that achieve $X=(V,E)$96-uniform distributions in bounded time (Levine et al., 6 May 2026). On the swift-walk side, necessary and sufficient conditions are still incomplete beyond the even-degree, cubic, complete, and complete-bipartite families already characterized (Frigerio et al., 2022). A plausible implication is that continuous-time chiral quantum walks now occupy a dual role: they are both a subject in spectral graph theory and a design framework for interference-controlled transport, mixing, and routing on quantum networks.