Perfect tunneling in equilateral triple-well potentials on graphs: the Lin–Lippner–Yau conjecture and the classification of triple-well tunneling

A quantum particle on a finite connected graph, evolving under the discrete Schrödinger equation with Hamiltonian $H=\Delta-QW$ ($\Delta$ the symmetrized graph Laplacian, $QW$ a deep potential supported on a few 'well' vertices), tunnels between wells in a way completely governed — in the double-well case — by the local symmetry of the graph around the wells: Lin, Lippner and Yau proved that tunneling between two wells is perfect, partial, or absent according to the cospectrality (agreement of random-walk return probabilities) of the wells and their distance, with tunneling time $\Theta(Q^{d-1})$. The triple-well case is only partially understood. For three equilateral, pairwise cospectral wells with equal couplings $c_{xy}=c_{xz}$, they proved an arithmetic trichotomy governed by $s=\sqrt{c_{yz}^2+8c_{xz}^2}$: tunneling from $y$ to $z$ is perfect when $s$ is irrational (though slower than the double-well scale), while in the rational case they could only prove that any perfect tunneling requires time at least of order $Q^a$ and left whether it occurs at all undetermined. Their Conjecture 1.6 asserts perfect tunneling always occurs when $c_{xy}=c_{xz}\neq c_{yz}$; they showed it can fail only if some graph realizes a delicate exceptional configuration — rational $s$ together with a $Q$-independent lowest-eigenvalue-gap ratio — which they believe unrealizable but could not rule out. They further proved that tunneling coefficients can be nonlocally unstable (altered by one-edge changes arbitrarily far from the wells) and restricted their classification to cospectral wells to avoid case analysis.

Background

The discrete model studied here is the finite-graph analogue of a quantum particle in a deep multi-well potential. Let $G=(X,E)$ be a finite connected graph with vertex degrees $d_v$. The symmetrized Laplacian is the matrix with entries $\Delta(x,x)=1$, $\Delta(x,y)=-1/\sqrt{d_xd_y}$ for an edge $xy\in E$, and $0$ otherwise; without a potential, the evolution $\varphi_t=e^{it\Delta}\varphi_0$ is the continuous-time quantum walk, the quantum analogue of the simple random walk introduced by Farhi and Gutmann in the context of quantum computation. Adding a potential $V:X\to[0,\infty)$ gives the discrete Schrödinger equation $-i\,d\varphi_t/dt=H\varphi_t$ with Hamiltonian $H=\Delta-V$, where $|\varphi_t(x)|^2$ is the probability of finding the particle at vertex $x$ at time $t$. Lin, Lippner and Yau study the deep-well limit: $V=Q\cdot W$ with $W$ fixed and $Q\to\infty$, so the vertices where $W$ attains its global maximum (normalized to $1$) become arbitrarily deep wells. A potential is simple if it is supported exactly on the wells. This mirrors the semiclassical analysis of double-well Schrödinger operators on $\mathbb{R}^d$, where Simon proved that the splitting between the two lowest eigenvalues is exponentially small in the semiclassical parameter and governs tunneling; on graphs the barrier instead becomes 'invisible' in a combinatorially quantified way, a discrete echo of Klein's paradox that the higher the barrier, the more transparent it becomes to the particle.

The quantitative target is the tunneling coefficient. If the particle starts at a well $x$, then

$$TC(x,y)=\liminf_{Q\to\infty}\;\sup_{t\ge 0}\;|\varphi_t(y)|^2$$

(the source's Definition 1.2). Perfect (asymptotic) tunneling occurs if $TC(x,y)=1$, partial tunneling if $0<TC(x,y)<1$, and no tunneling if $TC(x,y)=0$. Tunneling 'within time $f(Q)$' replaces $\sup_{t\ge0}$ by $\sup_{0\le t\le f(Q)}$, and one asks how the shortest such time scale grows with $Q$. The mechanism is eigenvalue splitting: the lowest eigenvalues of $H$ cluster near $1-Q$, and their differences are the beat frequencies that transfer amplitude between wells.

The governing local-symmetry notion is cospectrality. Two vertices $x,y$ are $m$-cospectral if the return probabilities of the simple random walk agree up to time $m$, that is $PR(x,k)=PR(y,k)$ for $0\le k\le m$, where $PR(x,k)$ is the probability that the walk started at $x$ is back at $x$ at time $k$; the cospectrality $\mathrm{co}(x,y)$ is the maximal such $m$, a quantitative measure of how similar the graph looks around the two vertices. For a simple double-well potential, Lin–Lippner–Yau completely classified the behavior (their Theorem 1.3): writing $d=d(x,y)$, tunneling between the two wells is perfect if $\mathrm{co}(x,y)\ge d$, partial if $\mathrm{co}(x,y)=d-1$, and absent if $\mathrm{co}(x,y)<d-1$; when it occurs, the tunneling time is $\Theta(Q^{d-1})$.

For three wells $x,y,z$ with mutual distances $a=d(x,y)\le b=d(x,z)\le c=d(y,z)$, the paper restricts to sufficiently cospectral configurations — the wells are assumed pairwise $c$-cospectral, that is cospectral up to time $c=d(y,z)$, the largest mutual distance — explicitly 'to avoid excessive case analysis'. The behavior is then governed by the coupling constants

$$c_{uv}\;=\;\sum_{P:\,u\to v,\;|P|=d(u,v)}\;\prod_{j=0}^{|P|-1}\frac{1}{\sqrt{d_{x_j}d_{x_{j+1}}}},$$

sums over all geodesics (shortest paths) between the wells of products of inverse geometric-mean degrees along the path. Their Theorem 1.4 settles the cases $a<b$ and $a=b<c$: in the latter, a particle started at $y$ tunnels perfectly to $z$ if and only if $c_{xy}=c_{xz}$, in time $\Theta(Q^{a-1})$ — a third well can substantially accelerate tunneling between the two more distant wells. The analysis runs through a $3\times3$ limit interaction matrix $M$, with zero diagonal and off-diagonal entries $c_{uv}$, whose eigenvectors are the limits of the low-energy eigenfunctions restricted to the wells; in the equilateral case with $c_{xy}=c_{xz}$ its eigenvalues are $\mu_{1,2}=(c_{yz}\pm\sqrt{c_{yz}^2+8c_{xz}^2})/2$ and $\mu_3=-c_{yz}$, and its eigenvalue ratio

$$\gamma=\frac{\mu_1-\mu_3}{\mu_1-\mu_2}=\frac{3c_{yz}+\sqrt{c_{yz}^2+8c_{xz}^2}}{2\sqrt{c_{yz}^2+8c_{xz}^2}}$$

controls the relative gaps between the three smallest eigenvalues $\lambda_1(Q)\le\lambda_2(Q)\le\lambda_3(Q)$ of $H$; perfect tunneling on the fast scale $O(Q^{a-1})$ requires $\gamma$ to be a rational of odd-over-even type (in lowest terms $p/q$ with $p$ odd, $q$ even), a lattice-approximation criterion, and their Claim 3.6 shows such rationals cannot occur in the equilateral triple-well case.

The equilateral case $a=b=c$ (their Theorem 1.5) exhibits an arithmetic trichotomy for tunneling from $y$ to $z$: (i) if $c_{xy}\neq c_{xz}$, only partial tunneling occurs; (ii) if $c_{xy}=c_{xz}$ and $s:=\sqrt{c_{yz}^2+8c_{xz}^2}$ is irrational, tunneling is perfect, but any tunneling time $f$ must satisfy $Q^{a-1}=o(f(Q))$ — strictly slower than the double-well scale; (iii) if $s$ is rational (in particular when $c_{xy}=c_{xz}=c_{yz}$), the paper proves only that if perfect tunneling occurs then $Q^a=O(f(Q))$: whether it occurs at all is left undetermined. In the fully symmetric sub-case, an order-3 symmetry of the graph permuting the three wells forces $TC(y,z)=4/9$ (their Lemma 3.8), so no perfect tunneling. Their Conjecture 1.6 asserts that in case (iii) with $c_{yz}\neq c_{xz}$ perfect tunneling always occurs, and their Remark 3.7 shows this is the only remaining uncertainty: the conjecture can fail only if some finite graph realizes the exceptional configuration in which $c_{xy}=c_{xz}\neq c_{yz}$, $s$ is rational, and at the same time the eigenvalue-gap ratio $(\lambda_1(Q)-\lambda_3(Q))/(\lambda_1(Q)-\lambda_2(Q))$ is independent of $Q$ — a configuration the authors believe no graph realizes but cannot rule out.

Finally, their Theorem 1.7 reveals a nonlocal instability with no continuum analogue: in a graph made of three equal-length paths joined at a common vertex with the wells at the three inner endpoints, $TC(y,z)=4/9$, and adding a single edge arbitrarily far from the wells changes $TC(y,z)$ to $1$. Hence the tunneling coefficient 'is not predictable from any fixed neighborhood of the wells'. Together these results leave open the complete characterization of triple-well tunneling: the arithmetic exceptional case of the equilateral conjecture, the degenerate equal-coupling case, the classification without the cospectrality hypothesis, and the boundary between locally predictable and nonlocally unstable behavior.

Adjacent literature: Colin de Verdière reformulated the paper's eigenvalue asymptotics through an interaction-matrix construction adapted from semiclassical multi-well theory, confirming the method but not addressing the conjecture; the quantum state transfer literature (Kempton–Lippner–Yau; Lippner–Shi) studies exact and asymptotic state transfer with vertex potentials, building on the double-well results but not the triple-well questions.

Problem Statement

Complete the characterization of deep-well tunneling for triple-well potentials on finite graphs as left open by Lin–Lippner–Yau. Throughout, $G=(X,E)$ is a finite connected graph with vertex degrees $d_v$; $\Delta$ is its symmetrized Laplacian, with $\Delta(x,x)=1$, $\Delta(x,y)=-1/\sqrt{d_xd_y}$ when $xy\in E$, and $0$ otherwise; $W$ is the simple triple-well potential, equal to $1$ on three distinguished vertices (the wells $x,y,z$) and $0$ elsewhere; and $H=\Delta-QW$ with $Q\to\infty$. A particle started at the well $y$ (initial state $\varphi_0$ with $|\varphi_0(y)|=1$) evolves by $\varphi_t=e^{itH}\varphi_0$, and $|\varphi_t(v)|^2$ is the probability of finding it at vertex $v$ at time $t$. The tunneling coefficient from $y$ to $z$ is

$$TC(y,z)=\liminf_{Q\to\infty}\sup_{t\ge 0}|\varphi_t(z)|^2,$$

called perfect tunneling if $TC(y,z)=1$, partial if $0<TC(y,z)<1$, and absent if $TC(y,z)=0$; tunneling within time $f(Q)$ replaces $\sup_{t\ge 0}$ by $\sup_{0\le t\le f(Q)}$. Let $\lambda_1(Q)\le\lambda_2(Q)\le\lambda_3(Q)$ be the three smallest eigenvalues of $H$. The wells are pairwise $c$-cospectral if the return probabilities of the simple random walk agree pairwise, $PR(u,k)=PR(v,k)$ for every pair $u,v$ of wells and all $0\le k\le c$, where $PR(u,k)$ is the probability that the walk started at $u$ is back at $u$ at time $k$ and $c$ is the largest mutual distance of the wells. The coupling between wells $u,v$ is the geodesic path-weight sum

$$c_{uv}=\sum_{P:\,u\to v,\;|P|=d(u,v)}\;\prod_{j=0}^{|P|-1}\frac{1}{\sqrt{d_{x_j}d_{x_{j+1}}}},$$

the sum over all shortest paths $P=x_0x_1\cdots x_{|P|}$ from $u$ to $v$ of the product of inverse geometric-mean degrees along the path.

(1) The equilateral perfect-tunneling conjecture (the sharply bounded core). Assume the wells are pairwise $c$-cospectral and equilateral, $d(x,y)=d(x,z)=d(y,z)=a$ (so $c=a$). Prove, or refute, Conjecture 1.6 of the source: if $c_{xy}=c_{xz}\neq c_{yz}$, then $TC(y,z)=1$ (perfect asymptotic tunneling from $y$ to $z$). By the source's Theorem 1.5, Claim 3.6 and Remark 3.7, the conjecture fails if and only if there exists a finite graph whose equilateral, pairwise $c$-cospectral wells satisfy simultaneously

$$c_{xy}=c_{xz}\neq c_{yz},\qquad \sqrt{c_{yz}^2+8c_{xz}^2}\in\mathbb{Q},\qquad \frac{\lambda_1(Q)-\lambda_3(Q)}{\lambda_1(Q)-\lambda_2(Q)}\ \text{independent of}\ Q.$$

A proof that no such graph exists establishes the conjecture; an explicit such graph refutes it — by the source's Remark 3.5 there is then no perfect tunneling from $y$ to $z$ at all, that is $TC(y,z)<1$.

(2) The degenerate equal-coupling case. In the same equilateral setting with $c_{xy}=c_{xz}=c_{yz}$ (so $\sqrt{c_{yz}^2+8c_{xz}^2}=3c_{xz}$ is rational and the limit interaction matrix has a double eigenvalue), determine whether $TC(y,z)=1$ can occur when the graph lacks an order-3 symmetry permuting the wells; the source proves only the conditional statement that any perfect tunneling must satisfy $Q^a=O(f(Q))$, and that order-3 symmetry forces $TC(y,z)=4/9$.

(3) The classification beyond cospectrality. Extend the classification of triple-well tunneling to configurations without the pairwise $c$-cospectrality hypothesis: for every finite graph and every triple of wells, characterize — in terms of effectively checkable graph data such as distances, cospectrality degrees, and geodesic path-weight sums — whether tunneling between each ordered pair of wells is perfect, partial, or absent, and on what $Q$-time-scale it occurs.

(4) Local predictability versus nonlocal instability. Within that classification, decide for each classification datum — the tunneling outcome (perfect, partial, or absent) of each ordered pair, its $Q$-time-scale, and the coupling and cospectrality data — whether it is locally predictable, meaning fixed by the ball of radius $r$ around the wells uniformly in $Q$ and in the graph for some fixed $r$, or nonlocally unstable, meaning not fixed by any bounded neighborhood, and identify the required radius (or prove that no bounded radius suffices). The source's Theorem 1.7 instantiates the unstable side, where a one-edge modification arbitrarily far from the wells moves $TC(y,z)$ from $4/9$ to $1$.

Sub-questions (1)–(2) address the equilateral, pairwise $c$-cospectral configurations, whose boundary is sharp. Sub-question (3) ranges over every finite graph and every triple of wells, hence over a strict superset of the configurations of (1)–(2).

References

1. Yong Lin, Gábor Lippner, Shing-Tung Yau, Quantum Tunneling on Graphs, Communications in Mathematical Physics 311(1) (2012) 113–132. DOI: 10.1007/s00220-012-1453-8; arXiv:1101.2660. https://doi.org/10.1007/s00220-012-1453-8
2. Barry Simon, Semiclassical Analysis of Low Lying Eigenvalues, II. Tunneling, Annals of Mathematics 120(1) (1984) 89–118. DOI: 10.2307/2007072. https://doi.org/10.2307/2007072
3. Edward Farhi, Sam Gutmann, Quantum computation and decision trees, Physical Review A 58(2) (1998) 915–928. DOI: 10.1103/PhysRevA.58.915. https://doi.org/10.1103/PhysRevA.58.915
4. Oskar Klein, Die Reflexion von Elektronen an einem Potentialsprung nach der relativistischen Dynamik von Dirac, Zeitschrift für Physik 53 (1929) 157–165. DOI: 10.1007/BF01339716. https://doi.org/10.1007/BF01339716
5. Yves Colin de Verdière, Tunneling on graphs: an approach "à la Helffer–Sjöstrand", preprint (2018). arXiv:1809.04973; DOI: 10.48550/arXiv.1809.04973. https://arxiv.org/abs/1809.04973
6. Andrea Casaccino, Seth Lloyd, Stefano Mancini, Simone Severini, Quantum state transfer through a qubit network with energy shifts and fluctuations, International Journal of Quantum Information 7(8) (2009) 1417–1427. DOI: 10.1142/S0219749909006085. https://doi.org/10.1142/S0219749909006085
7. Chris Godsil, State transfer on graphs, Discrete Mathematics 312(1) (2012) 129–147. DOI: 10.1016/j.disc.2011.06.032. https://doi.org/10.1016/j.disc.2011.06.032
8. Mark Kempton, Gábor Lippner, Shing-Tung Yau, Perfect state transfer on graphs with a potential, Quantum Information and Computation 17(3–4) (2017) 303–327. DOI: 10.26421/QIC17.3-4-7; arXiv:1611.02093. https://doi.org/10.26421/QIC17.3-4-7
9. Gábor Lippner, Yujia Shi, Quantifying state transfer strength on graphs with involution, Quantum Information Processing 23(5) (2024), article 166. DOI: 10.1007/s11128-024-04362-5; arXiv:2310.08690. https://doi.org/10.1007/s11128-024-04362-5
10. Giacomo Di Gesù, Spectral Analysis of Discrete Metastable Diffusions, Communications in Mathematical Physics 402(1) (2023) 543–580. DOI: 10.1007/s00220-023-04729-2; arXiv:2102.10053. https://doi.org/10.1007/s00220-023-04729-2
11. Matthias Keller, Lorenzo Pettinari, Christiaan J. F. van de Ven, Coupling of the continuum and semiclassical limit. Part I: convergence of eigenvalues, preprint (2026). arXiv:2602.23156. https://arxiv.org/abs/2602.23156
12. Yujia Shi, Continuous-Time Quantum State Transfer with a Generalized Laplacian, preprint (2025). arXiv:2509.05454. https://arxiv.org/abs/2509.05454
13. Ce Wang, Continuous-Time Quantum Walk on Locally Infinite Graph, preprint (2026). arXiv:2602.23970. https://arxiv.org/abs/2602.23970
14. Gábor Lippner, Yujia Shi, Strong quantum state transfer on graphs via loop edges, Linear Algebra and its Applications 704 (2025) 77–91. DOI: 10.1016/j.laa.2024.10.008. https://doi.org/10.1016/j.laa.2024.10.008
15. Gábor Lippner, Yujia Shi, Fast state transfer via loop weights, preprint (2026). arXiv:2601.20237. https://arxiv.org/abs/2601.20237
16. Xiaohong Zhang, Combinatorial and algebraic aspects of quantum state transfer (2019).