---
title: Hamiltonian Bypass in Graph Theory
url: https://www.emergentmind.com/topics/hamiltonian-bypass
type: topic
---

# Hamiltonian Bypass in Graph Theory

In graph theory, a Hamiltonian bypass is a spanning subdigraph obtained from a Hamiltonian cycle by reversing exactly one arc; equivalently, it may be described as a Hamiltonian path whose initial vertex dominates the terminal vertex [1404.5780], [2507.15435]. The notion refines Hamiltonicity by asking not merely for a directed cycle through all vertices, but for a Hamiltonian structure with a prescribed one-arc reversal. Within the supplied literature, the term also appears in broader Hamiltonian settings as an interpretive label for constructions that “bypass” obstacles in an energy landscape or an adiabatic evolution, although its canonical technical meaning is the digraph-theoretic one [2206.14072], [1501.02811], [2408.06916].

## 1. Definition, notation, and equivalent formulations

For a digraph $D$, a Hamiltonian cycle is a directed cycle that visits every vertex exactly once. If
$$
C=x_1x_2\cdots x_nx_1
$$
is a Hamiltonian cycle, then a Hamiltonian bypass is obtained by reversing exactly one arc, for example replacing $x_nx_1$ by $x_1x_n$ while keeping the remaining cycle arcs [1404.5780]. The paper "On Hamiltonian Bypasses in one Class of Hamiltonian Digraphs" denotes such a bypass by
$$
D(n;2)=[x_1\to x_n;\;x_1x_2\cdots x_n],
$$
that is, the subdigraph consisting of the $(n-1)$-arc directed path $x_1x_2\cdots x_n$ together with the reversed arc $x_1\to x_n$ [1404.5780].

A later paper adopts an equivalent path-based formulation: a Hamiltonian bypass is a Hamiltonian path $P=x_1x_2\cdots x_p$ such that the initial vertex $x_1$ dominates the terminal vertex $x_p$; equivalently, $P$ together with $x_1x_p$ forms a Hamiltonian cycle [2507.15435]. These formulations are compatible: the reversed-arc viewpoint emphasizes derivation from a Hamiltonian cycle, while the path viewpoint emphasizes endpoint domination.

The standard degree notation in this literature is
$$
d^+(v)=|N^+(v)|,\qquad d^-(v)=|N^-(v)|,\qquad d(v)=d^+(v)+d^-(v),
$$
with minimum out-degree $\delta^+(D)$, minimum in-degree $\delta^-(D)$, and minimum semi-degree $\delta^0(D)=\min\{\delta^+(D),\delta^-(D)\}$ [1404.5780]. Two vertices are non-adjacent if neither arc between them is present, and a common in-neighbour of $x,y$ is a vertex $z$ such that $zx,zy\in A(D)$ [1404.5780].

The significance of the concept lies in its placement between Hamiltonian cycles and more specialized spanning configurations. The supplied graph-theoretic papers consistently treat bypass existence as a stronger structural guarantee than Hamiltonicity alone [1404.5780], [1405.0002], [2507.15435].

## 2. Degree conditions with common in-neighbours

A central line of work studies Hamiltonian bypasses under the Bang-Jensen–Gutin–Li condition
$$
d(x)+d(y)\ge 2n-1
\quad\text{and}\quad
\min\{d(x),d(y)\}\ge n-1
$$
for every pair of non-adjacent vertices $x,y$ with a common in-neighbour [1404.5780]. Bang-Jensen, Gutin, and Li proved that a strong digraph satisfying this condition is Hamiltonian [1404.5780]. Darbinyan and Karapetyan then showed that if, in addition, $\delta^0(D)\ge 2$, then either the digraph contains a pre-Hamiltonian cycle of length $n-1$, or $n$ is even and the digraph is isomorphic to the symmetric complete bipartite digraph $K_{n/2,n/2}$ or that digraph minus one arc [1404.5780].

The main contribution of "On Hamiltonian Bypasses in one Class of Hamiltonian Digraphs" is a bypass strengthening of those earlier Hamiltonicity and long-cycle results. Its Theorem 12 states that if $D$ is a strong digraph of order $n\ge 4$, satisfies the same Bang-Jensen–Gutin–Li condition, has minimum out-degree at least two, and minimum in-degree at least three, then $D$ contains a Hamiltonian bypass [1404.5780]. The theorem therefore upgrades a Hamiltonian existence statement to a specific spanning subdigraph with one reversed arc.

The proof is organized around a pre-Hamiltonian cycle $C$ of length $n-1$ and the unique vertex $y$ outside $C$ [1404.5780]. Under the no-bypass assumption, Lemma 4 imposes stringent adjacency restrictions: $y$ cannot be adjacent with both members of any consecutive pair on $C$, forcing bounds such as $d^+(y)\le (n-1)/2$ and $d(y)\le n-1$ [1404.5780]. Lemmas 6 through 9 show that certain local configurations involving $y$ and consecutive vertices of $C$ immediately yield a bypass or a “good cycle,” with the minimum in-degree and out-degree assumptions used explicitly [1404.5780]. Lemmas 10 and 11 exclude shortcut arcs of the form $x_{i+1}x_{i-1}$ in a counterexample without a bypass, because such arcs lead to contradictions with the degree condition [1404.5780].

The argument then proceeds by induction on a block of consecutive vertices of $C$ that are non-adjacent to $y$, repeatedly applying partner-arc and multi-insertion lemmas to build a spanning path whose concatenation with a suitable arc through $y$ forms a Hamiltonian bypass [1404.5780]. Even the extremal symmetric bipartite case is handled: although a pre-Hamiltonian cycle may fail there when $n-1$ is odd, the paper states that it is still straightforward to check that a Hamiltonian bypass exists under $\delta^+(D)\ge 2$ and $\delta^-(D)\ge 3$ [1404.5780].

The degree assumptions are not merely cosmetic. The paper gives examples $D(7)$ and $D(n)$ in which $\delta^0=1$, the Bang-Jensen–Gutin–Li condition holds vacuously because there is no dominated pair of non-adjacent vertices, and yet no Hamiltonian bypass exists [1404.5780]. The authors conjecture that the minimum in-degree bound can be lowered from $3$ to $2$, but they do not provide a counterexample for $\delta^-(D)=2$ [1404.5780].

## 3. Manoussakis’ condition and the exceptional tournament $T(5)$

A second major sufficient condition is Manoussakis’ Condition $A_0$. For every triple $x,y,z$ with $x$ and $y$ non-adjacent, the condition requires:
- if there is no arc $x\to z$, then
  $$
  d(x)+d(y)+d^+(x)+d^-(z)\ge 3n-2;
  $$
- if there is no arc $z\to x$, then
  $$
  d(x)+d(y)+d^-(x)+d^+(z)\ge 3n-2
  $$
[1405.0002].

Manoussakis proved that every strong digraph satisfying $A_0$ is Hamiltonian [1405.0002]. Darbinyan had earlier shown that such a digraph either contains a pre-Hamiltonian cycle of length $n-1$, or, when $n$ is even, is isomorphic to the complete bipartite digraph with partite sets of size $n/2$ [1405.0002].

The paper "On Hamiltonian Bypasses in Digraphs with the Condition of Y. Manoussakis" strengthens this to a bypass theorem: every strongly connected digraph of order $n>4$ satisfying Condition $A_0$ contains a Hamiltonian bypass, unless it is isomorphic to a specific tournament $T(5)$ of order $5$ [1405.0002]. The exceptional tournament has vertex set $\{x_1,x_2,x_3,x_4,y\}$, a directed $4$-cycle $x_1\to x_2\to x_3\to x_4\to x_1$, the chords $x_1\to x_3$ and $x_2\to x_4$, and the additional arcs $x_3\to y$, $y\to x_2$, $y\to x_4$ [1405.0002]. According to the supplied details, this tournament satisfies $A_0$ but does not contain a Hamiltonian bypass [1405.0002].

The proof again begins with a pre-Hamiltonian cycle $C$ and a unique off-cycle vertex $y$ [1405.0002]. Under the assumption that no bypass exists, Lemma 7 imposes bypass-avoidance constraints: for every consecutive pair on $C$, at most one of the two arcs from $y$ to the pair and at most one to $y$ from the pair can be present; moreover $d^+(y)\le (n-1)/2$, $d^-(y)\le (n-1)/2$, and $d(y)\le n-1$ [1405.0002]. If both $x_k\to y$ and $y\to x_{k+1}$ occur, then $y$ has no arc to any other $x_i$ [1405.0002].

The paper combines these restrictions with path insertion lemmas and a degree trade-off lemma of Manoussakis type. One formulation given in the supplied data is that if $x$ has two distinct non-adjacent pairs $(x,y)$ and $(x,z)$ and one of those pairs has a relatively small degree sum, then the other must have a correspondingly larger one; in particular, if $d(x)+d(y)\le 2n-2$, then $d(x)+d(z)\ge 2n-1$ [1405.0002]. Through a detailed case analysis of all possible adjacency patterns between $y$ and the cycle $C$, the no-bypass assumption is shown to force either a contradiction with $A_0$ or the small exceptional structure $T(5)$ [1405.0002].

This theorem places bypass existence strictly above pre-Hamiltonian cycle existence in the $A_0$ class. The complete bipartite digraph remains extremal for the earlier pre-Hamiltonian result, but is not exceptional for bypasses: the paper states that it still contains a Hamiltonian bypass [1405.0002].

## 4. One-exception high-degree digraphs and balanced bipartite digraphs

The 2025 paper "On Hamiltonian bypasses in digraphs and bipartite digraphs" studies a different high-degree framework [2507.15435]. Let $D$ be a $2$-strong digraph of order $p\ge 3$, and fix a vertex $z$ such that every vertex other than $z$ has degree at least $p$:
$$
d(w)\ge p\qquad\text{for all }w\in V(D)\setminus\{z\}
$$
[2507.15435]. The paper introduces a conjecture asking for the smallest integer $k$ such that $d(z)\ge k$ forces a Hamiltonian bypass, and proves that one may take $k\le \lfloor (p-1)/3\rfloor+1$ [2507.15435].

Its first principal theorem states: if either $D$ is Hamiltonian or
$$
d(z)>\frac{p-1}{3},
$$
then $D$ contains a Hamiltonian bypass [2507.15435]. This improves an earlier threshold $d(z)>0.4(p-1)$ in the same one-exception large-degree setting [2507.15435]. The proof uses a long cycle $R$ of length $p-1$ containing all high-degree vertices, then derives adjacency inequalities around $z$ under the no-bypass assumption:
$$
d^-(z,\{x_i\})+d^+(z,\{x_{i+1},x_{i+2},x_{i+3},x_{i+4},x_{i+5}\})\le 1,
$$
$$
d^+(z,\{x_j\})+d^-(z,\{x_{j-1},x_{j-2},x_{j-3},x_{j-4},x_{j-5}\})\le 1
$$
for all appropriate indices [2507.15435]. Summing yields
$$
d(z)+5d^+(z)\le p-1,\qquad 5d^-(z)+d^+(z)\le p-1,
$$
hence
$$
6d(z)\le 2p-2,
$$
so $d(z)\le (p-1)/3$, contradicting the hypothesis [2507.15435].

The same paper also proves a bipartite theorem. Let $B(X,Y)$ be a strong balanced bipartite digraph of order $2a\ge 6$ with partite sets $X=\{x_1,\dots,x_a\}$ and $Y=\{y_1,\dots,y_a\}$. If for every pair $u\in X$, $v\in Y$ with $uv\notin A(B)$,
$$
d^+(u)+d^-(v)\ge a+1,
$$
then $B$ contains a Hamiltonian bypass [2507.15435]. The lower bound $a+1$ is sharp: the digraph $B_8$ of order $8$ satisfies the weaker condition $d^+(u)+d^-(v)\ge a$ for every missing cross-part arc and has no Hamiltonian bypass [2507.15435]. By contrast, the small exceptional digraph $B_6$ satisfies the sharp condition $A_1$, is not Hamiltonian, but still contains a Hamiltonian bypass [2507.15435].

These results link bypasses to classical high-degree Hamiltonicity theory. The paper recalls, for example, that every strong digraph of order $p\ge 3$ with minimum degree at least $p$ contains a $D(p,2)$, described there as a directed Hamiltonian bypass structure [2507.15435]. It also formulates broader conjectures, including the global conjecture that if $p-1$ vertices of a $2$-strong digraph of order $p$ have degrees at least $p$, then the digraph contains a Hamiltonian bypass with no degree requirement on the remaining vertex [2507.15435].

## 5. Structural techniques, extremal examples, and open questions

Across the digraph literature, bypass proofs rely on a recurring toolkit: cycle-extension lemmas, Bondy–Thomassen-style path insertion, and the multi-insertion lemma of Bang-Jensen–Gutin [1404.5780], [1405.0002], [2507.15435]. The operative idea is to locate “partners,” namely arcs $x_ix_{i+1}$ on a long path or cycle that permit insertion of an external vertex or short path through arcs $x_i\to y$ and $y\to x_{i+1}$ [1404.5780]. When sufficiently many such partners exist, repeated insertion yields a spanning path that closes to a bypass.

A second common theme is the analysis of a unique off-cycle vertex relative to a pre-Hamiltonian cycle. In both the Bang-Jensen–Gutin–Li and Manoussakis settings, a no-bypass assumption forces sparse and highly patterned adjacency between the outside vertex and consecutive vertices of the cycle [1404.5780], [1405.0002]. Those local restrictions then interact with global degree lower bounds to produce contradictions.

Extremal and exceptional constructions serve two functions: they show that bypass results are sharper than Hamiltonicity alone, and they delimit necessary hypotheses. The examples $D(7)$ and $D(n)$ in [1404.5780] show that if the minimum semi-degree drops to $1$, then the Bang-Jensen–Gutin–Li condition may hold vacuously while no Hamiltonian bypass exists. The tournament $T(5)$ in [1405.0002] shows that even Manoussakis’ strong triple condition does not force a bypass in every strong digraph. The bipartite example $B_8$ in [2507.15435] shows that the balanced bipartite threshold $a+1$ cannot be reduced to $a$.

Several open problems remain explicit in the supplied data. One is whether the theorem of [1404.5780] remains valid under $\delta^+(D)\ge 2$ and $\delta^-(D)\ge 2$, rather than $\delta^-(D)\ge 3$. Another is the characterization problem posed in [1405.0002]: characterize the digraphs satisfying Bang-Jensen–Gutin–Li or Bang-Jensen–Guo–Yeo Hamiltonicity conditions that do not contain a Hamiltonian bypass. The 2025 paper adds conjectures on one-exception high-degree digraphs, four-vertex degree-sum conditions, Woodall-type thresholds for $D(n,2)$, and a characterization problem for Hamiltonian balanced bipartite digraphs satisfying the weaker condition $A_0$ [2507.15435].

A plausible implication is that the bypass problem has become a fine-grained refinement of degree-based Hamiltonicity theory. The supplied results consistently move from Hamiltonian existence to a more rigid spanning configuration, and the exceptions that remain are small or highly structured.

## 6. Broader non-graph-theoretic uses of the expression

Outside graph theory, the supplied data uses “Hamiltonian bypass” more loosely as a descriptive label for modifying a Hamiltonian so as to evade an obstruction in optimization or adiabatic evolution.

In non-convex optimization, "Non-Convex Optimization by Hamiltonian Alternation" introduces an alternate Hamiltonian
$$
\widetilde H(x)=K(H(x)),
$$
with the explicit choice
$$
\widetilde H(x)=H(x)\bigl(2b-H(x)\bigr),
$$
so that minima of $H$ below an energy threshold $b$ remain minima of $\widetilde H$, while local minima of $H$ above $b$ become local maxima of $\widetilde H$ because
$$
K'(H)=2(b-H)
$$
changes sign at $H=b$ [2206.14072]. The paper describes Hamiltonian alternation as repeatedly minimizing $H$ and $\widetilde H$ in sequence, thereby allowing the optimizer to escape shallow minima. The supplied data states that this can be viewed as a “Hamiltonian bypass” mechanism and that the terminology “corresponds precisely” to that interpretation [2206.14072]. In Sherrington–Kirkpatrick spin-glass experiments with $N=100$ and $N=1000$, the paper used greedy single-spin flips and set $b$ to one percent less than the lowest energy attained so far; for $N=1000$, higher alternation ratios yielded lower energies than pure multi-start, whereas for $N=100$ alternation offered little advantage [2206.14072].

In quantum control, the supplied description of "Shortcuts to nonabelian braiding" explicitly frames counterdiabatic driving as a “Hamiltonian bypass” for finite-time nonabelian braiding [1501.02811]. The original time-dependent Hamiltonian $H_0(t)$ is augmented by a counterdiabatic term
$$
H_{\mathrm{cd}}(t)=i\sum_n[\dot P_n,P_n],
$$
equivalently
$$
H_1(t)= i \sum_n \left[\sum_\alpha |\partial_t\psi_\alpha^n\rangle\langle\psi_\alpha^n|
-\sum_{\alpha\beta}|\psi_\alpha^n\rangle
\langle\psi_\alpha^n|\partial_t\psi_\beta^n\rangle
\langle\psi_\beta^n|\right],
$$
so that the exact finite-time evolution reproduces the ideal adiabatic Wilczek–Zee holonomy within a degenerate manifold [1501.02811]. For the Majorana Y-junction model
$$
H_0(t)= i\sum_{\alpha=1}^3 \Delta_\alpha(t)\gamma_0\gamma_\alpha,
$$
the resulting correction simplifies to
$$
H_1(t)= i\frac{1}{2h_\Delta^2}\sum_{\alpha<\beta}
(\Delta_\beta\dot\Delta_\alpha-\Delta_\alpha\dot\Delta_\beta)\gamma_\alpha\gamma_\beta
$$
[1501.02811]. Exact implementation removes leakage and phase errors identically for any protocol duration, while approximate implementation with $\lambda=0.9$ or $\lambda=0.7$ still suppresses errors substantially, with residual errors scaling approximately as $(1-\lambda)^2$ [1501.02811].

A related control-theoretic usage appears in "Shortcuts to adiabaticity across a separatrix," where the supplied data states that the method presents a Hamiltonian bypass across a classical separatrix [2408.06916]. There the total Hamiltonian is
$$
H(q,p,t)=H_0(q,p;\lambda(t))+H_{\mathrm{cd}}(q,p,t),
\qquad
H_{\mathrm{cd}}(q,p,t)=p\,v(q,\lambda(t))\,\dot\lambda(t),
$$
with the velocity field chosen as
$$
v(q,\lambda)=-\frac{\partial_\lambda S(q,\lambda)}{\partial_q S(q,\lambda)}
$$
to transport every point on an initial energy shell to a target shell of the same enclosed Liouville volume, even across a separatrix where ordinary adiabatic transport fails [2408.06916]. The schedule is designed so that $\dot\lambda(t_s)=0$ at the crossing, keeping the product $v\dot\lambda$ finite despite the divergence of $v$ at the fixed point [2408.06916]. In the double-well erasure application, the work invested in the Hamiltonian stages is
$$
W=E_r-E_0,
$$
and both the energy cost and the fidelity are independent of protocol duration $\tau$ [2408.06916].

Finally, "Optimal shortcut approach based on an easily obtained intermediate Hamiltonian" is summarized in the supplied data as a “Hamiltonian Bypass” method for shortcut design [1705.08578]. Instead of adding the difficult counterdiabatic Hamiltonian of the original system, the method constructs an intermediate Hamiltonian
$$
H_I(t)=X(t)H_0(t)-K(t)H_{\mathrm{CD}}(t)
$$
that preserves the operator structure of $H_0$ while canceling problematic matrix elements, and then adds the counterdiabatic term of $H_I$ to obtain the final transitionless Hamiltonian
$$
H_T(t)=H_I(t)+\widetilde H_{\mathrm{CD}}(t)
$$
[1705.08578]. In the off-resonant $\Lambda$-system example, this removes the direct $|1\rangle\leftrightarrow |3\rangle$ coupling from the implementable Hamiltonian, leaving only modified pump, Stokes, and detuning controls [1705.08578].

Taken together, these non-graph-theoretic uses suggest a broader metaphor: a Hamiltonian is altered so that an otherwise obstructed path—through a local-minimum basin, a diabatic braiding process, or a separatrix crossing—can be traversed by an auxiliary or deformed dynamics. That broader usage, however, remains distinct from the established graph-theoretic definition of a Hamiltonian bypass as a one-arc reversal of a Hamiltonian cycle.

Source: https://www.emergentmind.com/topics/hamiltonian-bypass