Threshold-Based Cutoff in Network Flow

• Threshold-based cutoff mechanisms are network selection processes that convert distributed flow problems into single-path routing by applying local threshold conductances.
• The approach minimizes a linear sum of threshold costs instead of quadratic dissipation, ensuring that the entire current is routed along the path with the smallest cumulative threshold.
• This variational framework explains transient lightning behavior where initial branching gives way to a single discharge path as sharp, threshold nonlinearities suppress competing routes.

Searching arXiv for the primary paper and closely related threshold/cutoff work to ground the article. arXiv search query: "^^^^^^^^1ti:\1 optimizes: a threshold mechanism ensures minimum-path flow\"" A threshold-based cutoff mechanism is a network-selection mechanism in which monotone threshold-like conductances convert a distributed flow problem into a single-path routing problem. In the formulation developed for lightning and dielectric discharge, the network is a graph of capacitive links and nonlinear resistors, the energy source is concentrated at one node, and the steady-state current no longer minimizes the quadratic dissipation functional of linear circuits. In the threshold limit, it minimizes instead a linear min-sum threshold cost, with the consequence that all current is routed along a single source-to-ground path whose cumulative thresholds are minimal. The same framework also yields a transient explanation of lightning: multiple tentative branches appear first, but once one branch reaches ground, the threshold nonlinearity suppresses competing branches and retains only the optimal one (&&&^^^^1ti:\1^^^^&&&).

1. Network formulation and constitutive assumptions

The mechanism is posed on a connected electrical network modeled as a graph PRESERVED_PLACEHOLDER_^^^^1ti:\1^^^^ with $n+1$ nodes $N=\{0,\dots,n\}$ and $m$ links $L=\{0,\dots,m-1\}$. Node $n$ is the zero-potential ground, $v=0$, and node $0$ is the source where a generator is applied and injects an input current $d$, with its other terminal grounded. Each link $k=(i,j)$ is the parallel connection of a capacitor $n+1$^^^^1ti:\1^^^^ and a possibly nonlinear resistor with current–voltage characteristic $n+1$1. If the endpoint voltages are $n+1$2 and $n+1$3, then the link current is

$n+1$4

Using the node–link incidence matrix $n+1$5, with the ground row removed, the dynamics are written compactly as

$n+1$6

where $n+1$7 and $n+1$8. At steady state, Kirchhoff current law reduces to

$n+1$9

The constitutive assumption is that each $N=\{0,\dots,n\}$^^^^1ti:\1^^^^ is odd, monotonically increasing, and locally Lipschitz; monotonically non-decreasing suffices. Under this assumption, each $N=\{0,\dots,n\}$1 is invertible, with inverse $N=\{0,\dots,n\}$2. The network is connected internally and to the external environment so that $N=\{0,\dots,n\}$3 has full row rank. These hypotheses are the structural basis for both the variational characterization and the cutoff result (&&&^^^^1ti:\1^^^^&&&).

2. Variational structure of steady-state flow

For general monotone resistors, steady-state flow is characterized by a convex optimization problem rather than by an ad hoc path rule. Defining

$N=\{0,\dots,n\}$4

the steady-state current distribution minimizes

$N=\{0,\dots,n\}$5

subject to

$N=\{0,\dots,n\}$6

If the $N=\{0,\dots,n\}$7 are strictly increasing, the minimizer is unique (&&&^^^^1ti:\1^^^^&&&).

The linear resistive case provides the contrast that makes the cutoff mechanism precise. For $N=\{0,\dots,n\}$8,

$N=\{0,\dots,n\}$9

which matches $m$^^^^1ti:\1^^^^ for

$m$1

Because this quadratic functional is strictly convex, the minimizing flow is generally scattered over many branches. There is therefore no intrinsic single-path collapse in linear networks.

Regime	Minimized functional	Qualitative flow outcome
Linear resistors	$m$2	Currents scatter over many branches
Threshold limit	$m$3	Current concentrates on one minimum-threshold path

This distinction removes a common misconception: the concentration of lightning-like discharge into one branch is not a generic property of all passive networks. It is a consequence of threshold nonlinearity combined with single-point injection (&&&^^^^1ti:\1^^^^&&&).

3. Threshold law and minimum-threshold path selection

The ideal threshold law associated with dielectric breakdown assigns each link a local dielectric rigidity $m$4:

$m$5

Below threshold, the link is effectively insulating; above threshold, it becomes virtually infinitely conductive. Its inverse is multivalued at zero and constant away from zero:

$m$6

In this limit, the convex functional becomes

$m$7

For a single source at node $m$8 and ground at node $m$9, let $L=\{0,\dots,m-1\}$^^^^1ti:\1^^^^ be the family of simple source-to-ground paths. The optimal steady-state flow sends the entire current $L=\{0,\dots,m-1\}$1 along a single path $L=\{0,\dots,m-1\}$2 minimizing

$L=\{0,\dots,m-1\}$3

The optimality criterion is therefore the min-sum of thresholds, not a minimax rule (&&&^^^^1ti:\1^^^^&&&).

The proof reduces the threshold problem to the linear program

$L=\{0,\dots,m-1\}$4

which is the standard min-cost flow problem with unit supply at the source and costs $L=\{0,\dots,m-1\}$5 per unit flow on each edge. Its optimum is a single path flow. If multiple paths have exactly the same cumulative threshold, any convex combination of tied path flows is also optimal, but such ties occur with probability zero for random $L=\{0,\dots,m-1\}$6.

The same conclusion appears directly from flow decomposition. If a feasible flow is decomposed into path fractions $L=\{0,\dots,m-1\}$7 with $L=\{0,\dots,m-1\}$8 and $L=\{0,\dots,m-1\}$9, then

$n$^^^^1ti:\1^^^^

with equality only when all flow lies on minimum-threshold paths. Splitting current among multiple branches is therefore suboptimal except in exact ties. This is the cutoff mechanism in variational form: threshold penalties eliminate the quadratic incentive to spread current and replace it with a linear cost that structurally favors path concentration (&&&^^^^1ti:\1^^^^&&&).

The ideal law is not physically realizable, so the analysis also introduces sharp monotone approximations, for example the piecewise-linear law

$n$1

with $n$2 and $n$3 large. A convergence theorem states that the unique steady-state minimizers for such sharp monotone laws converge to the threshold minimizer that routes all current on the unique minimum-threshold path as $n$4.

4. Dynamic cutoff and lightning transient behavior

The transient model is the distributed-circuit dynamics

$n$5

With $n$6, the Lyapunov function

$n$7

satisfies

$n$8

where $n$9 is a diagonal matrix of strictly positive continuous functions arising from the monotonicity of $v=0$^^^^1ti:\1^^^^. Solutions therefore converge asymptotically to the steady state defined by

$v=0$1

(&&&^^^^1ti:\1^^^^&&&).

Within this dynamics, the cutoff mechanism has two qualitative phases for Category 1 cloud-to-ground lightning. In the stepped leader phase, currents are relatively low, capacitive terms dominate, voltages change rapidly, and multiple branches form tentatively. After connection to ground through a connecting leader, dielectric breakdown fully develops, above-threshold conduction dominates, and the long-duration discharge flows along the single optimal path while other branches are depleted.

The decisive event is that the first branch to reach ground captures the flow. Below threshold, the slope is $v=0$2, so many links remain effectively non-conductive. Once one branch crosses threshold along a source-to-ground sequence with the smallest cumulative threshold, its conductance jumps because the slope becomes $v=0$3. This sharply lowers the voltages along that branch and keeps competing branches below threshold, thereby cutting them off. This mechanism is consistent both with the min-sum variational criterion and with the global convergence result (&&&^^^^1ti:\1^^^^&&&).

The numerical study uses the sharp piecewise-linear characteristic with

$v=0$4

capacitances $v=0$5, and thresholds drawn uniformly from

$v=0$6

Larger $v=0$7 increases initial branching activity, but asymptotically a single branch survives and matches the minimum-threshold path. This suggests that stronger spatial heterogeneity amplifies exploratory transients without altering the final selection rule.

5. Conditions for single-path selection, multipath persistence, and model limits

The single-path outcome is guaranteed only under specific structural conditions. The source must be concentrated at a single node, threshold costs along competing paths must not be exactly tied, and the constitutive laws must be sufficiently sharp monotone threshold-like functions converging to the ideal threshold law. Under these conditions, the asymptotic current distribution concentrates on one path (&&&^^^^1ti:\1^^^^&&&).

Multipath persistence is not excluded in principle, but it requires exact degeneracy. If multiple paths have exactly the same cumulative threshold, $v=0$8 is convex but not strictly convex, so several minimizers exist. With smooth threshold-like characteristics, minor splitting may also occur transiently, but this vanishes as the sharpness parameter increases and the asymptotic steady state is approached. The non-uniqueness issue is therefore a tie phenomenon rather than a generic failure of the mechanism.

Several limitations are explicit. The ideal threshold law is an idealization, although sharp monotone characteristics converge to its solution. Inductances are neglected; this does not affect the minimum-path analysis because the latter is conducted at steady state, where $v=0$9, and the return stroke starts after the path has been selected. The mechanism also assumes a single-point source and zero-potential ground; distributed sources or sinks would generally produce different, possibly multipath, optimal structures.

The model is topology-independent: any connected network is admissible, not only lattices or grids. Conductive grounded objects can be incorporated as links with very small resistance and large capacitance, effectively adding zero-cost segments to some paths. Spatial variability in $0$^^^^1ti:\1^^^^ induced by humidity, temperature, pressure, or pollution explains why observed lightning paths appear irregular: the selected trajectory is optimal relative to the instantaneous threshold field, not relative to Euclidean distance (&&&^^^^1ti:\1^^^^&&&).

6. Conceptual significance and broader analogies

The threshold-based cutoff mechanism formalizes a decentralized optimization principle. Each link applies only a local monotone threshold rule, yet the network-level consequence is a shortest-path-type solution: the global outcome is the minimum-sum-threshold path from source to ground. In graph-theoretic terms, the ideal threshold problem is equivalent to min-cost flow with edge costs $0$1, and the resulting discharge path is the corresponding minimum-cost path (&&&^^^^1ti:\1^^^^&&&).

This interpretation also clarifies what the mechanism is not. It is not a restatement of minimum-energy routing in the linear sense, because the relevant functional is not quadratic once threshold-like conductances are present. It is not a minimax criterion, because the selected path minimizes the sum of thresholds rather than the largest threshold encountered. And it is not merely a transient instability, because the transient branching pattern and the final single-path outcome are both consequences of the same variational structure.

Beyond lightning, the same mechanism applies to dielectric breakdown and, by the paper’s own analogy, to gas discharges and other phenomena in which conductance emerges only after crossing local thresholds. A plausible implication is that whenever a networked transport process combines monotone threshold activation with single-point forcing, one should expect the governing functional to favor concentrated routing rather than branch sharing. In the setting analyzed here, that implication is fully explicit: threshold nonlinearity transforms scattered linear-network flow into a cutoff-driven, single-path discharge.