Post-Growth Voltage Ramping Overview

• Post-growth voltage ramping is an umbrella term for state-dependent voltage restoration after electrical disruptions such as breakdowns, faults, or renewable growth.
• Techniques include pulse-by-pulse linear recovery in DC systems, feeder-head hysteresis modeling, and query-sparse transformer forecasting, which together reduce secondary breakdowns and improve recovery performance.
• The research implies that accurate voltage recovery requires dynamic, state-dependent models that integrate transient energy delivery, spatial disorder, and flexible resource coordination.

Post-growth voltage ramping denotes the controlled restoration, prediction, or planning of voltage-related trajectories after a system has been displaced from nominal operation by breakdown, fault, stalled-motor dynamics, or growth in renewable-driven operating stress. The literature summarized under this label is technically heterogeneous. In a pulsed dc vacuum system, the problem is explicitly the gradual reapplication of voltage after a breakdown; in long radial feeders, it is closely related to feeder-head voltage restoration and fault-induced delayed voltage recovery; in post-fault monitoring, it appears as prediction of bus-voltage trajectories; and in expansion planning, it appears only indirectly through post-renewable-growth net-load ramping rather than voltage magnitudes themselves (Saressalo et al., 2021, Stolbova et al., 2014, Zheng et al., 2022, Li et al., 2016).

1. Terminological scope and domain structure

The phrase does not denote a single standardized model class. Instead, the available work partitions into direct, indirect, and analogical usages. The direct experimental treatment concerns post-breakdown voltage recovery in a pulsed dc system. Feeder studies address restoration of a depressed source voltage in a spatially extended, hysteretic load. Transformer-based forecasting addresses the shape of post-fault voltage evolution. Robust planning work addresses renewable-growth ramping stress, but with a DC power-flow model that omits voltage magnitudes. A nonlinear-process paper contributes a transferable ramping formalism, but not a voltage model.

Paper	Direct subject	Relation to post-growth voltage ramping
"Linear voltage recovery after a breakdown in a pulsed dc system" (Saressalo et al., 2021)	Post-breakdown voltage recovery in a pulsed dc vacuum system	Direct treatment of gradual voltage reapplication after an arc
"Fault Induced Delayed Voltage Recovery in a Long Inhomogeneous Power Distribution Feeder" (Stolbova et al., 2014)	Sudden feeder-head voltage depression and restoration on a long radial feeder	Indirect but highly relevant to restoration/ramping on feeders
"glassoformer: a query-sparse transformer for post-fault power grid voltage prediction" (Zheng et al., 2022)	Post-fault bus-voltage trajectory prediction	Predictive treatment of recovery shape
"Robust Coordinated Transmission and Generation Expansion Planning Considering Ramping Requirements and Construction Periods" (Li et al., 2016)	Hourly net load ramping uncertainty and expansion planning	Post-growth ramping as net-load/power, not voltage
"Demand Response for Flat Nonlinear MIMO Processes using Dynamic Ramping Constraints" (Baader et al., 2022)	Dynamic ramping constraints for flat nonlinear MIMO processes	Analogical state-dependent ramping formalism

A central unifying idea is path dependence. Whether voltage is reapplied after a vacuum arc, restored at the head of a long feeder, predicted from early post-fault data, or supported indirectly through flexible planning resources, the trajectory matters, not only the endpoint. This suggests that “post-growth voltage ramping” is best understood as an umbrella term for state-dependent transition management rather than as a single protocol.

2. Pause-free linear recovery after vacuum breakdown

The most direct experimental account of post-event voltage ramping studies a pulsed dc breakdown test stand in which a DC power supply fed a Marx generator that amplified the DC level and produced short square pulses. The standard operating regime was a pulse repetition rate of $2000~\mathrm{Hz}$ and pulse length of $1~\mu\mathrm{s}$. The electrodes were cylindrical copper electrodes in a vacuum chamber at about $7\times 10^{-8}~\mathrm{mbar}$, with fixed gap $d=40~\mu\mathrm{m}$. The anode had a $40~\mathrm{mm}$ contact diameter, the grounded cathode $60~\mathrm{mm}$, and together with the HV cable they formed an effective capacitance of roughly $700~\mathrm{pF}$. Voltages up to $6000~\mathrm{V}$, corresponding to an estimated field of $150~\mathrm{MV/m}$ via $E=V/d$, were available (Saressalo et al., 2021).

After each breakdown, the pulse voltage was dropped to a starting value $1~\mu\mathrm{s}$0, then increased linearly pulse-by-pulse until the target voltage $1~\mu\mathrm{s}$1 was reached. The control parameters were the starting fraction $1~\mu\mathrm{s}$2, the number of pulses used for recovery $1~\mu\mathrm{s}$3, and therefore the effective slope $1~\mu\mathrm{s}$4 and the per-pulse increment $1~\mu\mathrm{s}$5. Because the repetition rate was fixed at $1~\mu\mathrm{s}$6, the recovery duration was $1~\mu\mathrm{s}$7. In a representative case, $1~\mu\mathrm{s}$8, $1~\mu\mathrm{s}$9, and the ramp lasted 2000 pulses, i.e. $7\times 10^{-8}~\mathrm{mbar}$0, giving a nominal slope of $7\times 10^{-8}~\mathrm{mbar}$1 and $7\times 10^{-8}~\mathrm{mbar}$2 per pulse. The “linear” ramp was conceptually continuous in time, but the practically relevant quantity was the pulse-to-pulse increment. Because the source was RC-limited, the actual voltage trajectory deviated from the ideal one, especially for steep ramps: for the steepest 500-pulse recovery, the voltage had reached only about $7\times 10^{-8}~\mathrm{mbar}$3 of $7\times 10^{-8}~\mathrm{mbar}$4 by the nominal end of the ramp, while 1000, 2000, 4000, and 8000 pulses yielded about $7\times 10^{-8}~\mathrm{mbar}$5, $7\times 10^{-8}~\mathrm{mbar}$6, $7\times 10^{-8}~\mathrm{mbar}$7, and $7\times 10^{-8}~\mathrm{mbar}$8 of $7\times 10^{-8}~\mathrm{mbar}$9, respectively.

Breakdown statistics were based on the number of pulses between consecutive breakdowns. The probability density of inter-breakdown pulse counts was fit to

$d=40~\mu\mathrm{m}$0

and the crossing point $d=40~\mu\mathrm{m}$1 between the two exponentials was used to separate short-delay secondary breakdowns from longer-delay primary breakdowns. This classification made the recovery stage itself a quantitatively analyzable conditioning process rather than a simple reset.

Three protocol families were examined. In the first, only slope was changed while keeping $d=40~\mu\mathrm{m}$2. At $d=40~\mu\mathrm{m}$3, 500, 1000, 2000, 4000, and 8000 recovery pulses corresponded to nominal slopes of $d=40~\mu\mathrm{m}$4, $d=40~\mu\mathrm{m}$5, $d=40~\mu\mathrm{m}$6, $d=40~\mu\mathrm{m}$7, and $d=40~\mu\mathrm{m}$8, with $d=40~\mu\mathrm{m}$9 of $40~\mathrm{mm}$0, $40~\mathrm{mm}$1, $40~\mathrm{mm}$2, $40~\mathrm{mm}$3, and $40~\mathrm{mm}$4. In the second family, the slope was held nearly fixed at about $40~\mathrm{mm}$5 while the starting voltage varied. In the third, a two-stage protocol combined flat pulsing at a nonzero sub-threshold voltage with a subsequent linear ramp, always totaling 10000 pulses.

The central result was that gentler recovery with more pulses reduced both the fraction of secondary breakdowns and the overall breakdown rate. In the first measurement set, the breakdown rate fell from $40~\mathrm{mm}$6 breakdowns per pulse for the 500-pulse recovery to $40~\mathrm{mm}$7 for the 8000-pulse recovery. The secondary breakdown fraction $40~\mathrm{mm}$8 dropped from $40~\mathrm{mm}$9 to $60~\mathrm{mm}$0, and the mean number of consecutive breakdowns between primary events, $60~\mathrm{mm}$1, fell from $60~\mathrm{mm}$2 to $60~\mathrm{mm}$3. By contrast, lowering the number of pulses led to more dramatic voltage recovery and higher breakdown rates.

The starting voltage mattered, but not monotonically. The $60~\mathrm{mm}$4 case was by far the worst, with a breakdown rate of $60~\mathrm{mm}$5 bpp and $60~\mathrm{mm}$6 secondary breakdowns; more than half of all breakdowns occurred on the very first pulse after the previous breakdown, and the PDF could not be fit with the standard two-exponential model. Starting from $60~\mathrm{mm}$7, $60~\mathrm{mm}$8, $60~\mathrm{mm}$9, or $700~\mathrm{pF}$0 at the same slope gave broadly comparable behavior, and pulsing below about $700~\mathrm{pF}$1 did not appear to improve conditioning further in that measurement set. In practical terms, $700~\mathrm{pF}$2 of target emerged as a useful compromise.

Slope affected not only how many secondary breakdowns occurred, but when they occurred. Steeper ramps localized the secondary-breakdown probability near the end of recovery and into the first pulses at $700~\mathrm{pF}$3, while shallower ramps spread that probability more broadly and lowered the peak. The paper states that the maximum breakdown probability is typically reached a few hundred pulses after the voltage has reached its target value, and more generally reports that the peak breakdown probability is regularly observed around $700~\mathrm{pF}$4 after the end of the ramping period. Since the repetition rate is $700~\mathrm{pF}$5, that corresponds to about 2000 pulses. The fitted cross-point $700~\mathrm{pF}$6 was consistently shifted by about 2000 pulses beyond the end of the ramp.

Energy delivered during recovery was another organizing variable. The recovery energy was defined as

$700~\mathrm{pF}$7

with $700~\mathrm{pF}$8 for the analyzed dataset. Excluding the pathological $700~\mathrm{pF}$9 start case, the peak breakdown probability followed a strong empirical exponential decrease with ramping energy,

$6000~\mathrm{V}$0

with $6000~\mathrm{V}$1, $6000~\mathrm{V}$2, and $6000~\mathrm{V}$3. More energy delivered during recovery therefore correlated with a lower peak breakdown probability.

The paper also compared linear ramps with an earlier 20-step recovery protocol. In the stepwise case, the breakdown probability within each step decayed exponentially after the voltage change, with fitted half-life $6000~\mathrm{V}$4 pulses $6000~\mathrm{V}$5 and $6000~\mathrm{V}$6. That time constant is not a lifetime of the overall recovery process; it is the rapid decay time of the excess breakdown probability immediately after a discrete voltage step. The comparison supports pause-free linear recovery over stepped recovery with holds, because the stepwise protocol exhibited a strong breakdown-probability spike at the start of every step, and earlier work had implicated idle time through gas re-adsorption and step-change effects.

The mechanistic account is explicitly tentative. The authors propose that voltage changes activate short-lived field emitters, that sudden increases in electric field may promote local atomic migration and create fresh emitters, and that some emitters either die out harmlessly or enter thermal runaway and trigger full breakdown. They also suggest that repeated pulsing at moderate field may detach loosely bound impurities and effectively clean the surface. These ideas are interpretations rather than directly demonstrated mechanisms, but they organize the empirical finding that post-breakdown voltage recovery is an active conditioning stage rather than a return-to-service formality.

3. Feeder-head restoration, hysteresis, and delayed recovery

In long radial distribution feeders dominated by induction motors, voltage restoration after a disturbance is governed by hysteresis, spatial coupling, and propagating fronts rather than by a uniform local reset. The feeder model is a continuum on $6000~\mathrm{V}$7, with boundary conditions

$6000~\mathrm{V}$8

where $6000~\mathrm{V}$9 is the feeder-head voltage and $150~\mathrm{MV/m}$0, $150~\mathrm{MV/m}$1 are real and reactive power flows. The paper studies steady normal operation, a stalling event caused by suddenly lowering $150~\mathrm{MV/m}$2, restoration from a fully stalled state by suddenly raising $150~\mathrm{MV/m}$3 back to $150~\mathrm{MV/m}$4, and a fault-clearing scenario in which $150~\mathrm{MV/m}$5 is lowered for a finite time $150~\mathrm{MV/m}$6 and then restored (Stolbova et al., 2014).

The local induction motor is described by

$150~\mathrm{MV/m}$7

with slip $150~\mathrm{MV/m}$8, and mechanical dynamics

$150~\mathrm{MV/m}$9

For $E=V/d$0, especially $E=V/d$1, the torque-speed balance yields bistability over an intermediate voltage range $E=V/d$2: a stable normal state with $E=V/d$3, a stable stalled state with $E=V/d$4, and an unstable equilibrium between them. For the representative single-motor example with $E=V/d$5 and $E=V/d$6, the paper reports approximately $E=V/d$7 and $E=V/d$8. This hysteresis makes upward voltage restoration nontrivial. A motor that is already stalled remains stalled until the local voltage exceeds $E=V/d$9, even if that same voltage would have been adequate to keep the motor on the normal branch had it never stalled.

The feeder itself is governed by a continuous DistFlow-style model:

$1~\mu\mathrm{s}$00

The coupling is long range. Local voltage affects local motor torque and slip; slip determines local $1~\mu\mathrm{s}$01 and $1~\mu\mathrm{s}$02, especially the large reactive demand in or near stall; those power demands determine feeder flows; and feeder flows determine voltage drops along the line. The result is coexistence of spatially extended normal and stalled phases separated by a front.

The work introduces disorder through a spatially correlated Gaussian random field in the mechanical torque density,

$1~\mu\mathrm{s}$03

This makes the transition front somewhat random and creates a blurred domain in which neighboring motors may end up normal or stalled depending on the local disorder realization. The paper quantifies this by a front position $1~\mu\mathrm{s}$04 and a blurry-region width $1~\mu\mathrm{s}$05. At fixed $1~\mu\mathrm{s}$06, roughly doubling $1~\mu\mathrm{s}$07 approximately doubles the width of the distribution $1~\mu\mathrm{s}$08. The width $1~\mu\mathrm{s}$09 increases with increasing disorder amplitude $1~\mu\mathrm{s}$10 and increases as correlation length $1~\mu\mathrm{s}$11 decreases.

The fault-induced delayed voltage recovery mechanism is path dependent. During a temporary depression of $1~\mu\mathrm{s}$12, motors decelerate first near the remote end, because they are already closer to stall. As they slow, their reactive consumption increases, further depressing voltage and driving a stalling front inward. If the disturbance is cleared quickly enough, restoring $1~\mu\mathrm{s}$13 can launch a restoration front from the feeder head. But if motors at the end have slowed too much, restoring $1~\mu\mathrm{s}$14 to $1~\mu\mathrm{s}$15 may still leave them on the wrong side of the torque-speed balance. The paper illustrates this with snapshots: around $1~\mu\mathrm{s}$16, end-of-line motors had roughly $1~\mu\mathrm{s}$17 and $1~\mu\mathrm{s}$18, so restoring $1~\mu\mathrm{s}$19 then raised the end voltage enough that electrical torque exceeded mechanical torque and recovery became self-reinforcing. Around $1~\mu\mathrm{s}$20, end-of-line motors had roughly $1~\mu\mathrm{s}$21 and $1~\mu\mathrm{s}$22; restoring $1~\mu\mathrm{s}$23 then raised the end voltage only to about $1~\mu\mathrm{s}$24 at best, insufficient to reverse deceleration, so the motors continued toward stall after clearing.

When a fully stalled feeder at $1~\mu\mathrm{s}$25 is restored to $1~\mu\mathrm{s}$26, recovery begins at $1~\mu\mathrm{s}$27, where voltage recovers first and most strongly. As nearby motors accelerate, their reactive demand decreases, voltage drop lessens, and the front can move farther into the feeder. However, the front slows down and can become stationary, leaving part of the feeder still stalled. In the homogeneous case, the stopping point is sharp; in the disordered case, the front region is blurred. The paper does not directly study gradual ramps, because the feeder-head voltage changes are sudden step changes rather than slow upward ramps. Still, its quasi-static implication is clear: gradual source-voltage increase will not restore stalled motors until local voltages exceed the stalled-branch spinodal $1~\mu\mathrm{s}$28, and on a long feeder this threshold is mediated by collective reactive loading rather than by source voltage alone.

The clearing-time results sharpen the operational interpretation. The maximum clearing time $1~\mu\mathrm{s}$29 is the longest duration of lowered $1~\mu\mathrm{s}$30 for which the feeder still fully recovers after restoring $1~\mu\mathrm{s}$31 to $1~\mu\mathrm{s}$32. Increasing disorder amplitude $1~\mu\mathrm{s}$33 shifts $1~\mu\mathrm{s}$34 to shorter clearing times, while increasing correlation length $1~\mu\mathrm{s}$35 causes significant broadening of $1~\mu\mathrm{s}$36. This suggests that feeder restoration after growth or disturbance cannot be treated as purely local or purely quasi-static. Spatial voltage-drop structure, reactive burden, and motor-state hysteresis determine whether voltage restoration succeeds, stalls, or leaves embedded pockets of unrecovered load.

4. Trajectory forecasting after faults

A distinct research direction treats post-event voltage ramping as a forecasting problem: given a short initial observation window after a fault, predict the subsequent voltage trajectory of a bus. GLassoformer formulates this as single-bus voltage trajectory forecasting using local multivariate signals from electrically adjacent buses and lines. The benchmark is the New York/New England 16-generator, 68-bus system with 68 buses, 88 lines, and 2248 simulated fault events, each containing 10 seconds of signals. The target is the voltage of a bus; inputs use voltage with currents from locally connected buses and lines; and the input is zero-mean normalized (Zheng et al., 2022).

Architecturally, GLassoformer is an encoder-decoder transformer with convolutional embedding, 2 encoder layers with multi-head Group Sparse attention followed by temporal convolution, 1 decoder layer of Group Sparse self-attention, 1 standard cross-attention layer, and a fully connected output layer. Its core modification is query sparsification via a group Lasso penalty,

$1~\mu\mathrm{s}$37

with

$1~\mu\mathrm{s}$38

Because $1~\mu\mathrm{s}$39, zeroing column groups of $1~\mu\mathrm{s}$40 zeros corresponding rows of $1~\mu\mathrm{s}$41, inducing structured query sparsity. The paper states that if only $1~\mu\mathrm{s}$42 rows of $1~\mu\mathrm{s}$43 are nonzero, attention cost can drop to about $1~\mu\mathrm{s}$44.

The direct relevance to voltage ramping is not an explicit ramp-rate model, but the prediction of full post-fault voltage recovery curves. Those curves encode rapid drops, delayed recovery, oscillatory rebound, overshoot, and gradual return toward steady state. The method therefore treats ramping as trajectory shape rather than as a scalar stability label.

Model	Case I MSE / MAE	Case II MSE / MAE
GLassoformer	$1~\mu\mathrm{s}$45 / $1~\mu\mathrm{s}$46	$1~\mu\mathrm{s}$47 / $1~\mu\mathrm{s}$48
Informer	$1~\mu\mathrm{s}$49 / $1~\mu\mathrm{s}$50	$1~\mu\mathrm{s}$51 / $1~\mu\mathrm{s}$52
Lassoformer	$1~\mu\mathrm{s}$53 / $1~\mu\mathrm{s}$54	$1~\mu\mathrm{s}$55 / $1~\mu\mathrm{s}$56
1D-CNN	$1~\mu\mathrm{s}$57 / $1~\mu\mathrm{s}$58	$1~\mu\mathrm{s}$59 / $1~\mu\mathrm{s}$60
Prony	—	$1~\mu\mathrm{s}$61 / $1~\mu\mathrm{s}$62

These results show that GLassoformer tracks post-fault voltage evolution more accurately than the listed baselines. Efficiency comparisons place GLassoformer at 5.827M parameters and 18.98 ms inference, versus 7.257M and 29.76 ms for Informer, 5.707M and 14.92 ms for Lassoformer, and 0.706M and 0.5969 ms for 1D-CNN. Its pruning rate is 19.09%, compared with 4.220% for Informer and 2.674% for Lassoformer. The paper also reports that Informer can exhibit spurious spikes, attributed to its random sampling in sparse attention, whereas GLassoformer produces smoother and more stable predicted trajectories.

The limitations are equally important. The model predicts trajectories but does not explain physical mechanisms such as motor stalling, reactive deficiency, AVR or exciter interaction, or fault-clearing dynamics. It does not evaluate recovery-specific metrics such as recovery time, minimum post-fault voltage, ramp-rate error, damping ratio, or voltage-violation count. Its evidence therefore concerns sequence accuracy and stability, not causal explanation. In the context of post-growth voltage ramping, this makes the model useful for online monitoring and early warning, but not sufficient as a mechanistic theory.

5. Renewable-growth ramping and multi-year planning

A different meaning of “post-growth” appears in robust coordinated transmission and generation expansion planning under rapid wind integration. Here the central problem is not voltage ramping in the dynamic sense, but severe power-ramping events caused by renewable-driven net-load variability. The framework is a two-stage robust optimization model that coordinates transmission lines, generators, and FACTS devices while explicitly considering flexibility requirement, construction period, and cost. The network model is DC power flow, so it does not represent reactive power, voltage magnitudes, voltage stability margins, or voltage ramping explicitly (Li et al., 2016).

The key modeling distinction is between annual net load duration curve uncertainty (LDCU), which captures uncertainty in the level of annual net load, and hourly net load ramping uncertainty (HLRU), which captures uncertainty in hourly changes of net load. The HLRU set is

$1~\mu\mathrm{s}$63

and redispatch under HLRU is constrained by generator ramp-rate limits such as

$1~\mu\mathrm{s}$64

This is a rigorous model of post-growth ramping stress, but in active-power terms.

The 6-bus case uses PJM net-load data with a 5-year horizon, 5% annual net-load growth, 5% annual growth in hourly ramping range, and LDCU equal to 5% of each nodal net load. A strong result is that only the full combination of transmission, generation, and FACTS—denoted $1~\mu\mathrm{s}$65—was feasible. $1~\mu\mathrm{s}$66 only, $1~\mu\mathrm{s}$67 only, $1~\mu\mathrm{s}$68 only, $1~\mu\mathrm{s}$69, $1~\mu\mathrm{s}$70, and $1~\mu\mathrm{s}$71 were all infeasible for the stated reasons. This makes ramping stress a coordination problem across flexible resources rather than a single-resource adequacy problem.

Construction lead times are decisive. With generator and FACTS lead times zero, a line lead time of 0 years yielded line cost \$1~\mu\mathrm{s}$720, generator \$1~\mu\mathrm{s}$73280.41M. Increasing line lead time to 1 year decreased line investment to \$1~\mu\mathrm{s}$7415.46M, left generator cost unchanged, and raised total cost to \$300.81M. A line lead time of 2 years made the problem infeasible. If generator construction period increased to 1 year, the problem became infeasible even with zero line lead time, because no new generator could be installed in the first year to handle HLRU. The paper therefore identifies generator lead time as especially critical for post-growth ramping.

The comparison between planning models sharpens the same point. Model M1 is deterministic, M2 includes LDCU only, and M3 includes both LDCU and HLRU. Their total costs are \$1~\mu\mathrm{s}$75266.75M, and \$1~\mu\mathrm{s}$76261.53M and EENS 3936.96 MWh/yr, M2 had <a href="https://www.emergentmind.com/topics/explore-then-commit-etc" title="" rel="nofollow" data-turbo="false" class="assistant-link" x-data x-tooltip.raw="">ETC</a> \$1~\mu\mathrm{s}$77162.04M and EENS 0. Over 50 annual scenarios from PJM-based hourly data, M3 reduced average expected total cost by about 3.3% relative to M2 over five years.

For voltage-ramping questions, the significance is indirect but important. The paper demonstrates that renewable growth creates short-term ramping requirements that must be planned jointly with network expansion and flexible resources. However, because the model is DC and omits voltage magnitudes, any claim about post-growth voltage support remains an implication rather than a demonstrated result. A plausible implication is that planners concerned with actual voltage ramping would need to augment this framework with AC power flow, reactive capability, and voltage security constraints.

6. State-dependent ramping as a transferable abstraction

The most general ramping formalism in the cited corpus comes from work on flat nonlinear MIMO processes rather than electrical voltage itself. There, dynamic ramping constraints replace constant ramp-rate bounds by a high-order, state-dependent model in which

$1~\mu\mathrm{s}$78

together with bounds on $1~\mu\mathrm{s}$79, intermediate derivatives, and $1~\mu\mathrm{s}$80 that depend on the current operating point. The process is transformed into flat coordinates, an operating strategy couples the flat outputs to the production rate, and original state and input bounds are mapped into nonlinear feasible limits on $1~\mu\mathrm{s}$81. These limits are then conservatively approximated by piecewise-affine functions so that the resulting MILP remains feasible for the original nonlinear dynamics (Baader et al., 2022).

This is not a voltage paper. Its scheduling variable is production rate $1~\mu\mathrm{s}$82, not a bus voltage or converter voltage, and the case study is a heated reactor-separator process. Yet the formal lesson transfers. Across the directly voltage-related papers, admissible post-event trajectories are not well described by a constant slope bound. In the pulsed dc breakdown study, recovery quality depends on $1~\mu\mathrm{s}$83, $1~\mu\mathrm{s}$84, $1~\mu\mathrm{s}$85, and cumulative ramping energy rather than on endpoint voltage alone (Saressalo et al., 2021). In the feeder study, restoration depends on whether local voltages exceed a hysteretic recovery threshold and on whether motors have already decelerated too far, so the effective ramping limit is state dependent and spatially coupled (Stolbova et al., 2014). In the planning paper, the relevant post-growth limit is generator redispatch under HLRU and lead-time constraints, again showing that feasible recovery depends on dynamic flexibility rather than steady-state adequacy (Li et al., 2016). In the forecasting paper, the object being predicted is the entire trajectory, not a scalar recovery margin, which likewise treats post-fault voltage evolution as a structured time-dependent process (Zheng et al., 2022).

Several misconceptions are therefore contradicted by the literature. Immediate return to nominal voltage is not generally benign: in the pulsed dc system it sharply elevates secondary-breakdown risk. Nominal source-voltage restoration does not guarantee feeder recovery: stalled induction motors can remain trapped on the stalled branch because of hysteresis and reactive-power feedback. Accurate trajectory prediction does not by itself explain the underlying mechanism: GLassoformer is predictive rather than causal. Post-growth ramping is not automatically a voltage problem: in robust expansion planning it is explicitly a net-load/power-ramping problem under a DC model.

Taken together, these works describe post-growth voltage ramping as a family of state-dependent transition problems. The experimentally strongest claims concern statistical suppression of secondary breakdowns by gentler pause-free recovery, hysteretic front-mediated feeder restoration after voltage depression, improved post-fault trajectory prediction through query-sparse transformers, and the necessity of modeling renewable-growth ramping stress in long-term planning. The broader mechanistic and cross-domain connections—self-cleaning surfaces, short-lived emitters, blurred recovery fronts, or direct extension of dynamic-ramping formalisms to voltage—are best treated as interpretations or transferable abstractions rather than as already unified theory.