Modal Sensitivity Sweep¶
Analytical procedure — instantiates Invariant-Mode Decomposition Design
Perturbs each mode's gain or coordinate in turn to see which ones actually move the outcomes you care about — turning a raw spectrum into a ranked map of where intervention has leverage, and exposing where modes bleed into one another.
A spectrum tells you how large each mode is, not which one is worth touching — and those are different questions. Modal Sensitivity Sweep answers the second: it perturbs each mode's gain or coordinate, one at a time, and measures how much the outcome you actually care about moves in response. A mode with a huge eigenvalue that no outcome depends on drops down the list; a modest mode that swings the result rises to the top. Its defining contribution is that it produces a leverage ranking, not a description — it is the derivative of outcomes with respect to modes — and, as a by-product of perturbing, it catches where nudging one mode disturbs another, recording the couplings that make a "clean" single-mode fix backfire. It consumes a decomposition it does not produce; what it adds is the translation from which modes exist to where pushing does the most good.
Example¶
A grid operator holds the modal model of a multi-area power system. Among its electromechanical modes is a lightly-damped ~0.4 Hz inter-area oscillation — whole regions of generators swinging against one another — that threatens stability after a fault. The sweep perturbs each candidate control in turn: add damping at generator A, at generator B, retune this power-system stabiliser, shift that setpoint — and after each nudge it measures the outcome that matters, how quickly tie-line power settles after a disturbance. The ranking is not obvious from the spectrum: damping applied at one particular generator moves the dangerous inter-area mode far more per unit effort than the same damping applied elsewhere — a participation-factor result, telling them where the mode "lives." The sweep also logs that nudging the inter-area mode shifts a neighbouring local mode with it (a coupling), so the chosen fix has to be checked against both, not the target alone. The deliverable is a ranked intervention map plus a coupling register — where a control-effort dollar buys the most stability, and which mode-pairs must be reasoned about together.
How it works¶
Its distinguishing move is that it works on outcomes, not on the modes themselves:
- Take the modal model as given. Start from a decomposition produced upstream; the sweep does not compute modes.
- Perturb one thing at a time. Nudge a single mode's gain or coordinate, or a single candidate control, holding the rest fixed.
- Measure the outcome response. Record how far the outcome metric moves — a sensitivity, or leverage, score for that mode or control.
- Log the cross-effects. Note where perturbing mode i also moves mode j; those entries populate the coupling register.
Tuning parameters¶
The dials that adapt the sweep to a specific system:
- Perturbation size — infinitesimal (a local gradient / participation factor) versus finite (captures near-degenerate and mildly nonlinear effects). Too small misses coupling; too large leaves the regime where the ranking is valid.
- Outcome metric — what "important" means: settling time, safety margin, variance explained, cost. The entire ranking rotates with this choice, so it must be the real objective, not a convenient proxy.
- One-at-a-time vs. joint sweep — perturbing modes singly is cheap but blind to interaction; joint perturbation catches coupling directly at combinatorial cost.
- Coupling threshold — how large a cross-effect must be before a mode-pair is entered in the register; set it to keep the register meaningful rather than full of noise.
When it helps, and when it misleads¶
Its strength is separating the loud modes from the load-bearing ones. The largest eigenvalue is not always the one worth acting on, and this is the mechanism that tells them apart — while surfacing the couplings that would otherwise turn a tidy single-mode intervention into an unpleasant surprise elsewhere.
Its failure modes both come from taking a local reading as a global truth. Sensitivity is local — valid near the operating point the perturbation was taken at[1] — so a large real-world change can silently invalidate the ranking; and near-degenerate modes (a small spectral gap) trade identity under perturbation, making their individual sensitivities unstable and close to meaningless. The classic misuse is running the sweep after the intervention has been chosen, perturbing only the favoured knob to manufacture a leverage story. The discipline that guards against this is to sweep the full candidate set at the true operating point, and to treat a small spectral gap as a signal to analyse the mode pair jointly rather than pretending the ranking can separate them.
How it implements the components¶
Modal Sensitivity Sweep fills the leverage-and-coupling slice of the archetype — the components a perturbation procedure can genuinely produce:
modal_intervention_map— its primary output: the ranking of which modes and controls actually move the outcome, i.e. where intervention has leverage rather than merely where the gain is large.mode_coupling_register— its by-product: the record of which modes move together under perturbation, so a single-mode fix is never trusted in isolation.
It says where to push and what is coupled, not what the modes are or whether they are stable — the invariant_mode_basis and modal_gain_spectrum come from Eigendecomposition Workflow, and the growth/decay verdict from Modal Stability Analysis's stable_unstable_mode_partition. Its intervention map is the dynamical-leverage sibling of the node-targeting map Network Spectral Centrality Analysis builds on a graph.
Related¶
- Instantiates: Invariant-Mode Decomposition Design — it turns the archetype's modes into an actionable leverage ranking.
- Consumes: Eigendecomposition Workflow supplies the modes the sweep perturbs.
- Sibling mechanisms: Modal Stability Analysis · Reduced-Order Model · Eigendecomposition Workflow · Network Spectral Centrality Analysis · Spectral Gap Monitor
Editorial Notes¶
Form Classification¶
Form family: Experiment, Test & Rehearsal
Rationale: Modal Sensitivity Sweep operates as a bounded trial, probe, simulation, or rehearsal that generates evidence from performance because it perturbs each mode's gain or coordinate in turn to see which ones actually move the outcomes you care about — turning a raw spectrum into a ranked map of where intervention has leverage, and exposing where modes bleed into one another.
Independent corroboration: The frozen evidence defines Modal Sensitivity Sweep as 'Perturbs each mode's gain or coordinate in turn to see which ones actually move the outcomes you care about — turning a raw spectrum into a ranked map of where intervention has leverage, and exposing where modes bleed into one another', so its operative form is Experiment, Test & Rehearsal.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Engineering & Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Perturbing modal coordinates or gains to rank outcome sensitivity is rooted in engineering modal and control analysis.
Related originating lineages:
- Mathematics — Eigenvalue sensitivity and perturbation theory provide the formal machinery.
- Operations Research — Sensitivity analysis in optimization materially shaped intervention-oriented ranking of parameter leverage.
- Physics — Normal-mode response analysis provides the physical interpretation of modal influence.
Review resolution: Both independent reviews agree on primary origin engineering_design; reconciliation resolves secondary fields (reported_ambiguity, alternate_origin_disagreement, origin_mode_disagreement, domain_reach_disagreement, encyclopedia_synthesis_disagreement). Alternate origins retained (mathematics, physics, operations_research) are the union of reviewer-supported formative lineages with explicit rationales, not a list of later application domains. Present-day breadth is represented separately as domain_reach=multi_domain; origin_mode=cross_disciplinary_synthesis records the historical relationship among lineages. Confidence is conservatively reconciled to medium, and encyclopedia_synthesis=true preserves either reviewer's finding that the encyclopedia generalized the mechanism.
Attribution caveat: The exact sweep-and-intervention framing appears synthetic.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; medium confidence.
References¶
[1] G. W. Stewart and Ji-guang Sun. Matrix Perturbation Theory. Academic Press, 1990. Treats first-order sensitivity as a small-perturbation approximation about a specified unperturbed matrix, with validity limited by higher-order terms. registry ↩