The Gold Standard of Vibrations | Testing Stability from a Repeated Stiffness-Switching Cycle

Use one effective mass m with displacement x and velocity v. During each constant-stiffness phase, the equation is m ẍ + kx = 0, with positive k. There is no separately imposed additive driving force and no damping in this teaching model.

Stiffness changes at declared instants according to a repeating schedule. The displacement and velocity remain continuous at those ideal switches. Acceleration can change because the restoring-force coefficient changes.

The switches are mathematical changes in a parameter, not impacts. If an actual mechanism gives an impulse or changes another part of the system, it would require a different model.

Each fixed phase has natural angular frequency ω = √(k/m). Knowing those two frequencies does not yet specify the repeated motion. The durations, order and starting state are also part of the problem.

Write the schedule beside the equation. A statement that stiffness alternates between two values is incomplete if the question concerns growth over many cycles.

Contents · Next chapter

For constant k, mechanical energy is E = ½mv² + ½kx² and is conserved by the undamped phase equation. When k varies, differentiating gives dE/dt = ½k̇x² under the same ideal model.

At an instantaneous change from k_before to k_after, with x and v continuous, the energy jump is ΔE = ½(k_after − k_before)x².

Increasing stiffness at nonzero displacement adds mechanical energy. Decreasing it can remove energy. At x = 0, this ideal stiffness change makes no energy jump.

The parameter-changing agency therefore exchanges work with the oscillator. No additive force in the written equation does not mean the larger physical arrangement supplies no energy.

This distinction is central to parametric excitation. The source changes a property of the system rather than appearing as an independent right-hand-side force. We will account for that work explicitly, so amplification will not be mistaken for energy created from nothing.

Contents · Previous chapter · Next chapter

For a phase of frequency ω lasting τ, the exact solution from initial x₀ and v₀ is x(τ) = x₀ cos(ωτ) + (v₀/ω)sin(ωτ).

Velocity is v(τ) = −ωx₀ sin(ωτ) + v₀ cos(ωτ). These two expressions preserve both pieces of the initial state.

They are more informative than an amplitude alone. Two states at the same displacement but with different velocities generally produce different futures, including a different response to the same later stiffness switch.

Check τ = 0: the expressions return x₀ and v₀. Check a quarter-period, ωτ = π/2: position becomes v₀/ω and velocity becomes −ωx₀.

That quarter-period relation will make the first repeated example especially transparent. The choice is for exact arithmetic, not a recommended physical modulation schedule. A learner should understand the general phase expressions before treating the quarter-period result as a convenient special case.

Contents · Previous chapter · Next chapter

Choose a fixed reference frequency ω_r and represent the state as z = (x,v/ω_r). Both components now have length units, making the phase matrix dimensionless.

Its first row is (cos θ,(ω_r/ω)sin θ), and its second row is (−(ω/ω_r)sin θ,cos θ), where θ = ωτ. Multiplying this matrix by the starting z gives the ending z.

The reference frequency is a coordinate choice. It does not change the physical motion. In the main examples, ω_r = 1 per second, so the numerical velocity coordinate is easy to relate back to metres per second.

Changing ω_r requires rescaling the state and matrix consistently. It should not change cycle multipliers or whether motion is bounded. This is a useful later verification check.

The matrix notation is an extension route for learners ready to combine linear transformations. The first energy and quarter-period arguments can still be followed without treating matrix algebra as a prerequisite for every introductory vibration lesson.

Contents · Previous chapter · Next chapter

Set m = 1 kilogram. The low phase has k_L = 1 newton per metre, hence ω_L = 1 per second. The high phase has k_H = 4, hence ω_H = 2.

Hold the low phase for τ_L = π/2 seconds and the high phase for τ_H = π/4 seconds. Each duration is one quarter of that phase’s own natural period. At the end, switch back to low stiffness and repeat.

The cycle duration is T = 3π/4 seconds. The state is recorded immediately after the return to low stiffness. That establishes a consistent point in the repeating schedule.

These values specify a paper model, not a tested spring or feasible actuator. The exact calculation depends on the ideal linear phases and continuous state at each switch.

Contents · Previous chapter · Next chapter

With the chosen state scaling, the low quarter-period matrix L has rows (0,1) and (−1,0). The high matrix H has rows (0,1/2) and (−2,0).

The low phase acts first, so the complete-cycle matrix is M = HL. Matrix multiplication gives rows (−1/2,0) and (0,−2).

The rightmost operation acts first. Writing LH while intending low then high changes which state is being mapped from the declared cycle boundary.

The final return switch changes stiffness but not z under our ideal rule, so it does not add another state matrix. Its energy effect still belongs in the work account.

The two diagonal factors show what happens at successive cycle boundaries: the displacement coordinate changes by −1/2 and the scaled velocity coordinate by −2. A complete trajectory also contains the intervening phase motions; the cycle matrix records their net transformation over one period.

Contents · Previous chapter · Next chapter

Start first with x = 0 and v = 1 metre per second. At cycle boundaries, the scaled velocity becomes −2, then four, then −8, while displacement is zero at those particular times.

Start instead with x = 1 metre and v = 0. Boundary displacement becomes −1/2, then 1/4, then −1/8, with zero velocity at those boundaries.

The same repeated schedule therefore has an expanding direction and a contracting direction. It would be incorrect to say every nonzero initial state grows. The exactly contracting initial direction is a counterexample.

For a generic state with a nonzero expanding component, that component eventually dominates in this ideal linear model. Small perturbations away from the special contracting direction can matter over many cycles.

The mathematical instability concerns the existence of growth from arbitrarily small perturbations, not a promise that every prepared state grows or that a real oscillator continues indefinitely within the assumptions. Nonlinearity and other omitted processes can become important as motion grows.

Contents · Previous chapter · Next chapter

For the initial state x = 0, v = 1, low-stiffness energy is ½ joule. After the low quarter-period, x = 1 and v = 0, still with energy ½.

Switching from k = 1 to k = 4 at x = 1 adds ½(4 − 1)1² = 1.5 joules. The energy becomes two.

The high quarter-period takes the state to x = 0, v = −2, preserving those two joules. Switching back to low stiffness at x = 0 adds zero. The cycle ends with four times its initial energy.

All extra energy came through the stiffness change at nonzero displacement. The ideal parameter source supplied 1.5 joules in this first cycle.

On later cycles, the displacement at the energy-adding switch scales with the growing state, so the work also changes. This is why a fixed schedule can produce multiplicative growth without a fixed additive force amplitude. The written phase equation and the complete energy account are consistent.

Contents · Previous chapter · Next chapter

For initial x = 1 and v = 0 at low stiffness, energy is again ½ joule. The low quarter-period gives x = 0, v = −1.

The switch to high stiffness occurs at zero displacement and adds no energy. The high quarter-period then gives x = −1/2 and v = 0, still with ½ joule while k is four.

Returning to k = 1 removes ½(4 − 1)(1/2)² = 0.375 joule. The final energy is 0.125, one-quarter of the initial value.

The parameter source extracts energy from this phase relation rather than supplying it. That does not contradict the growing case: the states meet the same switches at different displacements.

Ask a learner to identify the position of each switch on the motion before calculating its work. The timing relative to the state is the mechanism. Looking only at how often stiffness increases and decreases would miss why their energy contributions do not cancel.

Contents · Previous chapter · Next chapter

The time-average stiffness in the main schedule is [1(π/2) + 4(π/4)]/(3π/4) = 2 newtons per metre.

Replacing the switched system by a constant stiffness of two gives a stable undamped oscillator of frequency √2. Its trajectories are bounded for every finite initial state.

That constant model does not reproduce the original cycle’s expanding factor of two. Equal time-average stiffness has not preserved the repeated dynamics.

The shortcut discards when each parameter value acts relative to displacement and velocity. Those associations determine work at the switches and the product of phase transformations.

An average model can be useful under justified approximation conditions, but the average alone does not establish those conditions. This article supplies a direct exact counterexample rather than a rule that averaging is always useless.

For independent checking, compare both models from the same initial state and at the same physical times. Do not confuse the average stiffness’s own natural period with the switching schedule’s period T.

Contents · Previous chapter · Next chapter

Now split each original phase into two equal pieces. Use low half-phase, high half-phase, low half-phase, high half-phase, then repeat.

Total low time remains π/2 and total high time π/4 within the same T = 3π/4. Both stiffness values and their total time weights are unchanged. Only the interleaving differs.

Each half-phase has angle π/4. Multiplying one high-half transformation by one low-half transformation gives S with rows (1/4,3/4) and (−3/2,−1/2).

The complete period contains two such pairs, so its matrix is S². Its rows are (−17/16,−3/16) and (3/8,−7/8), with trace −31/16 = −1.9375.

Unlike the original diagonal cycle, this interleaved schedule has bounded repeated motion under the ideal model. We will justify that statement from the trace criterion shortly. The comparison is stronger than changing a mean parameter: even the exact durations at each stiffness match while the repeated behaviour differs.

Contents · Previous chapter · Next chapter

Two linear transformations generally do not commute. For the full quarter phases, HL is diagonal with entries −1/2 and −2, while LH has entries −2 and −1/2.

Those two products correspond to different choices of where the repeating two-phase cycle begins. They have the same multipliers, although the state coordinates associated with growth are represented differently at the two cycle boundaries.

The interleaved schedule is not merely a cyclic relabelling. It replaces full holds by alternating half holds and changes the complete product itself. Its different stability cannot be removed by renaming the start of the day.

This distinction helps avoid an overstatement: changing the reporting phase of one periodic system should not change whether it is stable, while changing the physical schedule can.

For a learning exercise, give both operations without labels and ask which is a new schedule and which is a new observation point in the same schedule. The answer requires following the sequence, not only counting how many L and H symbols appear.

Contents · Previous chapter · Next chapter

Every undamped fixed-phase matrix has determinant cos²θ + sin²θ = 1. A product of such matrices also has determinant one.

For the main cycle, (−1/2)(−2) = 1. For S, (1/4)(−1/2) − (3/4)(−3/2) = 1. These checks expose an incorrect sign or missing frequency factor.

Determinant one means the two-dimensional state transformation preserves phase-space area in these coordinates. It does not mean mechanical energy is conserved across the stiffness switches.

One direction can expand while another contracts by the reciprocal factor. The product of those factors remains one, as the diagonal main example shows.

Keep mathematical invariants attached to their meaning. Treating every conserved quantity as energy would erase the parameter-source work already calculated. A good explanation can preserve area while changing the size and shape of an energy contour because the repeated map need not preserve the relevant quadratic energy form.

Contents · Previous chapter · Next chapter

For a real two-by-two cycle matrix with determinant one, its multipliers satisfy λ² − sλ + 1 = 0, where s is the trace.

If |s| < 2, the roots are distinct complex conjugates with modulus one. The matrix is diagonalizable over the complex numbers, and its integer powers remain bounded. The fixed finite phase transformations inside each period then keep the complete continuous-time response bounded as well.

If |s| > 2, the roots are real and reciprocal. One has magnitude greater than one and the other less than one, giving an expanding and contracting direction.

The main cycle has trace −2.5 and multipliers −2 and −1/2. The interleaved cycle has trace −1.9375, inside the bounded region.

This classification belongs to the exact undamped linear model. It does not supply a safety criterion for equipment or eliminate the need to examine damping, nonlinear response and parameter uncertainty in a real application.

Bounded does not mean that every coordinate decreases on every cycle, or that the state tends to rest. In the interleaved case, initial z = (1,0) becomes (−17/16,3/8) after one period. The displacement magnitude has increased from one to 1.0625. At the low-stiffness boundary, its energy has risen from 0.5 to 325/512, approximately 0.634766 joule. This short-term increase is compatible with bounded long-term powers of the cycle matrix. The matrix need not preserve the circular energy contour associated with the low phase. Distinguish bounded repeated motion from monotonic decay and from constant mechanical energy; the trace criterion establishes the first, not all three.

Contents · Previous chapter · Next chapter

At trace +2 or −2, the multipliers repeat at +1 or −1. Their magnitudes alone do not decide boundedness.

If M is exactly the identity or negative identity, repeated states are bounded. But a nontrivial repeated-root matrix can contain a Jordan contribution whose powers grow linearly with cycle count.

Therefore the casual rule all multipliers have magnitude at most one needs an additional check at these boundary cases. The strict interior condition |trace| < 2 avoids that ambiguity for our two-by-two determinant-one family.

The next original example constructs a boundary matrix explicitly. It is useful because no experimental noise or complicated numerical method is needed to expose the issue.

For learners, distinguish exponential growth, linear growth, bounded oscillation and decay along one special direction. A single word such as resonant can conceal those different mathematical behaviours. Describe what the repeated state actually does before applying a broad label.

Contents · Previous chapter · Next chapter

Retain the low quarter-phase angle π/2. Choose the high phase angle β = π + arcsin(4/5), so sin β = −4/5 and cos β = −3/5.

The corresponding high matrix has rows (−3/5,−2/5) and (8/5,−3/5). Multiplying by the low quarter matrix gives M with rows (2/5,−3/5) and (3/5,8/5).

Its trace is two and determinant one. Write M = I + N, where N has rows (−3/5,−3/5) and (3/5,3/5). Direct multiplication gives N² = 0 but N is not zero.

Therefore Mⁿ = I + nN. From initial state z₀ = (1,0), the cycle states are (1 − 3n/5,3n/5), which grow linearly in magnitude.

Both multipliers equal one. This construction demonstrates why their magnitudes alone were insufficient at the boundary. The result is exact for the supplied ideal phases; it is not an approximate instability chart for a particular device.

Contents · Previous chapter · Next chapter

Let α = ω_Lτ_L and β = ω_Hτ_H, with r = ω_H/ω_L. Multiplying the general phase matrices gives trace M = 2cos α cos β − (r + 1/r)sin α sin β.

The state scaling frequency ω_r cancels from this expression. It is therefore a useful consistency check on a derivation performed in another compatible state coordinate system.

For α = β = π/2 and r = 2, the trace is −(2 + 1/2) = −2.5, matching the original diagonal matrix. For the half phases, α = β = π/4 gives trace S = 1 − 1.25 = −0.25; squaring S gives trace S² = (trace S)² − 2 = −1.9375.

These two independent routes should agree. The trace relation is more compact than plotting many cycles, but its conditions still matter: exact constant phases, continuous state at switches and determinant-one undamped dynamics.

Contents · Previous chapter · Next chapter

Try m = 1 with low frequency two and high frequency three per second. The stiffnesses are four and nine. Hold each for a quarter of its own natural period: π/4 and π/6 seconds.

The cycle multipliers are −2/3 for displacement and −3/2 for the scaled velocity direction. From x = 0, v = 1, the first low quarter-phase gives x = 1/2, v = 0.

Increasing stiffness from four to nine there adds ½(9 − 4)(1/2)² = 0.625 joule. The initial energy was 0.5, so the high phase has 1.125. It ends at x = 0, v = −1.5, and the return switch adds zero.

The boundary energy multiplies by 2.25, consistent with the squared velocity multiplier. This changed case checks both the matrix and work reasoning with different ratios.

A learner should derive the new factors from the phase expressions, not assume that every quarter-phase pair doubles the growing state.

Contents · Previous chapter · Next chapter

For invertible phase maps, moving the observation boundary within the same periodic schedule produces a similar cycle matrix. For the two-phase case, LH = L(HL)L⁻¹.

Similar matrices have the same multipliers and trace. Their coordinate descriptions of expanding and contracting states differ, because the state has evolved between the two observation phases.

This explains why reversing which full phase is written first can leave the stability classification unchanged when it represents the same alternating cycle viewed from a different boundary.

It does not justify arbitrary rearrangement of pieces, as the interleaved example already showed. Only the appropriate cyclic change corresponds to choosing another point in the same repeating history.

For an evidence record, name the stroboscopic observation phase. A state called initial should specify whether it is just before or after a switch and which stiffness is in force. Without that label, even correct state numbers can be compared at different stages of the cycle.

Contents · Previous chapter · Next chapter

In the main cycle, the contracting initial state loses mechanical energy to the parameter-changing agency. There is no damping term in the phase equation.

For a general state, one coordinate may shrink while another grows. Observing only displacement at the cycle boundary can miss the expanding velocity component in our diagonal example.

Thus a shrinking sampled displacement does not by itself establish dissipative damping. The observation phase, state components and changing-parameter work matter.

This connects with the existing Vibrations guides on unknown forcing and sensor-location limits, but the mechanism here is distinct. We have supplied the changing parameter explicitly and derived its exact repeated map.

For a classroom check, provide only the displacement sequence zero, zero, zero from the growing initial velocity state. Ask whether the object is stationary. It is not: it moves between samples and has growing boundary velocity. A sparse observation can conceal motion even when its timestamps repeat consistently.

Contents · Previous chapter · Next chapter

The expanding main-cycle state multiplies in magnitude by two each period T = 3π/4. Its exponential rate at cycle boundaries is g = ln 2/T = 4ln 2/(3π), approximately 0.294181 per second.

The corresponding energy for that pure expanding state multiplies by four, so its logarithmic energy-growth rate is twice g. State amplitude and energy do not have the same growth exponent.

For the changed frequency pair two and three, the expanding state multiplier is 1.5 and T = 5π/12. Its rate is ln(1.5)/(5π/12), approximately 0.309753 per second. A smaller multiplier per cycle can therefore accompany a larger growth rate per unit physical time when the cycles are shorter.

These rates describe the exact mathematical schedules and appropriate expanding states. They do not establish practical growth rates for an apparatus. The comparison teaches why a multiplier should be reported with its period. Asking only which cycle has the larger multiplier can answer a different question from which state grows faster over a common duration.

Contents · Previous chapter · Next chapter

Keep the main frequency ratio r = 2 and the exact low angle α = π/2. Change the high angle to β = π/2 + δ. The general trace becomes s = −2.5cos δ.

Within −π/2 ≤ δ ≤ π/2, exponential expansion occurs when cos δ > 0.8, or |δ| < arccos(0.8), approximately 0.643501 radian. In the outer portions of that specified interval, the trace lies strictly between −2 and two and the ideal cycle is bounded.

At equality, the trace is −2. The matrix is not negative identity for this frequency ratio, so the repeated-root check matters. The boundary has a nontrivial Jordan contribution and linear growth for suitable states rather than the interior exponential behaviour.

These are statements about a phase parameter in a paper model, not operating tolerances or safe modulation margins. They hold only while the other stipulated quantities and switching assumptions remain fixed.

Now suppose δ is supplied only within [−0.1,0.1]. Since cos δ is at least cos(0.1), the trace stays below approximately −2.4875 throughout that family. Every permitted phase choice retains an expanding multiplier.

If the supplied range instead widens to [−0.8,0.8], it contains expanding, boundary and bounded choices. Selecting its midpoint δ = 0 does not establish the behaviour for every permitted case. The uncertainty statement has become too broad for a single family-wide classification.

No probability distribution was supplied for δ. The bounds therefore do not describe a percentage chance of stability. They identify which conclusions are supported uniformly by the allowed interval and which require more information.

For independent practice, give a phase interval entirely on one side of the boundary, then an interval crossing it. Ask the learner to state what is identified before plotting repeated states. A long numerical trace from one convenient parameter choice cannot settle the entire uncertain family.

Contents · Previous chapter · Next chapter

Set the low and high frequencies equal. The supposed stiffness change then disappears for fixed mass. Both phase matrices belong to the same oscillator and compose by adding their phase angles.

The general trace formula reduces to 2cos α cos β − 2sin α sin β = 2cos(α + β). Its value remains within [−2,2]. At the endpoints, the complete transformation is identity or negative identity rather than a nontrivial Jordan matrix.

All finite initial states therefore remain bounded. The switch-work expression also gives zero because k_after − k_before is zero. The matrix account and energy account recover the same constant-parameter physics.

This is a useful limiting-case test of the whole argument. A formula predicting exponential parametric growth when the two stiffness values become identical would contain a mistake or a different model assumption.

The limit also clarifies the mechanism. It is not the naming of two phases that creates amplification. The parameter difference, the state at the changes and the chronological transformation together determine the result. If a phase label changes while the physical coefficient does not, the oscillator does not receive work simply because the table has another row.

Contents · Previous chapter · Next chapter

Adding damping changes the phase equation and its transition matrices. The determinant-one and reciprocal-multiplier arguments cannot simply be carried over unchanged.

Damping can compete with energy added through modulation, but the resulting stability must be derived from the amended model. Multiplying our undamped answer by an arbitrary decay factor is not a justified replacement for that derivation.

Likewise, smooth stiffness changes require integrating the changing-coefficient equation rather than assuming instantaneous work jumps. The switch model can be a useful idealisation, but its accuracy needs evidence for the intended application.

The article’s unbounded growth is an ideal linear-model conclusion. A physical system can leave its linear range, change its parameters or respond through other mechanisms. The calculation supplies no claim of unlimited real motion.

A careful explanation states exactly which conclusion belongs to the model and which further observations would be needed to apply it. That boundary protects the useful mathematics from becoming an unsupported operating prediction.

Contents · Previous chapter · Next chapter

First check each phase at zero duration and at a known quarter-period. Confirm the chronological product and the return-to-start parameter convention.

Then check determinant, trace and a directly propagated initial state. For at least one state, reconcile the mechanical-energy change with the sum of the stiffness-switch work.

Finally compare repeated multiplication with any claimed multiplier formula. The boundary example requires the additional Jordan check; a numerical eigenvalue routine returning values near one is not enough to settle its exact behaviour.

Preserve rounding choices. Treating a value near |trace| = 2 as if it were certainly inside or outside can exaggerate what uncertain parameters establish. This article’s exact constructions avoid that measurement uncertainty; real inferred matrices would need it.

These checks test different parts of the reasoning. Agreement between two plots generated from the same incorrect phase matrix would not be independent confirmation of the physical derivation.

Contents · Previous chapter · Next chapter

For repair, begin with energy at one stiffness change. Compare the same change at zero displacement and at a turning point. Explain who supplies or receives work.

For stabilisation, use quarter-period phase maps and follow the two original initial states. Then compare the switched cycle with the constant average-stiffness model.

For extension, use the interleaved schedule, general trace relation and repeated-root boundary case. Ask the learner to distinguish a new schedule from a changed observation phase of the same schedule.

An independent task should change a frequency ratio or phase duration without naming the method. The learner should reconstruct the complete cycle before deciding whether a time-average description is adequate.

All tasks are mathematical investigations using supplied values. No apparatus needs to be driven, altered or tested. The useful outcome is an explanation that connects state, sequence, work and stability without treating any one summary as the whole dynamics.

Contents · Previous chapter · Next chapter

Can two stable fixed phases form an unstable repeated system?

Yes. The original quarter-phase product has multipliers −2 and −1/2 even though each fixed positive-stiffness oscillator is bounded.

Does no additive driving force mean no external work?

No. Changing stiffness at nonzero displacement exchanges work with the oscillator in this model.

Must every initial state grow in an unstable cycle?

No. The exactly contracting direction in the original example decays. Instability concerns the existence of amplifying perturbations.

Do equal time weights determine the cycle dynamics?

No. The original and interleaved schedules share those weights but have different repeated maps.

Are unit-magnitude multipliers always sufficient for boundedness?

No. At repeated-root boundary cases, a nontrivial Jordan contribution can produce linear growth.

Contents · Previous chapter · Next chapter

Return to Vibrations for the motion, energy and measurement foundations, and Resonance for the broader response concepts. The present work adds exact switched-cycle constructions rather than interpreting a spectral peak or fitting damping from an unknown force.

The primary mathematical reference is DaCunha and Davis, A Unified Floquet Theory for Discrete, Continuous, and Hybrid Periodic Linear Systems, submitted 24 January 2009 and checked 26 October 2026. It supplies the wider periodic-system framework. The two-phase oscillators, matrices, work ledgers, interleaving comparison and boundary example here are original educational derivations, not reproduced experiments or source figures.

For related but different vibration questions, read periodic response cancellation in a driven two-mass model, energy exchange between two freely coupled masses, unknown forcing and damping identification and mode participation from sensor locations. For the atomic-scale application, parametric excitation and trap modulation explains a separate experimental context; no trap procedure is implemented by these paper schedules.

Continue with the Education parent and How X Works when the next question asks what a sequence preserves that its separate averages lose.

Contents · Previous chapter · Return to Vibrations