Gunderson: Primer

Study document for the Graham A. Gunderson project. It adds no new verdicts; every verdict below is the arbiter's (arbiter.md). Source tags: [read] = from the two patent documents in the project folder (via the deep reads in deep/); [web] = Internet source with URL (mostly found by the Investigator, research/ and sites.md); [memory] = recalled textbook knowledge, with confidence.

Source gap (repeated from claims.md and arbiter.md). The project folder holds two patents and nothing else. Neither patent contains a measurement, a prototype description or an energy budget [read]. US 2006/0163971 has one unquantified sentence about experiments (para 0040). Every performance number in this primer comes from outside the patents: the inventor's own conference descriptions (2015, 2016), one forum witness thread, and the assignee companies' web pages. The 2016 "Magnetic Implosion Transformer" has different magnetics from either patent (resonant H-bridge drive, ferrite C-cores, versus a toroid or a bar). It shares one element with the 2012 patent: claim 1 of US 8,093,869 ends with a switched, time-varying-resistance load, and the 2016 output stage was a switched rectifier (see Section 7) [arbiter.md, QE revision]. It is related evidence, not a test of either patent.

Revised 2026-10-04 after the quality check (qe.md, verdict MINOR) and the arbiter's "QE revision". No verdict changed. Changes: a new Section 3.2b on the energy bound that fits patent 1's holed core; the corrected square-wave note in Step E; restored leads, related work and assignee statements in Sections 6, 7 and 9.

Run the worked example with: .venv/bin/python review/Gunderson/example.py

---

1. The idea in one paragraph

An ordinary generator makes electricity by moving a magnet past a coil, and the harder you pull current out, the harder the magnet pushes back (Lenz's law), so you must push harder to keep it turning. Graham Gunderson, an inventor from Spokane, Washington, proposed two ways around that brake. In the first (patent application 2006, granted 2010) a permanent magnet's field is steered back and forth inside a "soft" iron-like core by a small input coil, so the field sweeps across wires threaded through holes in the core, as if the magnet were moving when it is not. He argued that the push-back from the output current then lands on the permanent magnets rather than on the input coil, and can even help the input, so drawing more power needs less input, with the magnets supplying the energy until they wear out. In the second (granted 2012) a metal bar is made to ring like a bell at ultrasonic frequency; a coil around its middle picks up electricity because stress changes the bar's magnetism, and the coil's push-back travels back to the vibrator as a sound wave timed (90 + 180 + 90 = 360 degrees) to arrive in step and help it. Later, at conferences in 2015 and 2016, he showed transformer devices for which he and a witness reported more power out than in. The underlying effects (changing a magnetic path without moving parts, magnetostriction, resonance) are real and used every day; the question is the energy bookkeeping.

---

2. What you need to know first

2.1 Faraday's law and flux linkage. A voltage (electromotive force, EMF, in volts) appears around a closed loop whenever the magnetic flux through the loop changes: EMF = -N dPhi/dt, where Phi (webers, Wb) is the flux through one turn and N the number of turns. The flux can change because a magnet moves, because the current making the field changes, or because the material that guides the flux changes. Nothing in the law requires mechanical motion. Read: Griffiths, Introduction to Electrodynamics, ch. 7 [memory, high].

2.2 Lenz's law as energy bookkeeping. The induced current makes its own field that opposes the change that produced it. This is not an extra rule; it is what energy conservation requires. If the induced field aided the change, the change would speed up, inducing more current, without limit. In a generator the opposition is the drag on the rotor; in a transformer it is the extra primary current that flows when you load the secondary. Read: Griffiths ch. 7 [memory, high].

2.3 Magnetic circuits and soft versus hard magnets. Flux in iron behaves a little like current in wires: it follows the path of lowest "reluctance" (magnetic resistance). A soft magnetic material (iron, ferrite, amorphous metal) has high permeability that can be changed, even nearly switched off, by driving it toward saturation with a coil. A hard material (a permanent magnet, such as neodymium-iron-boron, NdFeB, or barium/strontium ferrite) holds a fixed magnetisation. So a coil on a soft core can redirect a permanent magnet's flux from one path to another. This is how magnetic amplifiers, fluxgate magnetometers and variable-reluctance pickups work. Read: Griffiths ch. 6 (magnetisation, H field); Cullity and Graham, Introduction to Magnetic Materials [memory, high].

2.4 AC power: real versus reactive. For sinusoidal voltage V and current I (root-mean-square, rms, values) with phase angle phi between them, the average power that actually leaves the source is P = V I cos(phi) (watts). The product S = V I is the apparent power (volt-amperes, VA). When phi is near 90 degrees, as in a resonant circuit, most of the energy just sloshes back and forth between inductor and capacitor (reactive power); very little is consumed. Then P is a small difference between large quantities, and a small error in phi gives a large error in P. Any circuit text (e.g. Nilsson and Riedel, Electric Circuits, AC power chapter) [memory, high].

2.5 Resonance, damping and impedance. A driven mass on a spring (or an LC circuit) responds most strongly at its natural frequency f0 = (1/2pi) sqrt(k/m). Impedance Z is the complex ratio of "push" to "flow" (voltage/current, or force/velocity). Its real part absorbs power, its imaginary part stores and returns it. Average power into an impedance driven by a force of amplitude F is (1/2) F^2 Re(1/Z). A passive load (resistor, coil, capacitor) always has Re(Z) >= 0: it can only absorb. Read: Taylor, Classical Mechanics, ch. 5 (oscillations) [memory, high].

2.6 Magnetostriction and magnetoelasticity. Ferromagnets change length slightly when magnetised (magnetostriction), and conversely their magnetisation changes when stressed (magnetoelasticity, also called the Villari effect). A bar with a bias magnet and a coil is therefore a transducer between sound and electricity, used in sonar since the 1930s-40s and in modern vibration harvesters [memory, high; the patent's own citations go back to Bell Labs and Hayes sonar patents, web: https://patents.google.com/patent/US8093869B1/en].

2.7 Standing waves in a bar. Sound travels along a bar at speed v (about 5000 m/s in iron, as the patent says [read]). A bar of length L free at both ends rings at f = v/(2L) and its multiples. A quarter wavelength of travel delays a wave by 90 degrees of phase. Read: any introductory waves text; Kinsler et al., Fundamentals of Acoustics [memory, high].

2.8 Fourier series and Parseval's theorem. A square wave is a sum of sine waves at odd multiples of the base frequency (fundamental, 3rd, 5th, ... harmonics) with amplitudes 4/(pi n). The average of the square of a sum of sinusoids at different frequencies equals the sum of the averages of each one squared: the cross terms average to zero. Read: Boas, Mathematical Methods in the Physical Sciences, ch. 7 [memory, high].

---

3. The textbook view

3.1 Flux switching with a permanent magnet (patent 1, claim 1)

Symbols: Phi = flux (Wb); N = turns; F_m = magnetomotive force of the magnet (ampere-turns); R_1, R_2 = reluctances (A/Wb) of two paths in the core; mu = permeability of the core (H/m).

A magnet with magnetomotive force F_m feeding two parallel paths splits its flux as Phi_1 = F_m/R_1 x (share), like current in two parallel resistors. If an input coil drives path 1 toward saturation, mu in that path falls, R_1 rises, and flux moves into path 2. An output wire that encircles path 2 sees dPhi_2/dt and an EMF = -N dPhi_2/dt. The magnet never moves. This is exactly what Gunderson calls "virtual motion": "The permanent magnet's apparent location ... will effectively move, vibrate, and appear to shift position when the magnetization of the soft magnetic material is modulated" [read: US 2006/0163971 para 0014]. All three reviewers and the arbiter accept this mechanism as Established [arbiter.md].

Where the energy comes from in the textbook view. The magnet is a bias, like the battery that sets the operating point of a transistor. Over one full cycle the magnet returns to its starting state, so its stored energy shows no net change and it does no net work. The power at the output is paid by the input coil, which has to do work to change the core's state. In the formal language of parametric devices (systems where one signal, the "pump", periodically changes a parameter such as inductance), the Manley-Rowe relations say the power at the output frequencies comes from the pump; a static magnet is not a pump [memory, high; general statement confirmed at web: https://en.wikipedia.org/wiki/Manley%E2%80%93Rowe_relations; the application "a static magnet does not pump" is the reviewers', per arbiter.md].

3.2 How a load shows up at the input

For two coupled coils with self-inductances L_1, L_2 and mutual inductance M, angular frequency w = 2 pi f, and a load impedance Z_L on coil 2, the impedance seen at coil 1 is

Z_in = Z_11 + (w M)^2 / (Z_22 + Z_L),

where Z_11 = R_1 + j w L_1 and Z_22 = R_2 + j w L_2 (j = sqrt(-1)). The second term is the "reflected" load. For a passive Z_L its real part is positive, so adding a load always adds real power drawn from the input. The mathematician states the same result as real input power >= real load power for any passive reciprocal network [mathematician.md; memory, high]. This linear picture fits the acoustic patent and a small-signal (linearised) view of the 2016 resonant circuit. It does not fit patent 1's holed core, for the reason in 3.2b [arbiter.md, QE revision].

3.2b Patent 1's holed core: why the picture above does not apply, and the bound that does

The geometry. In US 2006/0163971 the input coil (6) drives flux (a) round a ring-shaped (toroidal) core. The output wire threads radial holes drilled through the core's mid-plane, passing through them in alternating directions, with permanent magnets set beside the holes in a staggered pattern [read: paras 0029, 0037-0038; page image p.4]. When output current flows, it makes a reaction flux (c) that circulates in a small closed loop round each hole.

What the drawing shows. The deep read of FIG. 3 found that at all eight drawn holes the loop (c) runs against the input flux (a) on the magnet side of the hole and along (a) on the other side. The patent counts only the second side ("(c) aids (a)", para 0043) and says the first side is cancelled by the magnet flux (b) [read: page image p.4; para 0043]. The mathematician's point is confirmed: (c) aids on one side and opposes on the other, and the patent gives no reason why the two sides should count unequally for the input coil [arbiter.md, QE revision]. The patent also contradicts itself on this point. Para 0039 says the reaction is directed against the magnets and "not the input flux source (6, a)", that is, no effect on the input coil at all. Paras 0043-0044 say (c) "aids the input flux (a)", up to "strengthening and overall acceleration of the electrical input signal" [read: paras 0039, 0043-0044]. The arbiter records this as an inconsistency within the description; it does not change a verdict.

Why M is about zero. Because the holes sit on the core's mid-plane, a (c) loop either misses a cross-section of the core linked by coil 6 or crosses it twice, in opposite directions, so its net flux through that cross-section is zero. In a linear, mirror-symmetric core the ordinary mutual inductance M between the input coil and the output wire is therefore about zero, and the (wM)^2 term in 3.2 vanishes [arbiter.md, QE revision; analysis, magnetics memory high]. That does not mean the output is free. With M near zero, in a linear core the input flux and the magnet flux would simply add, and nothing would be induced in the output at all. An output appears only because the core is nonlinear: the magnets bias it partly toward saturation, so the input changes the core's permeability mu, and that shifts the magnet flux. This is parametric coupling (the input changes a parameter, mu, rather than linking the output directly). The staggered magnets make the bias unequal top to bottom, which is what lets an output appear.

Parametric is not "no back-action". At the biased operating point the core has a nonzero small-signal coupling between coils. For a core with no hysteresis, the effect of the input on the output and of the output on the input both come from the same stored-energy function, so they are equal (reciprocity). Load current therefore reflects back into coil 6. The patent's claim that the reaction lands only on the magnets would need that reciprocity to fail, and nothing offered shows that it does [arbiter.md, QE revision].

The bound that does fit: one full cycle. Symbols: i_k = current in winding k (amperes); lambda_k = flux linkage of winding k (weber-turns, i.e. turns x flux through each turn); the integral of i_k d(lambda_k) round a closed cycle (written ∮ i_k dlambda_k) = electrical energy delivered into winding k during one cycle (joules; since EMF = dlambda/dt, it equals the integral of current x EMF over one period, and dividing by the period gives the average power into that winding). Hysteresis loss is the energy turned to heat each cycle because the core's B-H curve is a loop rather than a single line; eddy loss is the heat from currents induced inside the core material. The rule:

sum over all windings of ∮ i_k dlambda_k = (hysteresis loss + eddy loss) per cycle >= 0.

In words: the core and magnets end each cycle in the same state, so the magnetic energy they store returns to its starting value. Whatever net electrical energy the windings put in over the cycle must therefore have gone into heat in the core, which can only be positive or zero. The output winding takes energy (its ∮ i dlambda is negative), so the input winding must put in at least that much plus the core loss. Hence input real power >= output real power + core loss, and the magnets supply nothing in steady state. This holds for nonlinear, saturating cores, so it does not rely on the linear formula of 3.2 [arbiter.md, QE revision; memory, high; this derivation is QE's and the arbiter's, not in the sources]. It reaches the same conclusion as the reviews by a route that suits this device.

The arbiter's proposed test for this geometry: build the toroid with the staggered magnets and again with the magnets replaced by non-magnetic spacers; measure the small-signal mutual inductance from output wire to input coil at the bias point, and the loaded input real power on the DC side. The predicted result, under the arbiter's verdict, is a small but nonzero coupling and input real power >= output real power + core loss.

3.3 Resonant input: why input power can look tiny

At resonance an LC input circuit circulates large current with voltage and current nearly 90 degrees apart. The real power taken from the supply is only what is lost or delivered. So a resonant drive genuinely reduces the supply's real power relative to the circulating VA, and the patent is right that "Power drawn from the square wave input-energy source ... will now be at a minimum" at resonance [read: para 0047]. What resonance cannot do is make the real input smaller than the real output.

3.4 Lenz reaction phase in a resistive coil (patent 2)

Let flux in the bar be Phi(t) = Phi0 cos(wt). Faraday: EMF = -dPhi/dt = w Phi0 sin(wt). With a coil of resistance R, current i = EMF/R is in phase with sin(wt), so the coil's reaction field is 90 degrees from Phi, not 180. Only a purely inductive loop (no resistance, no power consumed) gives a reaction exactly 180 degrees opposite. So the power-consuming part of the reaction is a 90-degree, damping-like force, and the 180-degree part draws no net power [physicist.md, derived from Faraday's law].

3.5 Square-wave power

A unit square wave is (4/pi)[sin wt + (1/3) sin 3wt + (1/5) sin 5wt + ...]. At certain instants all the harmonic peaks line up, and instantaneous power (proportional to the square of the summed current) is then large. But average power is the sum of the squares of each harmonic's rms value. The fundamental carries 8/pi^2 = 81.1% of a square wave's mean-square power; the whole square wave carries pi^2/8 = 1.234 times the fundamental alone, and the drive supplies that extra [example.py Part E].

---

4. Where the inventor departs

The exact point of departure. In both patents the step that departs from the textbook is the same: a statement about the direction of a magnetic field (flux arrows adding or cancelling at a place in the core, or a reaction wave arriving "in phase") is read as a statement about the direction of energy flow. Fields add as vectors; power is a product of a force-like and a flow-like quantity (EMF x current, force x velocity) averaged over a cycle. The patents never compute that product [read: deep reads, both documents]. The physicist's phrase is "flux-arrow bookkeeping, not an energy budget" [physicist.md].

Also note, in fairness and in weighing: the inventor's own account of the energy source changed over time. Patent 2006: the magnets ("motive pressure", para 0040) while also "no consumption of natural resources" (para 0036) [read]. Assignee Magnetic Power Inc. (MPI) 2008: zero-point energy [web: https://web.archive.org/web/20080514005109/http://www.magneticpowerinc.com/pdf/POWERGENIE_Technical-progress.pdf, via Investigator]. Gunderson 2018: part of the force "taken from the ambient energy" [web: https://web.archive.org/web/20180115005940/http://grahamgunderson.com/ou]. In 2016 he said of his newest device "I'm not quite sure why it works this well" [web: https://emediapress.com/shop/magnetic-implosion-transformer/, seller has a commercial stake; description signed by Gunderson]. The 2006 application's published claim 7 names a fourth possible source: "an externally generated magnetic field, such as the Earth's magnetic field" modulates the core "in substitution of, or in conjunction with" the input coil [read: pp.9-10; granted claim 8, web: https://patents.google.com/patent/US7830065B2/en]. If the Earth's field is static it does no net work over a cycle, so the cyclic bound of 3.2b covers it too; no test of this variant is on record [arbiter.md, QE revision].

Two phrases in the acoustic patent invite an over-unity reading but do not make one. The summary names the bar an "energy transfer and multiplier element" [read: US 8,093,869 p.10, col.2]; the term is used there as the bar's name, is never defined and carries no gain figure. Resonant build-up and horn amplification both "multiply" amplitude or stored energy without creating energy. FIG. 1 models a free chain of masses and springs given one push, whose push "has never been dissipated" and which moves in "an inchworm motion successively toward the right" [read: p.15, col.12]. Read with the textbook, this is one impulse J: the whole chain's centre of mass drifts at the constant speed J / (total mass), plus an internal vibration, as momentum and energy conservation require [memory, high]. The patent offers no numbers, thrust or experiment for it. The arbiter reads it as conservation-consistent wording, the seed of the later energy claim (p.20) that it does rule on (claim 3). Pages 18-19 ("increase in total momentum carried") and p.13 ("forward-thrusting") were not deep-read; whether they make a momentum claim is open [arbiter.md, QE revision].

---

5. Worked example: where could 9.5 W for an hour come from?

The reported observation. At the Energy Science and Technology Conference (ESTC), July 2016, Mark McKay, a professional engineer (PE) who has known Gunderson about ten years and is not a commercial party, wrote that two power analyzers and a Tektronix oscilloscope "agreed" on "1.53 watts in and 9.43 watts out", a coefficient of performance (COP = output/input) of 6.16, and that "an hour later the input wattage was down to zero (with four places of accuracy) with the output still at 9.5" [web: https://www.energeticforum.com/forum/energetic-forum-discussion/renewable-energy/13945-magnetic-implosion-transformer-replication]. The device: a 12.7 ampere-hour (Ah) gel battery, an H-bridge (four-transistor switch) driving a resonant step-up to about 800 V peak, a transformer of two ferrite C-cores "beefed up with about 20 Manganese Ferrite PM's (2" x 3" x 0.5")", about 50 kHz, a 12 V lamp load [same source]. The quality check re-fetched page 1 of this thread on 2026-10-04 and confirmed the readings, battery, magnet, voltage, frequency and instrument details quoted here against the post by "Spokane1" (McKay) of 11 July 2016 [web: same URL; qe.md]. This is still one witness's account, documented in one forum thread; no second source has been found. A second attendee, "Reiyuki", recorded about 3.5 W in and 8.5 W out, with results depending on a phase setting [same thread].

We test three candidate sources: the magnets, the battery, and the measurement.

Step A: could the magnets be the fuel?

Energy delivered: E_out = P t = 9.5 W x 3600 s = 34,200 J.

Magnet volume: one magnet is 2 x 3 x 0.5 in = 3 in^3. With 1 in = 2.54 cm, 1 in^3 = 16.39 cm^3, so one magnet is 49.16 cm^3 and twenty are 983 cm^3 = 9.83 x 10^-4 m^3.

Upper bound on magnet energy. The maximum energy product (BH)max of a magnet material (units: tesla x ampere/metre = joule per cubic metre) measures how much field energy it can sustain outside itself. Multiplying (BH)max by the volume gives a very generous ceiling on what you could ever get by using the magnet up completely (real extractable energy is less):

(Values: [web: https://en.wikipedia.org/wiki/Neodymium_magnet, checked by the arbiter: up to ~512 kJ/m^3 theoretical, table range 200-440, ferrite about 1/18 of NdFeB]; ferrite 25-35 kJ/m^3 [memory, high].) Even the impossible best case runs the lamp for under a minute, and the magnets were not reported demagnetised. A side note: the witness wrote "Manganese Ferrite PM's"; manganese-zinc ferrite is normally a soft material, so the actual magnet material is unknown [arbiter.md]. No conclusion here depends on it, because the table covers every candidate.

Conclusion of Step A. The magnets cannot be the energy source for this output. This is the "Refuted by calculation" part of claim 2 (magnets as the source).

Step B: could the battery hide it?

Battery energy: 12.7 Ah x 3600 s/h = 45,720 coulombs; x 12 V = 548,640 J (152 Wh). One hour at 9.5 W is 34,200 J, i.e. 6.2% of the battery. A battery voltage reading over one hour would barely move either way. So a one-hour run cannot by itself tell battery energy from new energy. That is why the arbiter's decisive tests are an integrating charge counter (coulomb counter) on the battery over many hours, and a closed loop with the battery removed. McKay himself wrote that, if the instruments were right, Gunderson "should be able to loop the system and have a self runner, but he ran out of time" [web: same thread].

Step C: how big a measurement error would it take?

Real power P = S cos(phi), with S = V_rms I_rms the apparent power. Differentiate: dP/dphi = -S sin(phi). Near phi = 90 degrees, sin(phi) ~ 1, so a 1-degree (0.01745 rad) error shifts the reading by about 0.01745 S watts: 1.75 W at S = 100 VA, 8.7 W at S = 500 VA.

Suppose the true input equalled the output (9.43 W, COP = 1). What phase error would make it read 1.53 W? Solve cos(phi_true) = 9.43/S and cos(phi_read) = 1.53/S. The circulating apparent power S of this device was never published, so S below is ASSUMED to show the sensitivity:

One degree at 50 kHz is 1/(360 x 50,000) s = 55.6 ns, about the delay of eleven metres of coaxial cable at roughly 5 ns per metre, or of a current probe with a different internal delay from the voltage probe [memory, high]. So if the input circulates a few hundred VA, a skew of tens of nanoseconds would explain the whole result; if it circulates only tens of VA, the needed error is large and an artefact becomes hard to believe. Reported narrow 1-2 kV spikes at the output switch [web: same thread, Reiyuki] add a second risk: bandwidth limits on the output measurement.

What Step C shows and does not show. It shows the measurement is sensitive in exactly the regime the device runs in. It does not show an error happened: the witness says two analyzers and a scope agreed, which lowers (but, if they shared probes or a measurement point, does not remove) this possibility [arbiter.md]. The negative input readings (Gunderson's -0.25 W; Reiyuki's -1 W at a 140-degree setting) are what phase error looks like, and also what genuine return of stored reactive energy looks like during part of a cycle. Only a phase-independent measurement decides it: count the DC charge leaving the battery.

Step D (shorter): why "less input as more output is drawn" can be honestly observed

This re-derives the mathematician's model, and explains a real effect behind the patents' wording. Take a mass m = 1 g on a spring k = 10^6 N/m (resonance 5033 Hz), mechanical loss coefficient c = 0.05 N s/m, driven by a force of fixed amplitude F = 1 N at resonance. A pickup coil with transduction constant G = 2 V s/m (equal to N/A), inductance L = 1 mH and load R adds a mechanical impedance Z_e = G^2/(R + j w L). (All values ASSUMED, illustrative.) Velocity v = F/(c + Z_e); input power P_in = (1/2) F^2 Re(1/(c + Z_e)); load power (1/2)|v|^2 Re(Z_e).

Loading the coil lowers the input power (10 W to 5.7 W) at fixed drive force, because the extra damping reduces the motion at resonance. An experimenter watching the drive meter would truthfully see "the input dropped when I connected the load". But the load always gets less than the input, and the books balance to rounding (script asserts P_in = P_load + loss). If the coil's reaction really "aided" (flip the sign of Z_e: negative resistance), the same model gives 21.4 W in and 135.7 W out (57.1 W to the load plus 78.6 W of mechanical loss) — a net energy source, which a passive coil and resistor cannot be because Re(Z_e) = G^2 R/(R^2 + w^2 L^2) >= 0 for every R and L. This is the conventional effect the arbiter rated Established for claim 3, and the energy-gain reading it rated Refuted by calculation.

Step E: square wave (claim 5)

From example.py: the fundamental carries 0.8106 of the square wave's mean-square power; harmonics 1 and 3 together carry 0.9006; harmonics 1, 3, 5 together carry 0.9331; all of them 1.000. A square wave delivers pi^2/8 = 1.2337 times the power of its fundamental alone, and the drive supplies every bit of it. A numeric check confirms the cross terms average away: the mean of (sin t + sin 3t / 3)^2 equals 1/2 + 1/18 = 0.5556. (Note for the record: arbiter.md first gave "0.9015" for harmonics 1, 3, 5; that was a slip, and an earlier version of this primer compounded it by calling 0.9015 the share of harmonics 1 and 3. The correct shares are 0.9006 for 1 and 3 and 0.9331 for 1, 3 and 5, as the mathematician had it; the arbiter corrected this in its QE revision. No verdict depends on it.)

---

6. Evidence table

---

7. What a learner should take away

What is solid. - Changing a magnetic path without moving anything does induce EMF. Gunderson's "virtual motion" is a vivid and correct picture of flux switching, the principle of magnetic amplifiers and fluxgates, and of the Motionless Electromagnetic Generator (MEG, US 6,362,718), which Google lists as a similar document [web: Google Patents via Investigator]. - Resonance really does make the supply's real power small compared with the energy circulating in the circuit; para 0047 of the first patent is right. - Magnetostrictive transduction with a bias magnet works, and the acoustic numbers in US 8,093,869 are correct (quarter wave 1 cm at 125 kHz and 5000 m/s; electromagnetic quarter wave ~600 m). Mainstream magnetostrictive harvester patents (US DOE, Oscilla Power, Sumitomo Riko, Mitsumi) cite it [web: https://patents.google.com/patent/US8093869B1/en]. - Opposed windings do cancel inductance (L1 + L2 - 2M), a real matching benefit.

What is wrong, and why it is instructive. - The central error is treating the direction of a field as the direction of energy flow. Learn to always ask: what is the product (EMF x current, force x velocity), and what is its average over a cycle? - The 180-degree Lenz reaction is the reactive (no-power) part; the power-absorbing part is 90 degrees. Phase arithmetic that ignores this proves nothing about energy. - A permanent magnet is a bias, not a fuel: Section 5A shows how quickly one would be "spent" if it were. - The inventor's proposed test (instantaneous v x i at quadrature, para 0048) measures reactive circulation, not real input. The test design matters as much as the device. - Counting only half of a field loop. In patent 1 the reaction flux circles each hole, opposing the input on one side and aiding it on the other; the patent counts only the aiding side (para 0043) and elsewhere says the reaction does not reach the input coil at all (para 0039). When a loop of flux is involved, ask what its net linkage with each winding is (Section 3.2b). A reader who takes para 0043 literally for the free side alone is reading the text as written; the reading fails only because the opposed half is dropped [arbiter.md, QE revision].

What is genuinely open. - The 2016 readings (COP 6.16, zero input for an hour) are reported instrument readings by an engineer, never independently tested. They are not refuted by any experiment. The arbiter judges a measurement artefact most compatible with known physics; a coulomb-counted multi-hour run or a closed loop would decide it, and could be done by a careful hobbyist. - The switched load: a thread from 2012 to 2016. Claim 1 of US 8,093,869 (granted 2012) ends with a load of "nonlinear time-changing resistance, by including semiconductor switching elements for purposes of periodically modulating the current" [read: page image p.22]. The description nowhere explains this element [read: deep read]. The 2016 transformer's output was a silicon-carbide transistor (SiC-FET) synchronous rectifier "opened briefly near the phase transition", and the witness Reiyuki reports that the COP depended on the phase of that interruption [web: research/2026-10-04-energeticforum.md]. So the 2012 granted claim already recites the element the inventor later tuned his demonstration on. Two things follow. First, the 2016 device is not wholly unrelated to the patents. Second, switching at a chosen phase is exactly where phase and bandwidth measurement errors concentrate, which is why a measurement on the DC side, independent of phase, is the test that decides. A time-varying resistor with R(t) >= 0 still only absorbs energy, so the element does not by itself imply gain. Open: was it added in response to the patent examiner, and does the 2016 behaviour depend on it? The file wrapper for application 12/327,719 would show when and why it was added; a run that varies the rectifier timing while counting DC energy in and out would show the second [arbiter.md, QE revision]. This rests on one forum witness for the phase dependence. - The record goes dark after 2016. Chava announced "several prototypes", "an additional, very broad patent application" and 30 more patents in three years; Gunderson filed a provisional in July 2016 (one source, McKay); yet no later patent family, data or test exists, and US 7,830,065 lapsed for an unpaid fee in 2022 while US 8,093,869 stays in force to 2029 [web: Google Patents]. Competing explanations (arbiter.md): (a) the effect did not survive validation — fits "still validating" (2016) and the softened "believed to" grant text, but silence alone cannot show it; (b) cost — fits the unpaid fee and part-time work, against: fees on the other patent were paid; (c) trade secrecy/NDA — MPI offered NDAs, against: an investment-seeking firm would show data; (d) funding loss or internal dispute — one anonymous comment that Mark Goldes was "ejected" from Chava, repeated re-registrations of Chava; (e) suppression — no evidence for or against found. Two more pieces of evidence: a thread on overunityresearch.com, "Graham Gunderson Energy conference High COP demonstration" (Chet K, 11 Jul 2016), asks why Gunderson "abandon[ed] a device that claimed high COP" [web: sites.md, search snippet; the page needs a login and was not read]. It is one poster's question, not an answer; it fits (a), and also (c)-(d), and does not exclude (b). And because the 2012 claim already recites the switched load, the "dark period" is dark for published data, not necessarily for the idea, which favours (a) or (c) over a simple loss of interest without deciding between them [arbiter.md, QE revision]. Background bearing on (c)-(d): an aboveunity.com snippet says "Graham Gunderson worked for Mark Goldes for many years" [web: research/2026-10-04-naturalphilosophy-wiki.md; one source, not checked further], which ties Gunderson's "paid verifier" statement below to the MPI / Chava circle; and Gunderson sat on the 2016 ESTC panel with Lindemann, Babcock and Moray King [web: research/2026-10-04-emediapress.md, search snippet]. Records that would decide: the USPTO file wrappers for applications 11/336,337 and 12/327,719, the 2016 provisional's status, Chava filings in Florida and Nevada, a 2009 FOIA entry on Magnetic Power Inc (agency unidentified), forum pages 2-4, the replies in the overunityresearch thread, and Gunderson's 2019 video. - Gunderson also reports that, as a paid verifier, he found one system "verifiably overunity", a Floyd-Sweet-VTA-like device that charged an electric car's battery bank in a week [web: https://emediapress.com/shop/overunity-disclosure/]. Unnamed and undated; no second source found. A lead, not evidence about these patents. - Assignee product plans (interested-party statements). MPI's 2008 report describes a "constant voltage" output signature and plans for 1 kW modules and 75-150 kW car units; Mark Goldes' 26 March 2007 proposal set two 1 kW Magnetic Power Modules (5" x 5" x 16") against the GM Volt [web: research/2026-10-04-wayback-magneticpowerinc.md]. These are the assignee's statements, with a commercial stake and no data; no test of such modules was found. - Related work by others (leads, no verdict on them here). Tae Chang N.E.T., "Sound wave resonance generator", US20110233941A1, priority 3 Dec 2008, the same day as Gunderson's non-provisional filing for the acoustic patent; no evidence of contact either way. Patents citing US 7,830,065: Southeast University's "time-varying magductance" generator (US12456936B2 / CN115173664B, 2022), an academic group building on flux modulation and the strongest mainstream-adjacent follow-on found; Lovshin's Earth-magnetic-field energy patents (compare the Earth-field claim in Section 4); Shaikh GB2446656A [web: sites.md; these documents were not read, and the shared idea is inferred from the Investigator's notes]. Forum witnesses compare the 2016 device with E.V. Gray's devices (McKay) and with the Russian TK / Kapanadze / Akula circuits (Reiyuki) [web: sites.md]; these are comparisons by enthusiasts, not tests [arbiter.md, QE revision]. - Open: do pp.18-19 ("increase in total momentum carried") or p.13 ("forward-thrusting") of US 8,093,869 make a momentum or thrust claim? Not yet deep-read.

---

8. Check yourself

1. A coil on a soft ferrite core drives part of the core into saturation 50,000 times a second, redirecting a permanent magnet's flux through a pickup loop. Over many cycles, which supplies the energy delivered to a lamp on the pickup: the magnet or the drive coil? Why?
2. A resonant input circuit has V_rms = 200 V, I_rms = 2 A and phase angle 89 degrees. What real power does it draw? What would a meter read if its current channel lagged by an extra 0.5 degree?
3. A coil on a vibrating bar has resistance only (no inductance). Is its reaction field 90 or 180 degrees from the bar's flux? Which of those two parts carries power?
4. How long could 50 cm^3 of NdFeB at (BH)max = 400 kJ/m^3 run a 5 W load if it were somehow used up completely as fuel?
5. In the Step D model, connecting a 100-ohm load made the input power fall from 10 W to 5.7 W. Does that support the patents' claim that loading "aids" the input? What number tells you?

---

9. Further reading

Mainstream. - D. J. Griffiths, Introduction to Electrodynamics, ch. 6 (magnetic fields in matter) and ch. 7 (electrodynamics: Faraday's law, inductance, energy in fields) [memory, high]. - J. R. Taylor, Classical Mechanics, ch. 5 (oscillations, resonance, damping) [memory, high]. - M. L. Boas, Mathematical Methods in the Physical Sciences, ch. 7 (Fourier series, Parseval) [memory, high]. - B. D. Cullity and C. D. Graham, Introduction to Magnetic Materials (hard and soft magnets, energy product, magnetostriction) [memory, high]. - L. E. Kinsler et al., Fundamentals of Acoustics (standing waves in bars) [memory, high]. - J. M. Manley and H. E. Rowe, "Some general properties of nonlinear elements", Proc. IRE, 1956 (parametric power flow) [memory, high]; overview at https://en.wikipedia.org/wiki/Manley%E2%80%93Rowe_relations [web]. - Any power-electronics measurement guide on probe deskew and power-analyzer bandwidth (the Step C problem).

The inventor's own key works. - G. A. Gunderson, US 2006/0163971 A1, "Solid State Electric Generator" (2006), granted as US 7,830,065 B2 (2010) [read; web: https://patents.google.com/patent/US7830065B2/en]. - G. A. Gunderson, US 8,093,869 B1, "Apparatus for generating electricity utilizing nondestructive interference of energy" (2012) [read; web: https://patents.google.com/patent/US8093869B1/en]. - ESTC 2015 "Overunity Disclosure" and schematic video; ESTC 2016 "Magnetic Implosion Transformer" [web: https://emediapress.com/shop/overunity-disclosure/ ; https://emediapress.com/shop/magnetic-implosion-transformer/] (seller has a commercial stake). - grahamgunderson.com/ou, archived 2018 [web: https://web.archive.org/web/20180115005940/http://grahamgunderson.com/ou]. - Witness thread: energeticforum 13945 [web: URL in Section 5]. - Related: Floyd Sweet's Vacuum Triode Amplifier (VTA), which Gunderson names as the root of his "motional field" idea; the MEG (US 6,362,718); E.V. Gray and the TK / Kapanadze / Akula circuits (named by the forum witnesses); Tae Chang N.E.T., US20110233941A1; Southeast University, US12456936B2 / CN115173664B; Lovshin's Earth-field patents; Shaikh GB2446656A (see Section 7); compilation at http://www.rexresearch.com/gunderson/gunderson.htm [web]. - Secondary: Manning and Manewich, Hidden Energy (2019), has a section on Gunderson [web: sites.md; not read for this review; obtain a legitimate copy].

---

Answers

1. The drive coil. The magnet returns to the same state every cycle, so its stored energy has no net change and it does no net work; it only sets the operating point. The drive must do work to push the core through its magnetisation cycle, and over each full cycle the electrical energy put into all windings equals the core's heat loss, so the drive must supply the pickup's output plus that loss (Section 3.2b; this cyclic bound holds even for a saturating core, where the simple reflected-load formula of 3.2 does not apply). Over minutes, the magnet's whole energy content would be far too small anyway (Section 5A).
2. S = 400 VA; P = 400 cos(89 deg) = 400 x 0.01745 = 6.98 W. At 89.5 degrees, P = 400 cos(89.5 deg) = 3.49 W: a half-degree skew halves the reading.
3. 90 degrees. Current i = EMF/R follows -dPhi/dt, which is a quarter cycle from Phi. That 90-degree part is the damping, power-absorbing part. A 180-degree reaction would come from a purely inductive (lossless) loop and would consume no average power.
4. Ceiling energy = 400,000 J/m^3 x 50 x 10^-6 m^3 = 20 J. At 5 W: 4 seconds.
5. No. The load received 2.40 W out of 5.69 W in (42%), and input = load + mechanical loss exactly. The input fell because loading damps the resonance at a fixed drive force, not because the reaction aided the source. "Aiding" would require load power above input power, which needs Re(Z_e) < 0, impossible for a passive coil and resistor.