Applications / Experiments read through FDT

Milligram gravity read across a torsion balance

Zheng-Jie An, Wen-Hai Tan, Wen-Can Dong, Cheng-Gang Qin and nine colleagues measure the gravitational interaction between two 7.6 mg silicon wafers with a miniature torsion pendulum. An attractor of sixteen source wafers turns so that each faces the test wafer in sequence, putting the signal at the sixteenth harmonic of the rotation, and a feedback loop holds the pendulum untwisted while its voltage reads the torque. At five face-to-face separations from 0.654 to 1.554 mm the measured torque falls with distance and agrees with the Newtonian torque the record calculates from the measured geometry; at the closest, the measured 11.16 aN m and the calculated 11.17 aN m differ by −0.01 ± 0.59 aN m, and the record determines G = 6.67(35) × 10⁻¹¹ m³ kg⁻¹ s⁻², agreeing with the CODATA-2022 value within the 1σ uncertainty of the measurement. FDT returns the measurement from the same source and data. With the same stress tensor the field equation returns general relativity’s curvature, and at this compactness the finite-body Newtonian torque is its leading weak-field reading: one wafer read as a spherical reference has αext = 1.1 × 10⁻²⁹ at 1 mm, deep in the general-relativistic regime, and the flow moves inward there at 1.0 μm/s. The pendulum’s two ends about the fibre are neighbours: a common acceleration cancels from the axial moment, and the balance reads the change of the field across its own extent. The record’s geometric factor is the calculated harmonic torque per unit reference constant and arm, so its G is the reference value times the measured-to-calculated torque ratio, 0.999 ± 0.053. Of five apparatus contrasts, two read the same mean in both theories at matched inputs, and three turn on the pairs a settlement joins, the mass-shadow response and the charge-to-force map, which enter no number here; none is counted as a prediction.

The record

What the article reports

The pendulum hangs from a 76-cm-long, 10-μm-diameter tungsten fibre and weighs about 243 mg with its reflector and fastenings. It is a symmetric structure with a silicon wafer at each end, the test mass and the counterbalance: the test mass measures 2.598 × 2.597 × 0.487 mm³ and has a mass of approximately 7.6 mg, the pendulum is 49.986 mm from end to end, and the average force arm from the test mass centre to the fibre is 23.694 mm. A larger wafer at its centre, 8 × 8 mm², reflects an autocollimator’s beam for the angular readout. The attractor, 38 mm across and 0.499 mm thick, is a 16-fold symmetrical structure of identical silicon source wafers, each with the test mass’s dimensions by design, connected to a rotary stage so that each source wafer in sequence faces the test mass; the Newtonian signal then falls at the sixteenth harmonic of the rotation, far from the fundamental disturbance frequency. Pendulum and attractor are fabricated integrally by plasma deep silicon etching, the system sits in a vacuum of about 10⁻⁵ Pa, and a 50-μm-thick gold-coated silicon membrane between test and source prevents direct electrostatic coupling, the source-to-membrane separation fixed at 0.268(3) mm and the pendulum-to-membrane gap at least 0.336 mm, where the record describes the Casimir and squeeze-film effects as negligible.

Patch potentials leave a residual electrostatic potential difference that introduces a negative stiffness at small separations, so two differential capacitive actuators beside the counterbalance, run by a proportional-integral-derivative algorithm, provide weak rotational control: the closed loop keeps the pendulum untwisted, the feedback voltage directly reflecting the torque acting on it. Free oscillation with the membrane about 6 mm away gives a natural period of 49.50(5) s and a quality factor of 1786(20); the moment of inertia, calculated from the measured geometry, is 12.61(1) × 10⁻⁹ kg m², and the fibre’s torsional constant is determined as 203.1(5) pN m/rad. The feedback gain is 33.4(5) fN m/V and the electrostatic stiffness 19.4(2.7) pN m/rad at 0.654 mm, the relative uncertainty of the transfer function at the signal frequency is estimated at about 2%, and the torque and force noise floors near 30 mHz are approximately 0.15 fN m/√Hz and 6 fN/√Hz. The sixteenth-harmonic component of the feedback voltage is extracted after a third-order polynomial fit removes the low-frequency drift, and divided by the transfer function.

The data runs at the separations 0.654, 0.854, 1.054, 1.354 and 1.554 mm last about 40 to 100 h each. The measured torques, 11.16, 9.80, 7.47, 6.16 and 4.94 aN m, are indicated near the data points of the record’s Fig. 3a, whose error bars at 2σ incorporate the statistical and other systematic errors, beside a red band of the Newtonian torque calculated from the measured geometric parameters. The Newtonian torque is a numerical integration over all volume elements of the pendulum assembly and all sixteen source masses with their supporting struts, using the measured geometries and densities of each component, and at 0.654 mm it is 11.17 ± 0.26 aN m in torque and 471 ± 11 aN in force, the spokes alone contributing 0.53 aN m, about 4.7%. Table 1’s total at that separation is 0.59 aN m at 1σ: 0.26 of Newtonian calculation, 0.46 of other systematic errors (electrostatic 0.27, magnetic 0.14, temperature 0.24, transfer function 0.22, voltage calibration 0.11) and 0.27 statistical with 100 hours of integration, which corresponds to a force uncertainty of 25 aN for the effective force arm of 23.694(3) mm; the combined uncertainty is 1.2 aN m in torque and 50 aN in force at 2σ.

At the closest separation the record infers G from the measured force as Fexp divided by its geometric factor, which it writes as the integral of ρ₁ρ₂/|𝐫₂ − 𝐫₁|² over the two volumes: with Fexp = (471 ± 23) aN and a factor of (7.06 ± 0.16) × 10⁻⁶ kg²/m² it deduces G = 6.67(35) × 10⁻¹¹ m³ kg⁻¹ s⁻² (1σ), agreeing with the CODATA-2022 value of 6.67430(15) × 10⁻¹¹ m³ kg⁻¹ s⁻² within the 1σ uncertainty of the measurement, a 5.3% relative uncertainty. Each of its 46 single determinations corresponds to 13 periods, about 7800 s, of source-mass rotation. With the rotation period extended from 600 to 3600 s the signal moves from 26.667 to 4.444 mHz, and 15 measurement sets of 13 full rotations give 11.41 ± 0.46 aN m, which within 1σ agrees with the 11.16 ± 0.27 aN m at 26.667 mHz, statistical uncertainty only. An identical measurement at about 3 mm shows a prominent signal only at the drive fundamental and no discernible signal at the sixteenth harmonic.

The record bounds each background at the closest separation. A maximum membrane displacement of 324 pm gives a parasitic electrostatic torque noise of at most 0.27 aN m. Magnetic coupling, modelled by integrating the dipole-dipole interaction energy with the remanent magnetization of a silicon sample, (2.4, 3.9, 5.6) mA/m, and the magnetization the ambient field induces, gives a torque below 0.14 aN m, and horizontal field fluctuations a torque noise below 0.02 aN m. A temperature response coefficient of −114.2 ± 5.7 fN m/°C with fluctuations of at most 2.1 μK at the signal frequency gives below 0.24 aN m. Its Yukawa bound is about two orders of magnitude weaker than its earlier work with two 8.7 g tungsten blocks, and the authors place its improvement at λ ~ 10 μm for a separation decreased to about 15 μm with the sensitivity and other systematic effects maintained. Its Discussion places the source and test masses of proposed tests of gravitation-induced quantum effects at 10⁻¹⁴ to 10⁻¹² kg, and expects a torque resolution of about 10⁻²¹ N m from an optimized balance integrated for 30 days.

The record’s values, each with the verb the record gives it.
QuantityValueIts verb
Test and source massestwo silicon wafers, 7.6 mg each; test mass 2.598 × 2.597 × 0.487 mm³report a measurement of the interaction between
Pendulum length; effective force arm49.986 mm; 23.694(3) mmhas; from the test mass centre to the fibre
Attractor16 source wafers; 38 mm across, 0.499 mm thickis designed as
Face-to-face separations0.654, 0.854, 1.054, 1.354 and 1.554 mmis measured at
Measured torques, Fig. 3a11.16, 9.80, 7.47, 6.16 and 4.94 aN m, bars at 2σindicated near the data points
Newtonian torque and force, 0.654 mm11.17 ± 0.26 aN m; 471 ± 11 aNyields
Spokes alone, 0.654 mm0.53 aN m, about 4.7% of the totalcontribute
Total uncertainty, 0.654 mm0.59 aN m (1σ); 25 aNTable 1; corresponds to
Combined uncertainty1.2 aN m; 50 aN (2σ)is characterized with
Signal frequencies26.667 mHz; 4.444 mHzset, with a 600 s period; extended, with 3600 s
Torques at the two frequencies11.16 ± 0.27; 11.41 ± 0.46 aN m, statisticalagrees within 1σ
Experimental force; geometric factor471 ± 23 aN; (7.06 ± 0.16) × 10⁻⁶ kg²/m²is utilized to obtain the G value
Gravitational constant6.67(35) × 10⁻¹¹ m³ kg⁻¹ s⁻² (1σ), 5.3%is deduced; agreeing with CODATA-2022
Natural period; quality factor49.50(5) s; 1786(20)yielding
Moment of inertia; fibre constant12.61(1) × 10⁻⁹ kg m²; 203.1(5) pN m/radis calculated; is determined
Feedback gain; electrostatic stiffness33.4(5) fN m/V; 19.4(2.7) pN m/radis determined
Transfer-function uncertaintyabout 2%is estimated
Noise floors near 30 mHz0.15 fN m/√Hz; 6 fN/√Hzachieves
Electrostatic, magnetic, field-noise and temperature terms≤ 0.27, < 0.14, < 0.02 and < 0.24 aN mis estimated; calculated

The FDT reading

I-⁠1 the pair, I-⁠3 the same curvature, I-⁠3a read between neighbours

There is one substance, the omnium: matter, energy and spacetime are three readings of it, and the chain E/m = c² = 1/(ε₀μ₀) = (d/t)² is one state read four ways, with c the strand rate, derived from the omnium’s two responses (0.1, 0.1a, 0.1d). Every step is a system-wide settlement, many-to-many and global, never pair by pair (0.1a). The universal force law and its ceiling are

F = Fmaxα₁α₂,   Fmax = c⁴/4Gthe universal force law (I-⁠1)

and on two separate spherical supports, each exterior reading taken across the centres, αi = Rs,i/dcc, the law reduces to Newton’s pair law exactly, I-⁠4’s layer L1:

F = (c⁴/4G)(Rs,1/dcc)(Rs,2/dcc) = G𝓜₁𝓜₂/dcc²exact on its domain

A face spacing between flat wafers gives neither a centre separation nor a spherical exterior, so the wafers of this record are read through the field equation, which stands unchanged with α downstream of its solution (I-⁠3):

Gμν = κTμν,   κ = 8πG/c⁴the same curvature

With the same stress tensor, cosmological terms, initial and boundary data and gauge, FDT returns the curvature general relativity returns. In the weak-field, slow-motion limit, I-⁠4’s layer L2, the field equation is linear in its source, and the potential of the whole rotated attractor is the volume integral of its density: one solved field of the whole system acting on the whole pendulum, the system-wide settlement read at L2 (0.1a). The element pairs of the record’s integration are the terms of that linear solution, never separate responses. The force on a pendulum element, the moment along the fibre about a point 𝐨 on it, and that moment’s harmonic over a turn of the encoder phase φrot are

d𝐅pair = G𝐛pair/rpair³ dmpendmsrc𝒩ax = 𝐳fib·∬(𝐫pen − 𝐨) × d𝐅pair𝒞ax = (1/π)∫02π𝒩ax(φrot) e−i𝖧ordφrot dφrot,   𝖧ord = 16the finite-body torque and its harmonic

with 𝐛pair the vector from a pendulum element at 𝐫pen to an attractor element, rpair its length and 𝐳fib the fibre’s positively oriented unit vector. Each element takes its own lever: the test and counterbalance forces need not be opposite for their axial moments to have opposite signs, and collapsing the assembly to one lever before integration erases those signs. Moving 𝐨 along the fibre leaves the axial component unchanged.

A pattern’s own window in the gravitational reading is its exterior compactness, with no numerical G (0.1c):

αext = 𝓜c²/(2Fmaxrf) = Rs/rf,   v = c√αextthe window and its flow

For one 7.6 mg wafer read as a spherical reference, Rs = 1.1 × 10⁻³² m and αext = 1.1 × 10⁻²⁹ at rf = 1 mm. That field-point reading lies deep in the general-relativistic regime of the A-⁠1 exterior branch, whose seam αext = 1/2 sits at rf = 2Rs (0.3a), and the small number is a correct reading of a laboratory pattern (I-⁠1). The bookkeeper reads the window as a flow: patterns falling from rest far away move inward at c√αext and fall at G𝓜/rf², so at 1 mm from the wafer the flow moves inward at 1.0 μm/s and falls at 5.1 × 10⁻¹⁰ m s⁻². The geometry holds at laboratory size; the encoder’s rate is imposed stage motion.

Moving at constant speed, on a pattern’s own clocks and rulers, through a density gradient is gravity: kept time and thread length rescale together by the clock factor 𝔪, and the bookkeeper reads acceleration toward slower clocks (0.1c),

∇𝒱𝒱μ = −𝔓μν∂ν ln 𝔪,   𝔓μν = c²gμν + 𝒱μ𝒱νthe extended equivalence principle

with 𝒱 a pattern’s four-velocity on its own clock. A single local reading gives no tide: curvature is read between neighbours, by differencing the accelerations of neighbouring patterns or comparing one with a reference path (I-⁠3a). On this balance the two ends and their common suspension are the neighbours. A common acceleration cancels from the axial moment when the pendulum’s first mass moment about the fibre vanishes, and that cancellation is the comparator’s zero of Use 7: what survives is the change of the field across the pendulum’s extent, read through the signed moments. The supported pendulum’s yaw is neither one geodesic of free fall nor a direct measurement of a local Riemann component, and the feedback holds it: a nonzero compensating voltage at the signal frequency requires a residual yaw harmonic, which the loop suppresses. The second figure draws the cancellation.

The record prints its geometric factor as the unsigned inverse-square integral over the two volumes. On the test wafer and the facing source wafer alone, at the 0.654 mm gap with the dimensions and density of Table 1, that integral is 2.2 × 10⁻⁵ kg²/m², 3.1 times the printed value, and every further volume element raises it. The printed value is the calculated harmonic torque per unit reference constant and arm,

Jgeom = 𝒩calc/(Grefℓarm) = (7.06 ± 0.16) × 10⁻⁶ kg²/m²the printed factor, identified

with 𝒩calc = 11.17(26) aN m, Gref = 6.67430(15) × 10⁻¹¹ m³ kg⁻¹ s⁻² and ℓarm = 23.694 mm, so the closest-gap quotient is

Ginf/Gref = Fexpℓarm/𝒩calc = 11.16/11.17 = 0.999 ± 0.053X-⁠1, in sample

The printed scalar alone does not fix the harmonic: a source whose whole mass is axisymmetric about the stage axis keeps a positive scalar integral while its sixteenth torque harmonic vanishes. The quotient’s uncertainty follows from Table 1. The statistical and other-systematic rows over the arm give the force 22.5 aN, printed as 23 aN; the Newtonian-calculation row gives the factor 0.26/11.17 = 2.3%; the arm cancels from the quotient and the remaining inputs sit in separate rows, so the covariance term vanishes, and the sum returns the record’s 5.3% and 0.35; rounding the force’s uncertainty to 23 aN first gives 0.36. Multiplied by its denominator the quotient returns the measured input identically: X-⁠1 restates the closest-gap comparison and tests neither a held-out separation nor a second force law.

A magnetic field is the helical density gradient of the omnium (P-⁠5), and a pattern whose rest energy depends on where it sits in that twist takes the energy as its clock factor, the helical window (0.1c):

E = E₀ + Eind − Σ 𝐦held·ℬext,   𝔪 = E/E₀𝓜 d²z/dt² = −∂E/∂zthe helical window (0.1c)

so the push on each sample belongs to its own moments, not to its mass, and a uniform twist turns a held moment and pushes nothing. The record’s magnetic bounds, 0.14 aN m for the synchronous torque and 0.02 aN m for horizontal field noise, are accepted in full; the samples’ magnetizations set them. Mass, energy, pressure and temperature enter one total gravitating mass, every channel once (I-⁠7):

𝓜 = M + E/c² + PV/c² + aΘ⁴V/c² + …each channel once

Rotation, support stress and electromagnetic and thermal energy belong to the same tensor; an assayed total already includes its constituents, and gravitational binding appears in the solved mass aspect. Each background of the record enters its own channel once, and none is a measured residual gravitational force.

Five gaps and the closest gap. Five gaps: the record’s measured torques against the face-to-face gap, from the labels of its Fig. 3a, with its bars at two standard uncertainties of the statistical and other-systematic rows, or at one; at the closest gap the calculated Newtonian torque, 11.17 ± 0.26 aN m at one standard uncertainty. The record prints no numerical calculated torque at the other four gaps, so no band is drawn there. The closest gap: Table 1’s three rows at one standard uncertainty, their quadrature sum for the measured-minus-calculated difference, 0.59 aN m, and the bar of Fig. 3a, 1.07 aN m. The slider scales the statistical row with the white-noise law, as one over the square root of the integration time, from its 0.27 aN m at 100 h, and carries the change through the difference, the bar, the force’s uncertainty over the 23.694 mm arm and the quotient Ginf/Gref = 11.16/11.17, whose relative uncertainty adds the force’s and the factor’s 2.3% in quadrature. At 100 h every value is the record’s; other integration times are the scaling law’s, never measurements in the source.

I-⁠1, I-⁠3, I-⁠4, X-⁠1, Use 2

What a discriminating test needs

Five contrasts, none counted as a prediction

A divergence is claimed only where FDT supplies an additional source, a constitutive input or a changed observable dictionary, and the record’s printed amplitudes carry neither a torque polarity nor an encoder phase. Each new test is therefore a signed component measured over whole rotations, with the reference phase and torque polarity fixed beforehand and each setting’s complex voltage response calibrated independently. Both models are matched in source state, drive history and readout dynamics, and both predicted torque traces pass through the same sampling, cubic subtraction, quadratures, transfer division and phase projection as the readings. Every design is a neighbour comparison, two settings of one apparatus read as the centred contrast

Kdiff = (y − w) − (p − 𝔮ref),   𝐡 = (1, −1, −1, 1)T𝗎(Kdiff)² = 𝐡TΣjoint𝐡the centred contrast

with y and w the new signed torque readings, p and 𝔮ref their independently estimated predictions and Σjoint the covariance of the four, readings and predictions together. A rejection in either direction requires |Kdiff| > 3𝗎(Kdiff) under a validated Gaussian model with independently characterized covariance. One shared offset cancels exactly, and a shared gain need not cancel when the two predicted torques differ. In the analysis’s single-pair design model, illustrative-assumed, propagation gives

𝗎(Kdiff)² = 𝗌g²(p − 𝔮ref)² + 𝗌r²(p² + 𝔮ref²)    + 2(𝗌b² + 𝗌k² + 𝗌n²)one pair’s variance

with means (10, 12) for the matched configurations and (20, 8) for the displaced pair, a common offset of 0.65, a shared fractional gain 𝗌g = 0.03 and a reset gain 𝗌r = 0.04, and background, calibration and readout of 0.2, 0.4 and 0.3, in normalized torque units. With 250 000 pairs in a stream, the rule’s nominal two-sided tail, erfc(3/√2) = 0.0027, expects 675 rejections with a binomial standard error of 26; the nine streams of the three clusters, period, spacing and bias, reject between 631 and 715 times, all within four standard errors. These counts are a size calculation under an assumed law and supply no sampling distribution for the printed amplitudes.

  1. DC-1
    Finite geometry: the pairs a settlement joins, at a held-out gap or after a remount. At a held-out gap or after an independently metrologized remount, both ends are compared about the common axis with each configuration’s calibrated dynamics and initial state, and each predicted signed finite-window torque is subtracted before differencing against the reference configuration. The conventional tensor source already includes multipoles, and granting a residual an arbitrary coupling would deprive the comparison of its independent normalization. FDT needs the pairs the many-body settlement joins (O-⁠5, register row SA-05) to predict a signed finite-window reading apart from the finite-body kernel, which carries the multipole and tidal remainders of the spherical pair expression at layer L2, derived in the derivation notes S16 and S17 of the visible-universe analysis; no such pairing enters a number here.
  2. DC-2
    Exterior source: the mass shadow, between held-out and reference separations. The signed residuals at held-out and reference separations of the same rotor and pendulum are compared with the conventional source content and metrology fixed, the gravitating mass calibrated independently and shared errors propagated. Every nonzero source carries its mass shadow, the exterior compressed-omnium response of 0.7, which gravitates through 𝓜, the part already present in the mass calibration entering once; an independently specified exterior density profile and closure would determine the additional torque trace at each separation, and the candidate is distinct where the two models predict different held-out minus reference contrasts. FDT needs the shadow profile and its closure (O-⁠14, SA-24); no profile enters a number here. The five-gap series is the measurement this contrast reads, at milligram mass and millimetre separation.
  3. DC-3
    Causal readout: the two rotor periods, matched. The same pendulum ends return to the closest setting at the two reported rotor periods, the order randomized within each interleaved pair, new readout noise drawn at each setting and the same prior drive and loop state used in both models. I-⁠1 separates a static force from its dynamic readout and supplies no distinct transfer for this pendulum, so at matched inputs the FDT and conventional mean residual contrasts both vanish at every positive precision: the normalized predicted residual is 0, the ablated residual is 0 and their difference is 0 against a threshold of 3 (SA-02). The contrast is inert.
  4. DC-4
    Assayed source energy: a heated rotor against a reference, matched. Heated and reference rotor states are compared after equilibration, with the change of source energy assayed independently and the heat capacity, expansion, density, support stress, true gap, rotation and detector gain calibrated for each state. The same total stress-energy is carried into general relativity, every channel once (I-⁠7), so the two means agree: 0, 0 and a difference of 0 against 3 (SA-31). The contrast is inert, and the record’s ambient temperature coefficient pertains to a different intervention.
  5. DC-5
    Electrostatic transduction: a bias scan against a reference bias. A specified bias and a reference bias are applied with a common encoder zero and nominal gap, the pendulum and controller state carried through each change, and contact and patch offsets, shielding, membrane displacement, actual gap, gain and calibrated capacitance derivatives tracked at each setting. The scan measures the torque response to applied potential; an FDT point for it needs the map from a charge or circulating charge to the I-⁠1 force input, beyond ordinary electrical transduction (O-⁠6, SA-26), and no such map enters a number here.

Two of the five contrasts read the same mean in both theories at matched inputs, and three turn on maps that enter no number here; none is counted as a prediction. A discriminating measurement pairs a prospectively calibrated signed observable with an FDT input that changes its predicted value.

Neighbours and contrasts. Across the balance: a schematic pendulum, not the record’s geometry, with end masses m₊ and m₋ at levers ±L about the fibre and a field along the gap direction, g₊ at the test end and g₋ at the counterbalance; the sliders set the common part, (g₊ + g₋)/2, the difference, g₊ − g₋, and the counterbalance’s mass. The axial moment is L(m₊g₊ − m₋g₋). With a balanced pendulum, its first mass moment about the fibre zero, the common part cancels, the comparator’s zero, and the moment reads the difference across the extent, mL(g₊ − g₋); an unbalanced one admits the common part through L(m₊ − m₋). Two settings: seeded draws of the centred contrast Kdiff under the analysis’s illustrative-assumed single-pair model, its offset, gains and additive terms, at the matched point (10, 12) or the displaced point (20, 8), each divided by its own standard uncertainty from the variance formula; the tails beyond ±3 are rejections, against the nominal 0.0027. The figure’s draws are illustrative and their count moves with the seed; the analysis’s receipt, 250 000 pairs per stream, gives the nine counts the text reports.

I-⁠3a, Use 7, U-11 to U-16

Readings declined

Use 6 Data yes, interpretation not necessarily

Standard reading

Magnetic coupling between the test and the source masses, from the remanent moment and the moment induced by ambient fields, is modelled by integrating the dipole-dipole interaction energy, a torque apart from gravitation.

FDT reading

The field is the helical density gradient of the omnium, never an entity apart from it (P-⁠5), and a moment’s push in that twist is motion through a density gradient read on the sample’s own rest energy, the helical window (0.1c). The bounds of 0.14 and 0.02 aN m are accepted in full.

Within the mass window of 10⁻¹⁴ to 10⁻¹² kg, gravitational interactions between micrometre-scale objects can produce measurable entanglement within experimentally achievable coherence times.

The universe is deterministic, with no probability amplitudes: a witnessed correlation of two masses is a bivector grade the two patterns share across one density gradient, superposition the ignorance of helical phase at observation and projection its fixation (I-⁠4). The witness is accepted as a prospective observable of that grade, and the torque this balance reads is a different observable.

Gravity awaits quantisation, and measuring the gravitational interaction between milligram-scale objects is a significant step toward bridging the classical and quantum regimes of gravitation.

One omnium runs through every regime, each consensus reading good in its own, and the geometry is carried by the one quantised excitation, each energy step adjusting mass, kept time and strand length together (I-⁠4, 0.1a, I-⁠2); the wafers’ field points sit deep in the general-relativistic regime. The measurement’s reach in source mass is accepted in full.

A Yukawa-type potential of strength α and range λ constrains extra-dimension models in the α–λ parameter space.

α is the curvature ratio of the Alpha Lock and carries no fitted coefficient, coupling or reference scale (0.3); the strength and range, gY and ℓY here, belong to the comparison potential. The exclusion at 95% confidence is accepted in full as an assessment of that specified potential.

Convergence

Use 2 Same source, same data

FDT agrees with the measured record of general relativity and quantum mechanics without reservation. At layer L1, substituting the two exterior gravitational radii at their common centre distance gives the inverse-square pair magnitude exactly on disjoint spherical monopoles. At layer L2, with the same source and data, I-⁠3 returns the same curvature as general relativity, and to leading weak-field, slow-motion order its finite-body volume kernel yields the wrench, the fibre-axis projection, the rotor harmonic and, after the same readout and reduction, the same equivalent-force estimate, on the actual wafer and strut geometry. The observed fall in torque and the reported Newtonian comparison are accepted at the record’s uncertainties: at the closest gap the measured and calculated torques differ by −0.01 ± 0.59 aN m, and the in-sample quotient restates that comparison as Ginf/Gref = 0.999 ± 0.053.

At layer L3 the finite-geometry remainder, the exterior density response and the charge input enter no number here (O-⁠5, O-⁠14, O-⁠6), and in the other comparisons matched energy content and matched mechanical response return the same means. FDT makes a divergence claim only where it supplies an additional source, a constitutive input or a changed observable dictionary, never through α, a chart or a branch tag alone. Where FDT, general relativity and quantum mechanics converge, FDT reproduces their results; where it declines their assumptions, here a field as an entity apart from the omnium, probability amplitudes and gravity awaiting quantisation, it is under no obligation to reproduce them.