Probing Yukawa Gravity with Modulated Newtonian Cancellation in the Torsion bar type detector Yuki Inoue, Hsiang-Yu Huang, Vivek Kumar, Daiki Tanabe Abstract We investigate the sensitivity of a torsion-bar gravitational-wave detector to Yukawa-type deviations from Newtonian gravity using a differential gravitational calibrator (GCal), where two rotating mass systems cancel the leading Newtonian torque. We derive an exact expression for the residual torque and map the Yukawa signal into a strain-equivalent response in the sub-Hz band. We evaluate the sensitivity in the ( α Y , λ ) (\alpha_{Y},\lambda) parameter space, finding optimal performance at scales comparable to the experimental geometry, reaching | α Y | = 2.4 × 10 − 5 |\alpha_{Y}|=2.4\times 10^{-5} at λ = 8 , m \lambda=8,\mathrm{m} . The sensitivity is limited by residual Newtonian torque from imperfect cancellation rather than statistical noise, with a systematic floor reached at T eq ≃ 9.25 × 10 4 , s T_{\rm eq}\simeq 9.25\times 10^{4},\mathrm{s} ( ∼ 26 \sim 26 hours). This limit is dominated by uncertainties in the source-mass geometry. The differential configuration retains sensitivity even at large interaction ranges, enabling constraints at meter-scale distances. These results establish torsion-bar detectors as a systematics-limited probe of non-Newtonian gravity in the sub-Hz band. I Introduction Testing gravity at short distances and low frequencies provides a unique window into physics beyond the Standard Model. While Newtonian gravity has been verified over a wide range of length scales, a variety of theoretical frameworks predict deviations from the inverse-square law at sub-meter scales Fischbach and Talmadge 1999 ; Adelberger et al. 2003 . Such deviations are often parameterized in terms of a Yukawa-type correction to the gravitational potential Fischbach and Talmadge 1999 , characterized by a strength (\alpha_{Y}) and a range (\lambda) . Experimental searches for Yukawa interactions have been actively pursued using torsion balances, atom interferometry, and precision force measurements Adelberger et al. 2009 ; Kapner et al. 2007 ; Geraci et al. 2008 ; Raffai et al. 2011 , placing strong constraints over a broad range of (\alpha_{Y},\lambda) parameter space. More recently, constraints have also been derived from planetary motion, satellite tracking, and astrophysical observations Konopliv et al. 2011 ; Hees et al. 2017 ; Lee et al. 2017 . In parallel, gravitational-wave detector technologies have advanced rapidly, opening new opportunities to probe weak forces with exquisite sensitivity. The direct detection of gravitational waves by LIGO Abbott et al. 2016 and the subsequent observations of compact binary mergers Abbott et al. 2019 ; Abbott et al. 2021 ; Abbott et al. 2023 have established interferometric detectors as precision probes of fundamental physics. In particular, torsion-bar antennas, such as TOBA and TorPeDO operating in the sub-Hz band offer a promising platform for detecting low-frequency gravitational signals Ando et al. 2010 ; Ju et al. 2019 . The CHRONOS detector, a cryogenic sub-Hz torsion-bar interferometer with a quantum non-demolition speed-meter configuration Inoue et al. 2026a ; Inoue et al. 2025 ; Inoue et al. 2026b , is also designed to achieve high sensitivity in the (0.1\text{--}10,\mathrm{Hz}) band. In this regime, the response of the torsion-bar system is dominated by inertia, providing a clean and well-understood transfer function between external torques and the measured strain-equivalent signal Saulson 1994 . A key technique for absolute calibration in such systems is the gravitational calibrator (GCal), in which rotating masses generate a well-defined, time-dependent gravitational torque. In existing gravitational-wave detectors such as LIGO, Virgo, and KAGRA, calibration is primarily performed using photon calibrators (PCal), which apply a known radiation-pressure force to the test masses Karki et al. 2016 ; Goetz et al. 2010 ; Inoue et al. 2023 . In parallel, Newtonian calibrators (NCal) generate a dynamic gravitational field using rotating masses and have been implemented as an independent calibration method Estévez et al. 2018 ; Acernese et al. 2018 . Early demonstrations of dynamic gravitational field generation and detection date back several decades Forward and Miller 1967 ; Hirakawa et al. 1980 ; Ogawa et al. 1982 ; Kuroda and Hirakawa 1985 ; Astone et al. 1991 ; Astone et al. 1998 , and have recently been developed for modern interferometric detectors as Newtonian calibrators Estévez et al. 2018 . Unlike conventional displacement-based calibration methods such as photon calibrators, the torque-coupled configuration directly excites the rotational degree of freedom of the torsion bar, enabling efficient calibration in the sub-Hz frequency band. Moreover, by employing a differential configuration with two rotating mass systems, it is possible to suppress the leading Newtonian contribution through geometric cancellation Inoue et al. 2018 ; Inoue et al. 2026c . This cancellation scheme provides a powerful method for probing non-Newtonian gravity. Because a Yukawa interaction does not follow a pure power-law dependence on distance, the cancellation condition for the Newtonian term does not generally eliminate the Yukawa contribution. As a result, a residual signal remains that is directly sensitive to the parameters (\alpha_{Y},\lambda) . This approach allows one to transform a calibration system into a probe of short-range gravitational physics. In this paper, we develop a framework for evaluating the sensitivity of a torsion-bar detector to Yukawa-type deviations from Newtonian gravity using a differential GCal configuration. We derive the exact expression for the residual torque after imposing the Newtonian cancellation condition, without relying on low-order approximations. The resulting torque is then converted into a strain-equivalent signal using the mechanical transfer function of the torsion-bar system. Based on this formulation, we compute the signal-to-noise ratio and project constraints on the Yukawa parameters (\alpha_{Y},\lambda) . An important aspect of this method is the role of systematic uncertainties. In particular, imperfect cancellation of the Newtonian contribution introduces a residual background torque that can limit the achievable sensitivity. This issue is closely related to calibration accuracy requirements in gravitational-wave detectors Hall et al. 2018 ; Cahillane et al. 2017 . We show that this residual sets a fundamental floor on the constraint on (\alpha_{Y}) , independent of the observation time, and must be carefully controlled in the experimental design. This paper is organized as follows. In Sec. II, we derive the exact expression for the Yukawa-induced torque in the differential GCal configuration. In Sec. III, we convert the torque into a strain-equivalent signal using the mechanical response of the torsion bar. In Sec. IV, we evaluate the sensitivity and present projected constraints in the (\alpha_{Y},\lambda) parameter space, including the impact of systematic uncertainties. II Torque Expansion with Yukawa Correction Before presenting the detailed derivation, we first summarize the core concept of the experimental configuration. Figure 1: Schematic diagram of the differential GCal configuration. Two source masses are placed under the torsion bar with a controlled geometric scaling. The configuration is designed such that the Newtonian torque cancels, while a residual Yukawa contribution remains. Large GCal can be tuned by lift to optimize the position. As shown in Fig. 1 , the experiment employs a differential configuration in which two gravitational field calibrators (GCals) are arranged symmetrically with respect to the torsion bar. By appropriately choosing the geometric scaling between the two GCals, the Newtonian contributions from each side cancel to a high degree. Such dynamic gravitational field configurations have been studied since early experimental works Forward and Miller 1967 ; Hirakawa et al. 1980 and form the basis of modern Newtonian calibrators Estévez et al. 2018 ; Acernese et al. 2018 as well as the torque-coupled GCal configuration developed in Ref. Inoue et al. 2018 ; Inoue et al. 2026c . This cancellation relies on the fact that the Newtonian interaction follows a pure inverse-distance law, allowing the torque contributions to be balanced by tuning the geometry. In contrast, a Yukawa-type interaction introduces an additional exponential factor, which depends on the absolute distance scale. As a result, the cancellation condition for the Newtonian term does not apply to the Yukawa contribution. Consequently, after the cancellation of the dominant Newtonian torque, a residual signal remains that is sensitive to the Yukawa interaction. The experiment therefore converts a large common-mode Newtonian signal into a small differential signal that directly probes deviations from Newtonian gravity. With this conceptual picture in mind, we now proceed to derive the Yukawa-corrected torque. We start from the modified gravitational potential Raffai et al. 2011 V(r)=-\frac{GmM}{r}\left(1+\alpha_{Y}e^{-r/\lambda}\right), (1) where G is the gravitational constant, m and M are the source and test masses, \alpha_{Y} is the Yukawa coupling strength, and \lambda is the interaction range. The potential can be separated into the Newtonian and Yukawa parts as V(r)=V_{\mathrm{N}}(r)+V_{\mathrm{Y}}(r), (2) with \displaystyle V_{\mathrm{N}}(r) \displaystyle=-\frac{GmM}{r},\ V_{\mathrm{Y}}(r) \displaystyle=-\alpha_{Y}\frac{GmM}{r}e^{-r/\lambda}. (3) Since the Newtonian contribution has already been derived in previous work Inoue et al. 2026c , we focus on the Yukawa contribution in the following. II.1 Geometrical parameterization The geometrical configuration used in the torque calculation is shown in Fig. 2 . Figure 2: Geometrical configuration used for the torque calculation. The parameters a_{i} , b_{i} , \rho_{i} and R_{i} define the relative positions of the rotor masses and the test mass, where i={S,L} . The relative angle between the rotor and the torsion bar is denoted by \Delta=\theta-\phi . We introduce a reference distance scale R^{2}=a^{2}+b^{2}+h^{2}+\rho_{\perp}^{2}, (4) which represents the typical separation between the source masses and the test mass. The actual distance can then be written as a modulation around this reference scale, r^{2}=R^{2}+\delta, (5) where \delta encodes the angular dependence of the geometry. The modulation \delta arises from the relative orientation between the rotor and the torsion bar. Defining the rotor phase \theta and the bar orientation \phi , the relative angle is \Delta=\theta-\phi. (6) The geometric coupling between the two directions appears through \cos\Delta . It is convenient to introduce a dimensionless parameter u\equiv-\frac{\delta}{R^{2}}, (7) so that the distance takes the form r=R\sqrt{1-u}. (8) The parameter u can be decomposed into contributions associated with different geometric effects. In particular, the dominant term arises from the relative-angle modulation, u\simeq\frac{2ab}{R^{2}}\cos\Delta. (9) This expression shows that the distance modulation is primarily governed by the relative phase \Delta between the rotor and the torsion bar, as shown in Fig. 2 . Other contributions, such as small offsets of the source masses, provide subleading corrections and are neglected in the following. Using this parametrization, the Yukawa part of the potential can be written as V_{\mathrm{Y}}=-\alpha_{Y}\frac{GmM}{R}\frac{e^{-\beta^{(Y)}\sqrt{1-u}}}{\sqrt{1-u}},\qquad\beta^{(Y)}\equiv\frac{R}{\lambda}. (10) The problem is therefore reduced to expanding the kernel K(u;\beta^{(Y)})\equiv\frac{e^{-\beta^{(Y)}\sqrt{1-u}}}{\sqrt{1-u}} (11) in powers of u . II.2 Expansion of the Yukawa kernel We first rewrite the kernel as K(u;\beta^{(Y)})=e^{-\beta^{(Y)}}\frac{\exp\!\left[\beta^{(Y)}\left(1-\sqrt{1-u}\right)\right]}{\sqrt{1-u}}. (12) This form is convenient because the factor e^{-\beta^{(Y)}} can be taken out, while the remaining expression is analytic around u=0 . We now expand K(u;\beta^{(Y)}) in a Taylor series, K(u;\beta^{(Y)})=e^{-\beta^{(Y)}}\sum_{k=0}^{\infty}c_{k}(\beta^{(Y)})\,u^{k}. (13) The coefficients c_{k}(\beta^{(Y)}) are known to be expressible in terms of the Bessel polynomials y_{k}(\beta^{(Y)}) : c_{k}(\beta^{(Y)})=\frac{y_{k}(\beta^{(Y)})}{2^{k}k!}. (14) Hence the Yukawa potential becomes V_{\mathrm{Y}}=-\alpha_{Y}\frac{GmM}{R}e^{-\beta^{(Y)}}\sum_{k=0}^{\infty}\frac{y_{k}(\beta^{(Y)})}{2^{k}k!}u^{k}. (15) II.3 Torque from the Yukawa potential for 2f components As in the Newtonian case, the variable u encodes the angular dependence of the rotor–test-mass configuration. We define the phase difference as \Delta=\phi+\theta, (16) where \phi is the rotation angle of the rotor and \theta is a constant geometrical offset. Since \theta is constant, one has \frac{\partial}{\partial\phi}=\frac{\partial}{\partial\Delta}. (17) In the present convention, we set \theta=0 , so that \Delta=\phi . For the symmetric quadrupolar geometry considered here, the angular dependence is such that only even powers of the basic geometrical quantity contribute to the leading torque term. After summing over the symmetric mass configuration and differentiating with respect to \Delta , the leading harmonic structure is proportional to \sin(2\Delta) . More specifically, the geometrical part of the derivation proceeds exactly as in the Newtonian case, and one finds that only the terms u^{2n} survive in the final quadrupole torque series. The Yukawa correction therefore preserves the same angular structure as the Newtonian torque, while modifying the coefficient of each multipole order. The torque is given by \tau_{\phi}^{(\mathrm{Y})}=-\frac{\partial V_{\mathrm{Y}}}{\partial\phi}=-\frac{\partial V_{\mathrm{Y}}}{\partial\Delta}. (18) Substituting Eq. ( 15 ), we obtain \tau_{\phi}^{(\mathrm{Y})}=\alpha_{Y}\frac{GmM}{R}e^{-\beta^{(Y)}}\sum_{k=0}^{\infty}\frac{y_{k}(\beta^{(Y)})}{2^{k}k!}\frac{\partial}{\partial\Delta}u^{k}. (19) At this stage, the same geometrical manipulations as in the Newtonian derivation can be used. The result is that the torque can be written in the form \tau_{\phi}^{(\mathrm{Y})}=\frac{2GmM}{R}\alpha_{Y}e^{-\beta^{(Y)}}\sum_{n=1}^{\infty}B_{n}(\beta^{(Y)})\left(\frac{2ab}{R^{2}}\right)^{2n}\sin(2\Delta), (20) where the Yukawa coefficient B_{n}(\beta^{(Y)}) is given by B_{n}(\beta^{(Y)})=\frac{y_{2n}(\beta^{(Y)})}{2^{2n}(2n)!}\frac{\binom{2n-1}{n-1}}{2^{2n-1}}. (21) Using this formulation, the derivation of the Newtonian contribution is summarized in Appendix A . The general Newtonian coefficient A_{n} has been derived in previous work Inoue et al. 2026c and is given in Eq. ( 55 ). Importantly, we show that this coefficient can be consistently recovered as the \lambda\to\infty limit of the Yukawa expression, i.e., \lim_{\beta^{(Y)}\to 0}B_{n}(\beta^{(Y)})=A_{n}, (22) demonstrating that Eq. ( 20 ) provides a unified generalization of the Newtonian expansion. Including both the Newtonian and Yukawa contributions, the total torque is written as \small\tau=\frac{2GmM}{R}\sum_{n=1}^{\infty}\left[A_{n}+\alpha_{Y}e^{-\beta^{(Y)}}B_{n}(\beta^{(Y)})\right]\left(\frac{2ab}{R^{2}}\right)^{2n}\sin(2\Delta). (23) Thus, the Yukawa correction introduces a scale-dependent modification through the factor e^{-R/\lambda} and the polynomial dependence encoded in the Bessel polynomials y_{2n}(R/\lambda) , while preserving the same quadrupolar angular dependence \sin(2\Delta) as in the Newtonian case. III Newtonian cancellation We consider a differential GCal configuration in which two GCals are placed on opposite sides of the torsion bar along the vertical direction and driven with opposite rotation senses. With this configuration, the leading Newtonian contribution can be suppressed, while a residual Yukawa contribution remains due to its non-power-law distance dependence. This provides a direct method to probe short-range deviations from Newtonian gravity. Such differential configurations have been proposed in the context of gravity-field-based calibration and tests of non-Newtonian gravity Inoue et al. 2018 ; Raffai et al. 2011 . We denote the GCal located farther from the torsion bar by the subscript L (‘long”) and the one located closer to the torsion bar by S (‘short”). The torque generated by each GCal is given by \displaystyle\tau_{L} \displaystyle=\frac{2Gm_{L}M}{R_{L}}\sum_{n}\Bigl[A_{n}+\alpha_{Y}e^{-\beta^{(Y)}*{L}}B_{n}\Bigr]\left(\frac{2ab*{L}}{R_{L}^{2}}\right)^{2n}\sin(2\omega t), (24) \displaystyle\tau_{S} \displaystyle=\frac{2Gm_{S}M}{R_{S}}\sum_{n}\Bigl[A_{n}+\alpha_{Y}e^{-\beta^{(Y)}*{S}}B_{n}\Bigr]\left(\frac{2ab*{S}}{R_{S}^{2}}\right)^{2n}\sin(-2\omega t), (25) where \beta^{(Y)}_{L}\equiv R_{L}/\lambda and \beta^{(Y)}_{S}\equiv R_{S}/\lambda . Using \sin(-2\omega t)=-\sin(2\omega t) , the total torque becomes \displaystyle\tau_{L}+\tau_{S} \displaystyle=2GM\sum_{n=1}^{\infty}\Bigl[A_{n},\mathcal{N}_{n}+\alpha_{Y},\mathcal{Y}_{n}\Bigr]\sin(2\omega t), (26) where \displaystyle\mathcal{N}_{n} \displaystyle\equiv\frac{m_{L}}{R_{L}}\left(\frac{2ab_{L}}{R_{L}^{2}}\right)^{2n}-\frac{m_{S}}{R_{S}}\left(\frac{2ab_{S}}{R_{S}^{2}}\right)^{2n}, (27) \displaystyle\mathcal{Y}_{n} \displaystyle\equiv e^{-\beta^{(Y)}_{L}}B_{n}(\beta^{(Y)}_{L})\frac{m_{L}}{R_{L}}\left(\frac{2ab_{L}}{R_{L}^{2}}\right)^{2n} \displaystyle\quad-e^{-\beta^{(Y)}_{S}}B_{n}(\beta^{(Y)}_{S})\frac{m_{S}}{R_{S}}\left(\frac{2ab_{S}}{R_{S}^{2}}\right)^{2n}. (28) The crucial point is that the Newtonian cancellation condition must be imposed on the fully resummed Newtonian torque, rather than order by order in the multipole expansion. This distinction is essential because the series does not truncate, and cancellation at a finite order does not guarantee suppression of the total Newtonian contribution. Using the generating function \sum_{n=0}^{\infty}A_{n}x^{n}=(1-x)^{-1/2},\qquad A_{n}=\frac{(2n)!}{(n!)^{2}4^{n}}, (29) the Newtonian part of the torque generated by a single GCal can be summed exactly as \tau_{N}(R,b)=\frac{2GmM}{R}\left[\frac{1}{\sqrt{1-\dfrac{4a^{2}b^{2}}{R^{4}}}}-1\right]\sin(2\omega t). (30) Therefore, for the differential configuration, the total Newtonian torque is \displaystyle\tau_{N,L}+\tau_{N,S}=2GM\Bigg[ \displaystyle\frac{m_{L}}{R_{L}}\left(\frac{1}{\sqrt{1-\dfrac{4a^{2}b_{L}^{2}}{R_{L}^{4}}}}-1\right) \displaystyle-\frac{m_{S}}{R_{S}}\left(\frac{1}{\sqrt{1-\dfrac{4a^{2}b_{S}^{2}}{R_{S}^{4}}}}-1\right)\Bigg]\sin(2\omega t). (31) Hence, the exact Newtonian null condition is given by \frac{m_{L}}{R_{L}}\left[\frac{1}{\sqrt{1-\dfrac{4a^{2}b_{L}^{2}}{R_{L}^{4}}}}-1\right]=\frac{m_{S}}{R_{S}}\left[\frac{1}{\sqrt{1-\dfrac{4a^{2}b_{S}^{2}}{R_{S}^{4}}}}-1\right]. (32) Since R_{L}^{2}=a^{2}+b_{L}^{2}+h_{L}^{2}+\rho_{L}^{2},\qquad R_{S}^{2}=a^{2}+b_{S}^{2}+h_{S}^{2}+\rho_{S}^{2}, (33) this condition determines the required long-side distance h_{L} at which the fully resummed Newtonian contribution vanishes. Thus, the Newtonian cancellation is achieved not by truncating the multipole expansion at a finite order, but by tuning the geometry such that the exact Newtonian torque becomes zero. In the practically relevant case m\equiv m_{L}=m_{S},\qquad 2b\equiv b_{L}=2b_{S}, (34) the null condition reduces to a single equation relating h_{L} and h_{S} . Although this condition can be written analytically, it does not in general yield a simple closed-form expression for h_{L} . Therefore, for a fixed short-side geometry, we determine the long-side distance h_{L} numerically by solving Eq. ( 32 ). It is then convenient to parametrize the null condition in the form \gamma h\equiv h_{L}=\gamma h_{S}, (35) where \gamma is the numerically determined ratio that realizes the exact Newtonian cancellation for the chosen geometry. Thus, the Newtonian-null configuration is fully specified by the root of Eq. ( 32 ), or equivalently by the corresponding value of \gamma . Having fixed the geometry such that the exact Newtonian contribution vanishes, we now evaluate the remaining Yukawa component without truncating the multipole expansion. Once this condition is imposed, the Newtonian contribution vanishes identically, and the residual signal is purely Yukawa: {\footnotesize\tau_{Y}(\lambda)=2GM,\alpha_{Y}\sum_{n=1}^{\infty}\mathcal{Y}_{n}(\lambda)\sin(2\omega t)} (36) For compactness, we define {R}_{S}\equiv R_{S}(a,b,\rho,h,\gamma),;{R}_{L}\equiv R_{L}(a,b,\rho,h,\gamma) , and express all quantities in terms of these functions. Then, \displaystyle\mathcal{Y}_{n}(\lambda) \displaystyle=\frac{m}{\mathcal{R}_{S}}\left(\frac{b}{\mathcal{R}_{S}}\right)^{2n}\Bigl[2^{2n}\left(\frac{\mathcal{R}_{S}}{\mathcal{R}_{L}}\right)^{2n+1}e^{-\mathcal{R}_{L}/\lambda}B_{n}\left(\frac{\mathcal{R}_{L}}{\lambda}\right) (37) \displaystyle-e^{-\mathcal{R}_{S}/\lambda}B_{n}\left(\frac{\mathcal{R}_{S}}{\lambda}\right)\Bigr] Therefore, the exact residual Yukawa torque at the Newtonian null is \displaystyle\tau_{Y}(\lambda) \displaystyle=2GM\alpha_{Y}\frac{m}{\mathcal{R}_{S}}\sum_{n=1}^{\infty}\left(\frac{2ab}{\mathcal{R}_{S}^{2}}\right)^{2n} (38) \displaystyle\times\Bigl[2^{2n}\left(\frac{\mathcal{R}_{S}}{\mathcal{R}_{L}}\right)^{2n+1}e^{-\mathcal{R}_{L}/\lambda}B_{n}\left(\frac{\mathcal{R}_{L}}{\lambda}\right) \displaystyle-e^{-\mathcal{R}_{S}/\lambda}B_{n}\left(\frac{\mathcal{R}_{S}}{\lambda}\right)\Bigr]\sin(2\omega t) Here, \mathcal{R}_{S} and \mathcal{R}_{L} denote the exact geometric quantities after imposing the parameter relations, so that the residual torque is written explicitly as a function of the Yukawa range \lambda . The conversion from the torque signal to the strain-equivalent signal h(t) is summarized in Appendix B . We next define the strain-equivalent signal generated by a possible Yukawa-type deviation from Newtonian gravity. Let the Yukawa torque amplitude at angular frequency \Omega=2\omega be denoted by \tau_{Y}(\Omega;\alpha_{Y},\lambda) , where \alpha_{Y} is the Yukawa strength and \lambda is the Yukawa interaction range. The induced angular response of the torsion bar is given by \theta_{Y}(\Omega;\alpha_{Y},\lambda)=\chi(\Omega)\,\tau_{Y}(\Omega;\alpha_{Y},\lambda), (39) where \chi(\Omega) is the mechanical susceptibility of the torsion-bar degree of freedom. Using the same strain-equivalent definition from Appendix B , the Yukawa-induced strain signal is defined as h_{\rm Yukawa}(\Omega;\alpha_{Y},\lambda)=\left|\frac{2\,\chi(\Omega)\,\tau_{Y}(\Omega;\alpha_{Y},\lambda)}{\eta_{g}\,|F_{\rm eff}|}\right|. (40) This quantity represents the apparent detector strain corresponding to a Yukawa torque signal at frequency \Omega . In the high-frequency regime relevant for CHRONOS, \Omega\gg\omega_{0} , the mechanical susceptibility reduces to \chi(\Omega)\simeq-\frac{1}{I\Omega^{2}}, (41) where the typical torsional resonance frequency \omega_{0} is smaller than 0.01\,\mathrm{Hz} , which is sufficiently below the operational frequency range considered in this study. Substituting this into Eq. ( 40 ), we obtain h_{\rm Yukawa}(\Omega;\alpha_{Y},\lambda)\simeq\frac{2\,|\tau_{Y}(\Omega;\alpha_{Y},\lambda)|}{\eta_{g}\,|F_{\rm eff}|\,I\,\Omega^{2}}. (42) Therefore, once the exact Yukawa torque \tau_{Y}(\Omega;\alpha_{Y},\lambda) is evaluated, the corresponding strain-equivalent signal follows directly through the same transfer function as for the standard GCal signal. This formulation is particularly convenient for the sensitivity analysis in the next section, where constraints on (\alpha_{Y},\lambda) can be obtained by directly comparing h_{\rm Yukawa} with the detector strain sensitivity. IV Systematic error limit and sensitivity Table 1 summarizes the fiducial parameter set used in this analysis, together with their associated absolute uncertainties. These parameters define both the gravitational field generated by the GCal and the response of the torsion-bar detector, and therefore determine the level of residual Newtonian contamination after cancellation. Parameters Values Absolute uncertainty Fundamental constant G 6.67\times 10^{-11} 1.50\times 10^{-15} GCal (source masses) m_{S} 1.8842 9.04\times 10^{-5} m_{L} 1.8842 9.04\times 10^{-5} h_{S} 8.000 1.00\times 10^{-3} h_{L}=\gamma h_{S} 10.553 1.00\times 10^{-3} b_{S} 0.250 2.50\times 10^{-5} b_{L} 0.500 2.50\times 10^{-5} \rho_{S} 0.0010 1.00\times 10^{-3} \rho_{L} 0.0010 1.00\times 10^{-3} Torsion-bar detector M 85.50 1.41\times 10^{-4} a 0.302 2.50\times 10^{-5} Table 1: The parameter set used in the sensitivity analysis. All uncertainties are given as absolute values. The long-baseline distance h_{L} is determined by the Newtonian cancellation condition h_{L}=\gamma h_{S} . In this section, we quantify how the parameter uncertainties listed in Table 1 propagate into the residual Newtonian signal after differential cancellation, and determine the corresponding systematic error floor. Rather than repeating the definition of the strain-equivalent signal h_{N} , we focus on its uncertainty \sigma_{h_{N}} and its decomposition. To identify the dominant sources of uncertainty, we decompose the total variance into contributions from individual parameters. The fractional contribution of each parameter is defined as C_{i}=\frac{\sigma_{h_{N},i}^{2}}{\sum_{j}\sigma_{h_{N},j}^{2}}\times 100\penalty\ \%, (43) where \sigma_{h_{N},i} is the standard deviation induced by varying only the parameter p_{i} while keeping others fixed. This decomposition is evaluated using a Monte Carlo propagation, which captures the full nonlinear dependence of the Newtonian cancellation condition. Parameter \sigma_{h_{N}} ,[-] Contribution (%) h_{S} 4.63\times 10^{-18} 58.3 h_{L} 3.51\times 10^{-18} 33.6 b_{S} 1.47\times 10^{-18} 5.9 b_{L} 7.38\times 10^{-19} 1.5 m_{L} 3.58\times 10^{-19} 0.3 m_{S} 3.57\times 10^{-19} 0.3 a 1.86\times 10^{-21} 0.0 \rho_{S} 7.13\times 10^{-22} 0.0 \rho_{L} 4.10\times 10^{-22} 0.0 G 1.34\times 10^{-31} 0.0 M 2.15\times 10^{-42} 0.0 Table 2: Noise budget of the residual Newtonian contribution expressed in strain. The uncertainty \sigma_{h_{N}} is dimensionless. As shown in Table 2 , the residual uncertainty is overwhelmingly dominated by the longitudinal geometric parameters h_{S} and h_{L} , which together account for more than 90\% of the total variance. The source-size parameters b_{S} and b_{L} contribute only at the few-percent level, while all other parameters are negligible. This hierarchy directly reflects the structure of the Newtonian cancellation condition, h_{L}=\gamma h_{S}, (44) which suppresses the leading-order Newtonian contribution. Because the cancellation is achieved by balancing the longitudinal geometry, small perturbations in h_{S} or h_{L} shift the cancellation point and generate a first-order residual in h_{N} . In contrast, variations in b_{S} and b_{L} enter only through higher-order geometric dependence and therefore have a subdominant impact on the systematic floor. The residual systematic uncertainty defines the integration time beyond which further averaging no longer improves the measurement. We therefore introduce the equivalent integration time T_{\rm eq} as the observation time at which the statistical strain sensitivity reaches the residual Newtonian systematic floor. IV.1 Dependence on source geometry To study the geometric dependence of the systematic floor, we vary the source radius b_{L} and, for each configuration, determine the corresponding cancellation parameter \gamma , the residual Newtonian uncertainty \sigma_{h_{N}} , and the equivalent integration time T_{\rm eq} . Table 3: Dependence of the residual Newtonian uncertainty on the source radius b_{L} . The equivalent integration time T_{\rm eq} is evaluated for each configuration. b_{L} (m) \gamma \sigma_{h_{N}} T_{\rm eq} (sec) 0.10 1.3199 4.06\times 10^{-19} 2.05\times 10^{7} 0.20 1.3198 1.14\times 10^{-18} 2.58\times 10^{6} 0.30 1.3196 2.32\times 10^{-18} 6.27\times 10^{5} 0.40 1.3194 3.95\times 10^{-18} 2.16\times 10^{5} 0.50 1.3191 6.04\times 10^{-18} 9.25\times 10^{4} Table 3 shows that the residual Newtonian uncertainty increases monotonically with b_{L} , while the equivalent integration time decreases accordingly. Thus, a larger source radius raises the systematic floor and reduces the amount of statistically useful integration time. At the same time, the scaling parameter \gamma remains nearly constant ( \gamma\simeq 1.319 ) over the explored range, indicating that the cancellation condition is controlled primarily by the longitudinal geometry and only weakly modified by the transverse source size. Here, \gamma is obtained by solving the exact Newtonian cancellation condition given in Eq. ( 32 ). This behavior implies a nontrivial design trade-off. A larger b_{L} generally enhances the Yukawa signal amplitude, but it also increases the residual Newtonian background. Conversely, a smaller b_{L} lowers the systematic floor and allows a longer effective integration time before the measurement becomes systematics limited. The final sensitivity must therefore be evaluated by consistently taking both effects into account. In this analysis, we assume tungsten as the source mass material. The detailed mechanical and gravitational response of tungsten is discussed in Inoue et al Inoue et al. 2026a ; Inoue et al. 2025 ; Inoue et al. 2026b . IV.2 Sensitivity curve The projected constraint is obtained from \alpha_{Y{\rm lim}}(\lambda)=\frac{2\sqrt{S_{h}(\Omega)}}{h_{\rm Yukawa}(\Omega;1,\lambda)\sqrt{T_{\rm eq}}}, (45) where T_{\rm eq} is the equivalent integration time defined by the condition that the accumulated statistical sensitivity reaches the residual Newtonian systematic floor. Thus, the final sensitivity is evaluated not for an arbitrarily chosen observation time, but for the maximum effective integration time allowed by the Newtonian systematic uncertainty for each configuration. Figure 3: Projected sensitivity to the Yukawa coupling strength \alpha_{Y} as a function of the interaction range \lambda . Different curves correspond to different values of b_{L} . Existing experimental and astrophysical constraints (Lab, geophysical, LAGEOS, LLR, planetary, and Galactic Center) are adapted from Refs. Lee et al. 2017 ; Adelberger et al. 2009 ; Konopliv et al. 2011 ; Hees et al. 2017 ; Borka et al. 2013 ; Psaltis et al. 2016 ; Boehle et al. 2016 ; Gillessen et al. 2017 . The sensitivity is evaluated at the equivalent integration time T_{\rm eq} for each configuration and is therefore limited by systematic uncertainties. Here we assume the CHRONOS sensitivity at 0.5 Hz, S_{h}^{1/2}\simeq 10^{-17}\,{\rm Hz}^{-1/2} , as presented in Inoue et al. 2026a ; Inoue et al. 2025 ; Inoue et al. 2026b . Because the statistical sensitivity is evaluated at T_{\rm eq} , the projected limit in Fig. 3 already includes the effect of the residual Newtonian systematic floor. The convergence properties of the Yukawa expansion with respect to the truncation order n are discussed in Appendix C . In the present analysis, we truncate the series at n=4 . Figure 3 shows the resulting sensitivity curves together with existing experimental and astrophysical constraints on Yukawa-type deviations from Newtonian gravity Adelberger et al. 2009 ; Fischbach and Talmadge 1992 ; Konopliv et al. 2011 ; Hees et al. 2017 ; Borka et al. 2013 ; Psaltis et al. 2016 ; Boehle et al. 2016 ; Gillessen et al. 2017 ; Lee et al. 2017 . We find that the experiment exhibits strong sensitivity at large interaction range \lambda . This behavior arises because the Yukawa potential asymptotically approaches a Newtonian-like interaction in the limit \lambda\to\infty . Even in this regime, however, a measurable residual remains due to the difference between the L and S source configurations, allowing the differential setup to retain sensitivity to long-range deviations. This feature becomes visible only when higher-order contributions are properly included. Previous studies, such as Raffai et al. Raffai et al. 2011 , mainly emphasized leading-order terms, whereas the inclusion of higher-order geometric contributions reveals an additional sensitivity in the large- \lambda regime. A detailed discussion of this asymptotic limit is provided in Appendix D . Importantly, even for very large \lambda , meaningful constraints can be derived from the residual signal, which constitutes a unique aspect of the present work. Combining Table 3 with Fig. 3 , we find that the optimal configuration is determined by the competition between signal enhancement and systematic growth. Increasing b_{L} strengthens the Yukawa response, but it also shortens the equivalent integration time by raising the residual Newtonian floor. The final projected constraint is therefore obtained only after both effects are combined through Eq. ( 45 ). In particular, the best sensitivity is achieved at |\alpha_{Y}|=2.4\times 10^{-5}\quad\text{at}\quad\lambda=8\,\mathrm{m}, (46) evaluated at the equivalent integration time T_{\rm eq}\simeq 9.25\times 10^{4}\,\mathrm{s}\;\;(\simeq 25.7\,\mathrm{hours}). (47) The optimal interaction scale \lambda\sim 8\,\mathrm{m} originates from the characteristic geometry of the setup, in particular the source height h_{S} , which sets the effective distance scale of the interaction. The input parameter uncertainties used in the error budget are adopted from previous studies Inoue et al. 2018 ; Inoue et al. 2026c . We also note that a characteristic feature appears around \lambda\sim 2.5\,\mathrm{m} , where the residual contributions from the large and small GCals change sign. This sign reversal leads to a partial cancellation of the Yukawa signal and produces a corresponding structure in the sensitivity curve. In this sense, the sensitivity curves presented here summarize the maximum achievable constraint for each geometric configuration once the available statistical integration has been exhausted up to the systematic-error limit. V Discussion In this work, we have proposed a method to probe Yukawa-type deviations from Newtonian gravity using a differential gravitational calibrator (GCal) configuration in a torsion-bar detector. A key feature of this approach is that the leading Newtonian contribution can be suppressed through geometric cancellation, while a residual signal remains for non-Newtonian interactions. In this sense, the present method transforms a calibration system into a direct probe of new physics Raffai et al. 2011 ; Inoue et al. 2018 ; Matone et al. 2007 ; Inoue et al. 2026c . Compared to conventional experiments searching for short-range modifications of gravity, such as torsion-balance measurements and atom interferometry, our approach operates in a distinct regime Raffai et al. 2011 ; Forward and Miller 1967 ; Hirakawa et al. 1980 ; Ogawa et al. 1982 ; Kuroda and Hirakawa 1985 ; Astone et al. 1991 ; Astone et al. 1998 . The measurement is performed dynamically at a well-defined frequency, rather than as a static force measurement, and directly probes the torque acting on the rotational degree of freedom. This frequency-domain approach provides complementary sensitivity and offers a new avenue for exploring deviations from the inverse-square law. The sensitivity to the Yukawa range (\lambda) is determined by the geometric configuration of the system. In particular, the signal is maximized when \lambda is comparable to the characteristic distances of the apparatus. In the present configuration, the optimal sensitivity is achieved around \lambda\sim 8\,\mathrm{m} , which is set primarily by the source height h_{S} that defines the effective interaction scale. For \lambda\ll\mathcal{R}_{S} , the Yukawa interaction is exponentially suppressed, while for \lambda\gg\mathcal{R}_{L} , the interaction approaches the Newtonian limit. Nevertheless, even in the large- \lambda regime, a finite residual remains due to the asymmetry between the L and S configurations, allowing the differential setup to retain sensitivity. A central result of this work is that the achievable sensitivity is not limited by statistical noise, but by systematic uncertainty arising from imperfect cancellation of the Newtonian torque, consistent with general calibration requirements in gravitational-wave detectors Hall et al. 2018 . We find that the statistical uncertainty reaches the systematic floor within an equivalent integration time of T_{\rm eq}\simeq 9.25\times 10^{4}\,\mathrm{s}\;\;(\simeq 25.7\,\mathrm{hours}), (48) beyond which further integration does not improve the constraint on \alpha_{Y} . This establishes that the experiment operates in a systematics-dominated regime on timescales of order one day. The error budget analysis shows that the dominant contribution arises from uncertainties in the geometric parameters of the source masses, in particular b_{S} and b_{L} , while all other sources are negligible. The input parameter uncertainties adopted in this analysis are based on previous calibration studies Inoue et al. 2018 ; Inoue et al. 2026c . This identifies precision control of the mass distribution and geometry as the primary requirement for improving sensitivity. We also find that the sensitivity curve exhibits a nontrivial structure around \lambda\sim 2.5\,\mathrm{m} . This feature originates from a sign change in the residual contributions from the large and small GCals, leading to a partial cancellation of the Yukawa signal. Such behavior highlights the importance of accurately modeling the full geometry of the system and cannot be captured by simplified leading-order approximations. An important implication of the frequency-domain operation is that the characteristic size of the calibration system is not limited by static constraints. At an operating frequency of \sim 0.5\,\mathrm{Hz} , the required rotational speed is sufficiently low that larger rotor radii, b_{L}\sim 0.1 – 0.5\,\mathrm{m} , can be implemented without introducing significant dynamical limitations, as discussed in torsion-bar detector studies Ando et al. 2010 . This has two important consequences. First, increasing the geometric scale enhances the Yukawa signal, since the torque scales with higher powers of b . Second, it shifts the optimal sensitivity toward larger values of \lambda , thereby extending the accessible parameter space. However, as demonstrated in this work, such scaling also increases the residual Newtonian background and shortens the effective integration time. Therefore, the optimal design must be determined by balancing signal enhancement against systematic growth. Another advantage of the present formulation is that it does not rely on a truncated multipole expansion. The exact expression for the Yukawa-induced torque includes contributions from all orders, ensuring the validity of the result even in regimes where low-order approximations break down. We have verified that the expansion converges rapidly and that truncation at finite order does not affect the final sensitivity within the systematic uncertainty. Although the present work focuses on Yukawa-type interactions, the method can be extended to a broader class of non-Newtonian forces. Any deviation from the inverse-square law with a nontrivial distance dependence can, in principle, be probed using a similar differential configuration. Future work will include a detailed evaluation of systematic uncertainties under realistic experimental conditions Tanabe et al. 2025 , as well as optimization of the calibration geometry. In particular, improving the control of geometric parameters and refining the differential configuration will be key to achieving significantly stronger constraints on deviations from Newtonian gravity. VI Conclusion In this work, we have developed a quantitative framework for probing Yukawa-type deviations from Newtonian gravity using a differential gravitational calibrator coupled to a torsion-bar detector Raffai et al. 2011 ; Inoue et al. 2018 . We derived the Yukawa-induced torque based on an exact formulation of the gravitational interaction and expressed it as a function of the interaction parameters (\alpha_{Y},\lambda) . The resulting torque was converted into a strain-equivalent signal through the mechanical response of the torsion-bar system Saulson 1994 ; Ando et al. 2010 , enabling a direct comparison with the detector noise. Using this framework, we evaluated the signal-to-noise ratio and constructed projected sensitivity curves in the (\alpha_{Y},\lambda) parameter space. Our main result is that the experiment reaches a projected sensitivity of |\alpha_{Y}|=2.4\times 10^{-5}\quad\text{at}\quad\lambda=8\,\mathrm{m}, (49) evaluated at the equivalent integration time T_{\rm eq}\simeq 9.25\times 10^{4}\,\mathrm{s}. (50) T_{\rm eq}\simeq 9.25\times 10^{4}\,\mathrm{s}\;\;(\simeq 25.7\,\mathrm{hours}). (51) This corresponds to approximately one day ( \sim 26 hours), indicating that further integration does not improve the sensitivity and that the experiment operates in a systematics-dominated regime. A central finding of this study is that the sensitivity is not limited by statistical noise, but by systematic uncertainty arising from imperfect cancellation of the Newtonian torque, consistent with calibration limits in gravitational-wave detectors Hall et al. 2018 ; Cahillane et al. 2017 . The statistical noise reaches the systematic floor within \sim 10 hours, beyond which further integration does not improve the constraint. This establishes that the experiment operates in a systematics-dominated regime. The error budget analysis shows that the residual uncertainty is overwhelmingly dominated by the geometric parameters of the source masses, in particular b_{S} and b_{L} , each contributing nearly 50\% of the total variance. In contrast, uncertainties in the gravitational constant and detector parameters are negligible. This identifies precision control of the source geometry as the key requirement for improving experimental sensitivity. From the theoretical side, we have demonstrated that the Yukawa torque can be expressed as a unified expansion that consistently reduces to the Newtonian limit. The convergence of the series is rapid due to the small geometric parameter (b/R)^{2} , and truncation at finite order introduces an error well below the systematic uncertainty. Thus, the theoretical prediction can be regarded as effectively exact within the experimental precision. A particularly important and unique aspect of this work is that the differential configuration retains sensitivity even in the large- \lambda limit. While the Yukawa interaction asymptotically approaches Newtonian gravity for \lambda\to\infty , a measurable residual signal remains due to the geometric asymmetry between the L and S configurations. This allows meaningful constraints to be placed even at very large interaction ranges, a feature that has not been fully exploited in previous studies. The resulting sensitivity curve exhibits three distinct regimes: exponential suppression at short interaction ranges, optimal sensitivity at \lambda\sim\mathcal{O}(1\,\mathrm{m}) , and a finite asymptotic sensitivity at large \lambda . This structure reflects the interplay between Yukawa suppression, geometric enhancement, and incomplete Newtonian cancellation. In summary, the framework presented here establishes a direct and quantitative connection between detector performance, calibration systematics, and constraints on non-Newtonian gravity. The key result, |\alpha_{Y}|=2.4\times 10^{-5} at \lambda=8\,\mathrm{m} , demonstrates that sub-Hz torsion-bar detectors provide a powerful and complementary probe of gravity at meter-scale interaction ranges. Future improvements will focus on reducing geometric uncertainties and optimizing the differential configuration, which will directly translate into stronger constraints on Yukawa-type deviations from Newtonian gravity. Acknowledgements. We would like to express our sincere gratitude to Y-C.Lin, M.Hasegawa, T.Kanayama and M.Hazumi for their valuable discussions and continuous support throughout this work. We also acknowledge the support and collaborative environment provided by the Department of Physics and the Center for High Energy and High Field (CHiP) at National Central University, the Institute of Physics, Academia Sinica, the National Institute of Physics, University of the Philippines Diliman. Y.I. is supported by the National Science and Technology Council (NSTC) of Taiwan under Grant No. 114-2112-M-008-006, and by Academia Sinica under Grant No. AS-TP-112-M01. Appendix A Newtonian limit Before proceeding, it is useful to verify that the Yukawa expansion correctly reduces to the Newtonian expression in the limit \beta\to 0 . From the definition of the Bessel polynomials Grosswald 1978 (see Eq. ( 14 )), we have y_{k}(0)=\frac{(2k)!}{2^{k}k!}. (52) Therefore, \frac{y_{k}(0)}{2^{k}k!}=\frac{(2k)!}{4^{k}(k!)^{2}}=\frac{1}{4^{k}}\binom{2k}{k}. (53) Substituting this into the Yukawa kernel expansion (Eq. ( 11 )), we obtain \frac{1}{\sqrt{1-u}}=\sum_{k=0}^{\infty}\frac{1}{4^{k}}\binom{2k}{k}u^{k}, (54) which is exactly the standard Newtonian expansion. We now consider the coefficients appearing in the torque expansion. The Newtonian coefficient can be written as A_{n}=\frac{1}{2^{4n-1}}\binom{4n}{2n}\binom{2n}{n-1}. (55) Using Eq. ( 53 ), the Yukawa coefficient B_{n}(\beta) can be expressed in a compact form as B_{n}(\beta)=A_{n}\frac{y_{2n}(\beta)}{y_{2n}(0)}. (56) This relation makes the structure of the expansion transparent: the Yukawa interaction preserves the Newtonian geometrical dependence encoded in A_{n} , while modifying only the radial dependence through the ratio y_{2n}(\beta)/y_{2n}(0) . In the limit \beta\to 0 , we have y_{2n}(\beta)\to y_{2n}(0) , and therefore B_{n}(\beta)\to A_{n}, (57) showing that the Yukawa expansion smoothly reduces to the Newtonian one. Appendix B Mechanical transfer function of the torsion-bar mode In this appendix, we derive the mechanical transfer function used to convert an external gravitational torque into the corresponding angular response of the torsion bar. We consider the rotational degree of freedom \theta(t) of a torsion bar with moment of inertia I , torsional spring constant \kappa , and viscous damping coefficient \Gamma . The equation of motion under an external torque \tau(t) is I\ddot{\theta}(t)+\Gamma\dot{\theta}(t)+\kappa\theta(t)=\tau(t), (58) which is the standard form of a damped harmonic oscillator Saulson 1994 . Assuming harmonic time dependence, \theta(t)=\theta(\Omega)e^{-i\Omega t},\qquad\tau(t)=\tau(\Omega)e^{-i\Omega t}, (59) Eq. ( 58 ) becomes \left(-\Omega^{2}I-i\Omega\Gamma+\kappa\right)\theta(\Omega)=\tau(\Omega). (60) Therefore, the mechanical susceptibility is given by \chi(\Omega)\equiv\frac{\theta(\Omega)}{\tau(\Omega)}=\frac{1}{I\left(\omega_{0}^{2}-\Omega^{2}\right)-i\Gamma\Omega}.. (61) where the torsional resonance frequency is obtained as \omega_{0}^{2}=\frac{\kappa}{I}, (62) In the high-frequency regime relevant for the present study, (\Omega\gg\omega_{0}) , the inertial term dominates over the restoring term, and the susceptibility reduces to \chi(\Omega)\simeq-\frac{1}{I\Omega^{2}}, (63) where damping has also been neglected since (|\Gamma\Omega|\ll I\Omega^{2}) in this limit. This is the transfer function used in the main text to convert the external GCal or Yukawa torque into an angular response of the torsion bar, consistent with standard treatments of mechanical response in gravitational-wave detectors Saulson 1994 . The corresponding strain-equivalent signal is then obtained through the geometrical conversion factor (\eta_{g}F_{\rm eff}/2) introduced in Inoue et al. Inoue et al. 2026a . Appendix C Convergence of the Yukawa expansion In this appendix, we examine the convergence of the Yukawa torque expansion by directly comparing numerical results obtained with different truncation orders. The Yukawa torque is evaluated as a finite sum truncated at order n_{\rm max} . To assess convergence, we compute the projected sensitivity \alpha_{Y{\rm lim}}(\lambda) using truncation orders n=1 – 4 , while keeping all other parameters fixed to the values used in the main analysis. The results are shown in Fig. 4 . Figure 4: Sensitivity curves \alpha_{Y{\rm lim}}(\lambda) evaluated with truncation orders n=1 – 4 . The curves progressively converge as n increases. The results for n\geq 4 are visually indistinguishable over the entire range of \lambda . The figure shows that the sensitivity curves rapidly approach a stable solution as the truncation order increases. At low truncation order ( n=1 – 2 ), visible deviations are present, indicating that higher-order terms contribute non-negligibly to the signal. However, these differences decrease systematically as n increases. For n\geq 4 , the curves overlap within the resolution of the plot over the full range of \lambda considered. This demonstrates that the series has effectively converged at this order. A localized deviation between truncation orders is observed around \lambda\sim 0.3 – 1\,\mathrm{m} , where higher-order contributions temporarily modify the signal. Nevertheless, this difference becomes negligible once n\geq 4 , and does not affect the final sensitivity curve. These results demonstrate that the Yukawa expansion converges rapidly in the present configuration. We therefore adopt n_{\rm max}=4 in all numerical evaluations. Importantly, the residual differences between successive truncation orders are much smaller than the systematic uncertainty discussed in the main text. Therefore, the truncation error does not limit the sensitivity of the experiment. Appendix D Large- \lambda limit of the Yukawa torque In this appendix, we derive the asymptotic form of the Yukawa torque in the large- \lambda limit and extract the leading residual contribution. According to Eq.( 38 ), we define F_{n}(\beta^{(Y)})\equiv e^{-\beta^{(Y)}}B_{n}(\beta^{(Y)}). (64) Then Eq. ( 38 ) becomes \displaystyle\tau_{Y}(\lambda) \displaystyle=2GM\alpha_{Y}\gamma\frac{m}{R_{S}}\sum_{n=1}^{\infty}\left(\frac{2ab}{R_{S}^{2}}\right)^{2n} \displaystyle\quad\times\left[2^{2n}\left(\frac{R_{S}}{R_{L}}\right)^{2n+1}F_{n}(\beta^{(Y)}_{L})-F_{n}(\beta^{(Y)}_{S})\right]\sin(2\omega t). (65) In the limit \lambda\to\infty , we have \beta^{(Y)}_{S},\beta^{(Y)}_{L}\to 0 , and F_{n}(\beta^{(Y)})=A_{n}+\mathcal{O}((\beta^{(Y)})^{2}). (66) Substituting this, we obtain \displaystyle\tau_{Y}(\lambda) \displaystyle=2GM\alpha_{Y}\gamma\frac{m}{R_{S}}\sum_{n=1}^{\infty}\left(\frac{2ab}{R_{S}^{2}}\right)^{2n} \displaystyle\quad\times A_{n}\left[2^{2n}\left(\frac{R_{S}}{R_{L}}\right)^{2n+1}-1\right]\sin(2\omega t)+\mathcal{O}(\lambda^{-2}). (67) Focusing on the leading contribution ( n=1 ), we obtain \displaystyle\tau_{Y}^{(n=1,\infty)} \displaystyle=2GM\alpha_{Y}\gamma\frac{m}{R_{S}}\left(\frac{2ab}{R_{S}^{2}}\right)^{2}A_{1} \displaystyle\quad\times\left[4\left(\frac{R_{S}}{R_{L}}\right)^{3}-1\right]\sin(2\omega t). (68) The coefficient A_{n} is given by A_{n}=\frac{1}{2^{4n-1}}\binom{4n}{2n}\binom{2n}{n-1},\qquad A_{1}=\frac{3}{4}. (69) Therefore, \displaystyle\tau_{Y}^{(n=1,\infty)} \displaystyle=6GM\alpha_{Y}\gamma m\frac{a^{2}b^{2}}{R_{S}^{5}} \displaystyle\quad\times\left[4\left(\frac{R_{S}}{R_{L}}\right)^{3}-1\right]\sin(2\omega t). (70) This result shows that, unless the first Newtonian-like term is exactly canceled, a finite residual remains even in the large- \lambda limit.