Enhanced magnetic moment discrimination for multiplex nanoparticle quantification via dual-frequency nonlinearity probing Authors: Timur I Bikulov, Ulrich M Engelmann, Andreas Offenhäusser, Hans-Joachim Krause Published: 2026 ABSTRACT AbstractAdvanced applications of Magnetic nanoparticles (MNPs) in biomedicine based on multiplex MNP distinction require accurate, model-agnostic characterization of their magnetic moment distributions (MMDs). However, the resolving power of conventional MMD reconstruction from a static magnetization curve remains literarily underexplored. Moreover, due to particle-particle interactions, the response of the particle mixture might differ from the linear combination of the original constituents. We explore resolution enhancement in magnetic-moment space by directly probing higher-order magnetization derivatives, benefiting from their ever-increasing field-domain localization. Nonetheless, the direct derivative probing, as it is inevitably conducted dynamically, poses an interpretive problem for the origin of the nonlinearities. Spectral symmetries arising solely under dual-frequency excitation reflect the corresponding origins of amplitude- and rate-related nonlinearities. Using a dedicated experimental setup capable of synchronous demodulation of intermodulation terms, the method is tested on commercial MNP samples and benchmarked with conventional AC-susceptometry and static magnetization data. The binary mixture ratio was quantified with 8.9% deviation, without any prior information about the initial constituents differing by a factor of 3 in their average magnetic moments, potentially allowing the accommodation of three independent contrast channels for multiplex MNP applications as well as qualitatively probing magnetic interaction effects. Subject terms: Biomedical engineering, Nanoscale biophysics, Nanoscale materials Introduction Single-domain magnetic nanoparticles (MNP) have found broad applications across diverse fields, including chemical processing1 and the oil industry2,3, as well as medicine4,5. Some noteworthy examples are: using static magnetic fields, MNPs are effectively employed for precision separation on the microscale, either to remove ionic contamination at parts-per-billion levels or to isolate cells directly from the bloodstream with efficiencies exceeding 70%6–10. Moreover, in response to periodically alternating magnetic fields (AMF), MNP can be remotely heated2 with an alternating magnetic field while providing immediate temperature feedback. This technology appears particularly promising for magnetic fluid hyperthermia (MFH)-based cancer therapy4,11, where cytostatic drugs can be selectively activated through local heating, while not exceeding 42 °C to maintain physiological functions. Finally, using the AMF-driven magnetic particle spectroscopy (MPS) read-out principle, Magnetic Particle Imaging (MPI) allows localization of as little as ~100 pg of MNP with a spatial resolution of 0.7mm12,13. An auspicious approach is MPS’s unique multi-contrast capability, which exploits differences in MNP magnetic moments in response to an AMF to extract rich, contrast-enhanced information from heterogeneous particle ensembles. In the future, such a multicontrasting capability would, for instance, enable simultaneous remote viscosity/temperature measurement, as it provides two or more fully separated telemetry channels and magnetic moment (i.e., MNP core size) analysis based on MNP signal-generating capabilities in different contexts14. Foundational works by Shasha et al.15, Sun et al.16, and others17–19 imply the feasibility of this principle. However, this problem requires a method that yields the so-called core size distribution, which translates in magnetic measurements into a magnetic moment distribution (MMD) for the entire sample particle population. Reconstruction of MMD can be challenging due to a minor contamination of the sample with particles of larger core sizes that can significantly alter the nonlinear response, which is dominated by large particles20. At the same time, the methods of direct core size evaluation (such as TEM) do not allow non-destructive sampling of the whole population. In contrast, indirect methods (such as static magnetization measurement) lack sufficient resolution to precisely distinguish sizes, as explained below. A common analysis of MNP magnetic cores employs transmission electron microscopy (TEM), particularly 3D TEM tomography. It offers21 sub-angstrom resolution (<0.5Å), which, when expressed as the magnetic moment of spherical magnetite, approaches the scale of a single Bohr magneton (9.2⋅10−24Am2). TEM-image-based core size analysis commonly exhibits remarkably narrower size distributions (i.e., lower variance) than their magnetic moment counterparts derived from magnetic techniques using a static field21. This discrepancy is usually attributed to TEM’s inability to access MNP’s total magnetic moment, anisotropies, and inter-particle interactions. The latter two factors also shape the magnetization curve from which the magnetic moment distributions are indirectly obtained through a regularized inverse problem solution, unavoidably introducing a priori information by employing a static magnetization model, i.e., the Langevin curve21. Sometimes, to reduce the problem to an analytical solution, the distribution is inferred from statistical assumptions across the entire MNP ensemble22. Berkov23 noted that the inverse problem for determining the magnetic moment distribution is a Fredholm equation of the first kind, which can be reformulated as a deconvolution problem. In this special case, applying the Fourier transform to the inputs simplifies the solution, as deconvolution reduces to an ordinary division. The convolutional form of an equation can be achieved by assuming odd symmetry in the magnetization curve and by switching to logarithmic variables for the fields and magnetic moments, since the magnetic moment and external field appear as a product in the equations. In other words, the information about magnetic moments is distributed not linearly, as in size measurements with TEM, but logarithmically, requiring logarithmic field sampling. Moreover, Berkov draws attention to the biasing effects of regularization and advocates the use of low-level selection criteria, such as the uncorrelatedness of the fit error. Bender24 employs a similar approach for continuous Tikhonov regularization, aiming to reduce the influence of subjective factors. Nevertheless, the use of regularization is unavoidable due to the presence of noise in measured data. Noise causes deviations from an ideal monodisperse response, leading to peak broadening in the distribution. The degree of regularization sets the interpretation threshold for the noise level and, hence, determines the effective widths of the peaks and resolution in the magnetic moment space. The resolution limit, as analysis of the inverse problem in the Fourier space shows, arises from a limited spectral width of the static magnetization curve. Basically, the wider and smoother the shape of the static magnetization curve (deconvolution kernel), the more sensitive the solution becomes to random additive noise. In this work, we propose that resolution improvement in magnetic-moment space can be achieved by directly probing the higher-order derivatives of the magnetization curve. Due to the sigmoidal topology of a magnetization curve, its derivatives get increasingly field-localized with increasing order. The direct assessment of a higher-order derivative is only accessible from a dynamic excitation of MNP by an alternating magnetic field, which, however, poses an additional problem of how to interpret the measured magnetic response. To address this challenge, this work provides an interpretational framework founded on modulation theory using basic geometrical assumptions about the dynamic magnetization trajectory of an MNP ensemble excited by an improved MPS technique using synchronized dual-frequency field, called frequency mixing magnetic detection (FMMD)14. This framework is assessed experimentally (using a setup described in “Setup of intermodulation spectral term measurement”) by measuring samples in liquid colloid state, as in this state, the inter-particle interactions can be controlled. By employing the dual-frequency measurement scheme, the colloid remains in a magnetization state that is asymptotically close to that of an ac-susceptometry measurement with a dc offset. That is why we verify the response of our proposed technique with static magnetization and ac-susceptometry as described in the Methods section. Two physical models of variable magnetic moment and variable media viscosity are employed to demonstrate the feasibility of directly interpreting their effects from the raw FMMD data. Finally, we experimentally demonstrate the resolution improvement in magnetic moment space by measuring the mixtures of different MNP types using the proposed technique. From this, we conclude that FMMD enables non-destructive, direct quantification of the magnetic moment distribution in unknown MNP samples without requiring prior assumptions on the sample properties. For a well-characterized MNP system, the framework could potentially provide a systematic approach to qualitatively detect the onset of interparticle magnetic interactions. Collectively, these capabilities would offer a foundation for enhancing quality control and assurance in MNP synthesis, characterization, and reproducibility assessment. Methods Theory of dual-frequency particle spectroscopy This section provides the fundamental background necessary for interpreting dual-frequency MNP responses used in this work. Magnetic particle spectroscopy (MPS) commonly employs single-frequency alternating magnetic fields (AMF) with frequencies in the range of fL = 10–50,000 Hz and amplitudes up to hL≤50mT/μ0 to excite dynamic magnetic responses from MNP25. MPS can thereby directly detect the presence of superparamagnetic MNP, evident from their distinct magnetic moment exhibiting higher order harmonics 3fL,5fL,…, which are absent otherwise in biological tissues without MNP26. This pecularity allows MPS to selectively localize MNP by contrast-positive detection. These harmonics are typically extracted by synchronous demodulation with respect to the excitation field (EF). Interpretation commonly assumes a linear approximation of the MPS response27, where the signal amplitude of the fundamental harmonic fL is proportional to the MNP concentration cs, and energy dissipation is extracted from its phase28. However, as this work will demonstrate, such an interpretation is valid only under special conditions and does not readily generalize for higher-order harmonics with n⋅fL,n\inN>1. Instead, a deeper understanding of the physical origin shall be developed and applied here, as demonstrated in this section. In a diluted MNP colloid, where particle-particle interaction is negligible, dissipative forces dominate over inertia, and restoring forces are not present. This prevalence ensures a non-resonant response from higher-order harmonic terms. Within this approximation, the emergence of the higher harmonics at the integer multiples of the fundamental harmonic can be solely attributed to the periodicity of the excitation field het=het+n/fL,n\inZ, which induces a similarly periodic response from the MNPs with mdt=mdt+n/fL. The uniqueness of the period defines a unique phase within each excitation, from which the phases of the higher-order harmonics are coherently derived. The emergence of higher harmonics is caused not only by finite magnetizability, i.e., the saturation magnetization of the MNP (i.e., magnetization is bounded ∣mdt∣10∣A1,8floor∣. For each LF amplitude value hiL, regression parameters were calculated: (a) Pearson’s correlation, (b) p-value, and (c) slope of the phase “rad/order”. Regularized reconstruction To reconstruct the distributions of relaxation times (from single-frequency dynamic magnetization using ACS) and magnetic moments (from static magnetization using MPMS and dual-frequency dynamic magnetization using FMMD), we apply a unified approach to solve the inverse problem with regularization, without any a priori assumptions on the shape of the distribution (e.g., log-normal). Instead, the histogram is directly reconstructed. In this way, assumptions such as mono- or polydispersity are ultimately avoided. The range of parameters this approach can reconstruct with trustworthiness is determined in advance based on the signal’s geometrical features. Consider a bell-shaped curve displaying a global maximum (e.g., the peak in the imaginary part of ac-susceptometry). One can only precisely determine the peak position if it is observed within the method’s operating range (e.g., 10 to 10,000 Hz for ac-susceptometry in this work). Thus, the reconstruction range covers the operating range, with a slight overlap on both sides, and is extended by a factor of 10. The trustworthy region equals the operating range, and the residual regions at the sides are treated as background that cannot be precisely reconstructed. A zero-order Tikhonov regularization is applied to stabilize the solution. Regularization of a higher order (e.g., constraining the higher order derivatives) is not used to avoid imposing physically unreasonable constraints on the reconstructed distributions’ smoothness and to exclude the appearance of artifactual peaks. Reconstruction reduces to solving the following matrix equation using the bounded least-squares algorithm, where b⃗meas is the measured data, 0⃗ is the zero vector, Mrec is the reconstruction matrix, E is the unit matrix, and λ is the regularization parameter. The exact form of the matrix and the input vectors vary across measurement methods (ACS, MPMS, FMMD). A standard non-negative least-squares algorithm implemented in BLAS/LAPACK with Python bindings was used to find the solution vector x⃗. Reconstruction of magnetic moment distribution from FMMD Distribution of magnetic moments, ρmFMMDmiFMMD was reconstructed from FMMD measurements in the range mminFMMD,mmaxFMMD with logarithmic steps miFMMD=mminFMMDmmaxFMMD/mminFMMDi−1M−1, M=300. The range was selected based on the field range of LF-fields, with the range extended ten times on both sides: mminFMMD=0.1⋅αFM/hmaxFMMD=3.11⋅10−20Am2 and mmaxFMMD=10⋅αFM/hminFMMD=8.09⋅10−17Am2. Here αFM=4.04⋅10−21Am2T−1 is a proportionality coefficient connecting the curvature peak of the Langevin curve and the magnetic field19. The reconstruction follows formulation (22), defining the reconstruction matrix Mrec∈RN×M filled with corresponding values of LF-field and magnetic moment: Mreci,j=∣A^1,2(hiL,mjFMMD)∣/maxx∣A^1,2(x,mjFMMD)∣. Measurement vector b⃗meas∈CN was filled with processed measurements bimeas=A1,2hiL. The resulting distribution is obtained by taking the absolute value of the solution x⃗ ρmmiFMMD=x⃗i. The regularization parameter λFMMD was selected using the L-curve method60, optimized in the range λFMMD∈10−3,10. Reconstruction of magnetic moment distribution from MPMS Distribution of magnetic moments ρmMPMSmiMPMS was reconstructed from MPMS measurements in the range mjMPMS∈mminMPMS,mmaxMPMS with M=300 logarithmic steps. The range was selected based on the field range of MPMS: mminMPMS=0.1⋅αFMhmaxMPMS=8.09⋅10−22Am2 and mmaxMPMS=10⋅αFMhminMPMS=8.09⋅10−17Am2. The reconstruction follows formulation (22), defining the reconstruction matrix Mrec∈RN×M filled with corresponding static normalized magnetization curves and moment values: Mreci,j=LmjMPMSμ0hims/kBT. Measurement vector b⃗meas∈RN was filled with processed measurements bimeas=mmshims, as described above. The resulting distribution is obtained by taking the absolute value of the solution x⃗ : ρmmiMPMS=x⃗i. The regularization parameter λMPMS was selected so to achieve the same Full-Width Half Maximum in terms of magnetic moment distribution as for FMMD reconstruction to achieve comparability between both methods. Relaxation time distribution reconstruction from ACS Distribution of cut-off frequencies ρωωiACS was reconstructed from ACS measurements in the range ωiACS∈ωminACS,ωmaxACS with M=100 logarithmic steps. The range was selected based on the frequency range available for this specific ACS: ωminACS=0.1⋅10Hz⋅2π=1Hz⋅2π and ωmaxACS=10⋅104Hz⋅2π=105Hz⋅2π. The following susceptometry model, which includes two cut-off frequencies, was used for fitting the ACS results, This model includes one relaxation time τN lying outside of the ACS working range, nominally called Néel relaxation, and a set of relaxation times τBi expected within the working range are called Brownian relaxation times. Scaling factors aN and aBi constitute a vector of unknowns: x⃗=(aN,(aB)0,(aB)1,…)∈RN+1. The reconstruction follows formulation (22), defining the reconstruction matrix Mrec∈CN+1×M: Measurement vector b⃗meas∈CN+1 was filled with processed measurements bimeas=χ(fjACS). The resulting distribution is obtained by taking the absolute value of the solution x⃗: ρωωiACS=x⃗i,i=1…N. The regularization parameter was selected using the L-curve, optimized in the range λ∈10−3,10. Measurement of dynamic magnetic properties AC susceptometry responses of all MNP stock sample types plus “Co-Fe” samples with varying media viscosity were measured inductively using a Physical Property Measurement System (Quantum Design), ranging from 10 Hz to 10 kHz in logarithmic steps with 5 points per decade (51 points total). The amplitude of the excitation field was 1 mT. The measurement result was averaged over five measurements. Each susceptibility spectrum was measured at zero DC field and at a bias field of 30 mT, imitating the end point of an FMMD measurement. For the three selected samples, AC susceptometry was measured at 10 points across the same frequency range, with bias fields logarithmically distributed from 0.5 mT to 30 mT in 5 steps. Evaluation of asymptotic correspondence between ACS and FMMD Because dynamic measurement is unavoidable to increase magnetic moment resolution, dynamic effects should be considered. Asymptotically, FMMD measurement is expected to converge to an ac-susceptometry with an offset field as fL tends to zero. Thus, a comparison should be conducted between these methods using a physical model in which the dissipation factor is varied in a controlled manner. A model with variable viscosity serves this purpose. As we assume that amplitude effects are dominated by the LF EF component and rate effects by its HF counterpart, the corresponding frequency points of ac-susceptometry are correlated with the IMTs of the FMMD response. The phase and amplitude changes in both methods are compared. A viscosity experiment should reveal how the dissipation factor influences the nonlinear response of FMMD without varying the interaction between MNPs. A coarse response shape comparison is done using characteristic points, such as the field where IMT peaks. A sample for this experiment should be sensitive to viscosity changes, showing changes in the FMMD signal, and at the same time should have a Brownian relaxation time τB in the working range of the Physical Property Measurement System (PPMS) system, so changes in it can be precisely identified at both LF and HF frequencies. Sensitivity to the changes in viscosity is characterized by αR parameter that is defined as the ratio of the mean change between the LF sample points for the liquid sample and the sample fixed in gypsum, normalized by the maximum signal amplitude of the liquid sample, In case αR=0, the particle has a pure Brownian relaxation, and when αR=1, is a pure Néelian particle. Note that αR implicitly depends on the field frequencies applied to the particles. The viscosity model was verified in two ways: by linear regression of the estimated relaxation time on viscosity. In the first case, the media viscosity is estimated using an empirical formula, and the viscosity power is compared to 1. In the second case, the proportionality between relaxation time and resulting viscosity was enforced, and the original syrup viscosity was estimated. Finally, the hydrodynamic diameter is estimated from viscosity changes, assuming it remains invariant throughout the measurements. For most extreme viscosity cases, AC susceptibility is measured with varying offset field and compared to the FMMD response in the complex plane. The asymptotic correspondence is investigated by ac-susceptometry measurements with an additional dc-offset component hDC∈0.5,1.39,3.87,10.8,30.0mT. In each case, the average Brownian relaxation time is estimated using model (23) from a varying part τBi. Field dependence of relaxation time τBhDC as a function of offset component hDC is fitted by a curve using the least-squares algorithm, obtaining a pair of parameters τ0,ξ35, Measurement of the static magnetization curve A superconducting magnetometry Magnetic Property Measurement System (Quantum Design, San Diego, USA) was used to measure the static magnetization curve mmshims of all types of MNP in liquid colloidal form. The magnetic field points hims were varied in geometric progression h±ims=±hminmshmaxms/hminmsi−1N−1, where hminms=0.1mTμ0−1 and hmaxms=1Tμ0−1 in the following sequence: first, (a) even-indexed points h±2ims were measured in descending order, then (b) odd-indexed points h±2i+1ms were measured in ascending order. This scheme provides better coverage of the field than the symmetric scheme and allows for drift evaluation. The rate of field change was limited to ∣h°∣≤10mTs−1μ0−1. The measurements were performed in the Reciprocating Sample Option mode, in which the sample was moved through the input gradiometer at 4 Hz and 4 cm amplitude, making 12 consecutive passes. To improve accuracy, each point was measured three times. The resulting magnetic moment value is obtained in MPMS under the assumption of negligible sample dimensions by fitting a point-spread function to the registered voltage profile. The data were processed as follows: (a) the curve mmshims was centered with respect to zero, then (b) the negative part of the curve was mirrored into the positive half-plane, exploiting the odd symmetry of the curve. Finally, (c) the measurement points mmshims corresponding to 10% near maximum hims≥0.9Tμ0−1 were used to estimate the parameters of the asymptote of the Langevin function mms=msat⋅1−hchr/h+χ∞⋅h, where msat is the saturated magnetic moment of the sample, hchr is the characteristic field and χ∞ is the susceptibility of the linear background. The resulting asymptote fh=χ∞⋅h was subtracted from the magnetization data for the removal of the linear background. Calibration of the magnetic moment reconstruction Calibration coefficients, interconnecting intensity-weighted magnetic moment distributions of FMMD ρmFMMDx and of MPMS ρmMPMSx with saturated magnetic moment, were obtained using two types of MNPs, specifically “Syn-D 50 nm” and “Syn-D 70 nm”. These samples were selected due to their narrow magnetic moment distributions, compliance with a Wiener model, and full coverage of their distribution by the FMMD range mminFMMD,mmaxFMMD. Volumes (~3 µL each) were estimated optically for both samples, “Syn-D 50 nm” and “Syn-D 70 nm” prepared for MPMS measurement, and the total magnetic moment msat was obtained from a static magnetic curve measurement. Calibration coefficients cmFMMD and cmMPMS were obtained first using “Syn-D 50 nm” and evaluated afterwards using “Syn-D 70 nm” by comparing the msat obtained from reconstruction with the one obtained from the MPMS measurement. First, both sample types were sealed in glass capillaries and then measured using MPMS, from which their msat were identified. Following the reconstruction procedure of MPMS, the distribution ρmMPMSx with calibration coefficient cmMPMS was obtained. Subsequently, a capillary with “Syn-D 50 nm” was measured using FMMD; the ρmFMMDx distribution was reconstructed, integrated and cmFMMD was extrapolated using (27), accounting for the sample volume. Then the same type, “Syn-D 50 nm”, was measured in the FMMD test tube in a volume of 50 µL, and the coefficient cmFMMD was obtained again. Finally, calibration is evaluated by repeating the procedure for “Syn-D 70 nm”. Distribution comparison The reconstructed magnetic moment distributions ρmFMMDx∣λ from FMMD and static magnetization curve measurements ρmMPMSx∣λ were compared per-quantile and depicted with quantile-quantile (QQ) plots. The relative error that covers 95% of all points was used as a figure of merit for evaluating the similarity between distributions. Linear regression was used to assess QQ-plots and numerically characterize the difference between the distributions. Evaluation of the system’s resolution for mixed particle systems The magnetic moment resolution in ρmFMMDx was experimentally estimated using the samples “Syn-D 50 nm,” “Syn-D 70 nm,” and their mixtures. The resolution was evaluated using intensity-weighted distributions ρmFMMDx∣λ obtained for various regularization parameters λ. At first, the series were measured, and the ratio that gave approximately equal contributions to the FMMD signal was chosen. Then the peak magnetic moments were identified for both samples, m0SynD50 and m0SynD70. The resolution RFMMDλ was calculated as the ratio for various λ for the sample that gave approximately equal contributions of both peaks, The reconstruction was performed at different values of the regularization parameter λ. The mixture was considered unresolvable if RFMMDλ≤1. Results Measurement system evaluation Our measurement system (described in “Setup of intermodulation spectral term measurement”) demonstrated maximum achievable hL amplitude equal to 26.1±0.1mTμ0−1@10Hz, verified by measurement using a Hall probe (APS11760, Allegro Microsystems). The amplitude for the hH field was 1.6±0.1mTμ0−1@40.5kHz, measured using a 5 mm wire loop. Assuming a uniform magnetization, a cylindrical sample with 5 mm diameter and 5 mm height with magnetization of 1Am−1 (total magnetic moment 9.81⋅10−8Am2) creates a numerically estimated flux of 2.03⋅10−9Wb in the detection coil (DTC). Estimated coupling at 40.5kHz equals 1.89⋅10−4Am2/Vp or 2.68⋅10−4Am2/VRMS in Root-Mean-Square (RMS) terms. With the help of Spectral Analyzer (SA) HP35670A, the system’s noise measured at the end of Low-Noise Amplifier (LNA) and recalculated to its input equals 27nVRMS/Hz or 7.37⋅10−14Am2/Hz (recalculated to the sample volume). The SA’s effective noise bandwidth of 51 kHz and LNA gain of 72 were considered in calculations. The LNA’s noise floor remained unchanged when the measurement head was replaced with a 50-ohm shunt. SA’s self-noise floor measured with a 50-ohm shunt was estimated to 0.526nVRMS/Hz. The ratio between SA measurements and the values obtained after demodulation with the integrated device is 2.56⋅10−2VFMMD/VRMS/Hz. VFMMD is an RMS value, too, and the subscript “FMMD” emphasizes the source. The predicted noise floor for the FPGA setup based on the measurement with SA is 6.912⋅10−10VFMMD. The noise floor of the integrated device obtained from the background measurements equal 2.89⋅10−9VFMMD or 0.112μVRMS/Hz if recalculated at the LNA’s input or 3.02⋅10−11Am2/Hz at the sample’s volume. To increase the dynamic range of the measurement, the active balancing (described in “Active cancellation of interference”) of the system was completed in five iterations, reducing the fundamental harmonic by 79 times down to 91.3μVRMS/Hz (17.6⋅10−11Am2/Hz). The levels of background intermodulation harmonics were 3.87μVRMS/Hz (0.74⋅10−11Am2/Hz) for the second-order harmonic A1,1 and 1.89μVRMS/Hz (0.364⋅10−11Am2/Hz) for the third-order harmonic A1,2. Evaluation of power-supply coupling by replacing the detection coil with a 50-ohm shunt reveals that the background fundamental harmonic reduces ten times in amplitude. A pair of EF phases φH,φL required for estimation of physically accurate phase response and further comparison with AC-susceptometry was obtained by modified FMMD measurements (cf. “Obtaining physically accurate phase”) of the four largest MNP samples of Ocean Nanotech “ON (15-30) nm”. Figure 5 shows the raw measurement (top row) and the result after consideration of the obtained phase pair (bottom row). Determination of the phases of the EFs. a–c Ocean Nanotech samples at IMTs A1,0, A1,±2,A1,±4 before consideration of the EF phases. d–f After correction of EF phases. Table (d) contains identified phase values. The phase difference between the corresponding points of all four measurements is minimized. A constant discrepancy between the values at A1,±2 IMT, 0.160±0.018 rad (N=77;±3σ) is observed, which can be attributed to background asymmetry between A1,−2, and A1,2 caused by self-nonlinearities of the device (cf. “Determining the systematic error of FMMD measurement”) For A1,±4 harmonics the difference is 0.028±0.024rad(N=47;±3σ). MNP samples evaluation All MNP samples (cf. “Samples of magnetic nanoparticles”) were measured at logarithmically decreasing concentrations ranging from 1 mg/ml to 50 ng/ml with a dilution factor of 7.25 to identify the Limit of Detection (LoD). Dysprosium Oxide (III) was measured as a negative control sample. The concentration intercept point of crossing the noise floor and corresponding MNP mass were determined for all samples (Fig. 6a LoD). Samples “ON 15 nm” and “SynD-70” showed the highest detectability with 98 ng and 99 ng of dried particle mass of the detectable limit in 100 µL volume, respectively. Measurement parameters of samples: (1) “ON 30 nm”, (2) “ON 25 nm”, (3) “ON 20 nm”, (4) “ON 15 nm”, (5) “ON 10 nm”, (6) “Co-Fe”, (7) “Syn-D 50 nm”, and (8) “Syn-D 70 nm”. a Limit of detection. b Relaxation type αR. c Short-term SNR level determined using Rice-Nakagami distribution Eq. (16). d Long-Term Drift evaluated using Eq. (20). e The difference between the minimal and maximal argument of the A1,2 response on the complex plane. f Number of points of LF-scan that have 20 dB SNR considered for reliable evaluation of linear phase dependency. g Distribution of slope of the phase vs. order of the harmonic obtained as a result of regression. h Distributions of p-values obtained as a result of regression. Gray line shows 5% confidence level. i Distribution of Pearson’s correlation coefficients obtained for LF-scans. j Evaluation of magnetic-moment distribution comparison. Denotes the relative error range that covers 95% of all distribution bins. k Imaginary part of ac-susceptometry normalized by the full amplitude of the response. Red stroke denotes the maximum of the imaginary part. l Average magnetic moments calculated in the FMMD-accessible range from the Magnetic Property Measurement System (MPMS) and FMMD techniques. Signal quality and stability parameters were determined for all MNP types and a linear reference, Dysprosium Oxide (III). Short-term drift (SNRm) is obtained using Eq. (16) for the measurement A1,2 and is provided in the Fig. 6c for samples at stock concentrations. For comparison, the SNRm value of a linear sample (Dysprosium Oxide) is SNRmDy2O3=1.543dB, and the value for the detuned measurement at A1,1.5: SNRmdetune=1.560dB. Long-term deviation (see Eq. (20)) on the level of a single measurement, Dm was obtained for both stock concentrations and diluted samples (1 mg/mL) shown in Fig. 6d. For a linear sample, the long-term drift coefficient is DmDy2O3=49%, due to signal amplitude comparable with noise floor level. A regression was conducted to assess linearity between IMT’s phase arg(A⋅,n) and its order n for all MNP samples at stock concentrations. The number of points considered for regression is given in Fig. 6f. For “ON 10 nm” linear regression could not be performed due to the low amplitude of the higher harmonics. Phase slope, regression p-value, and Pearson’s correlation are given in Fig. 6g, h, i, respectively. It was observed that the linear dependence of the phases is stronger at the onset and end of the hook response. Sample “ON 25 nm” demonstrated the highest linearity metric Fig. 6i, defined in the “Evaluation of linearity of higher order terms” section, and almost no change in phase in LF-scan (arg(A⋅,n)≈const) compared to other samples Fig. 6e. Varying viscosity and interrelation between FMMD and AC susceptometry By evaluating the responses of all MNP types in liquid and gypsum-fixed form, the relaxation degree parameter αR (Eq. (25)) was identified in Fig. 6b. The sample “Co-Fe” turned out to be the most sensitive to the changes in the Brownian relaxation time upon immobilization, with 54.4% variation. The least sensitive sample was “ON 20 nm”. All MNP sample types were measured with ACS, and the identified peak position (red line) in the imaginary part, shown with a gradient normalized to the amplitude of the ACS response, is provided in Fig. 6k. The sample “Co-Fe” showed the lowest Brownian relaxation time with a peak in the imaginary part at 4.9kHz. Figure 7 shows ACS results contrasted to FMMD hL-scan measurements for the “Co-Fe” MNP system under gradual immobilization with increasing viscosities induced by increasing concentrations of glucose syrup in the carrier medium. Subfigures (a, b) depict the real and imaginary parts of the sample’s susceptometry χACS=χ′+iχ″, and subfigures (c, d) depict FMMD’s amplitude and phase at A1,2. Relationship between AC susceptometry (ACS) and FMMD response for “Co-Fe” MNPs with dominating Brownian relaxation under gradual immobilization from increasing viscosities. Viscosity values are given in legend (a). Figures (a) and (b) show the real and imaginary components of AC susceptometry, respectively. Dots denote measured data, and the solid line denotes the model fit of the superposition of linear models with different relaxation times. Normalized distribution of the relaxation times is presented in (e), from which the average relaxation time for each viscosity case was obtained and plotted versus actual viscosity in (f). Plots (c) and (d) show the amplitude and phase of the FMMD signal, respectively, at the IMT A1,2. A thick red line shows the position of the maximum amplitude point (turnaround point of the hook) of the IMT A1,2. g Shows phases of FMMD IMTs and LF and HF phases derived from ACS. Solid lines denote asymptotic phases of the FMMD responses near small LF amplitudes. Dashed lines denote the phase of the turnaround point of the hook corresponding to the thick red line in (c). Phases for various IMTs are depicted simultaneously: (green) A1,2, (orange) A1,4, (light brown) A1,6, (dark brown) A1,8. Dash-dot lines denote ACS phases at HF (red) and LF (blue). The respective HF and LF frequencies are shown on ACS plots (a) and (b) with gray vertical lines. Due to the measurement device, the Physical Property Measurement System (PPMS), having a frequency range of 10 kHz, the HF phase was extrapolated from the fitted linear model. Despite increasing viscosities from water (ηwtr=9.532⋅10−4Pa⋅s) to (ηsyr=39.8⋅10−3Pa⋅s; estimated for glucose syrup), the peak positions with respect to hL-field amplitude remains almost unaltered in the range hpeak=11.7±0.6mTμ0−1±3σ,N=9, while the peak value dropped by approximately one order of magnitude from the pure water viscosity to the maximum syrup viscosity. The distribution of Brownian relaxation times ρωωiACS was obtained (cf. “Relaxation time distribution reconstruction from ACS”) in the form of normalized histograms given in Fig. 7e. Solid lines in Fig. 7a, b show the back-fitting of the model. From the inverse average of each histogram, the average Brownian relaxation time was calculated in dependence on the viscosity shown in Fig. 7f. Linear regression (dashed line in Fig. 7e) was conducted to verify the viscosity model. The viscosity of the mixture ηmixα was calculated as a geometric proportion between the viscosity of water and syrup ηmixα=ηsyrαηwtr1−α following the approach of Möddel61. The resulting relationship obtained using linear regression τBηmix=ηmix0.906⋅79.4⋅10−3sp−value1.8e−5. Here τB denotes the average value obtained using Eq. (24). For Brownian-relaxation-dominated particles, like “Co-Fe” here, direct proportionality between viscosity and (Brownian) relaxation time holds, τ∝ηmix. Using the back-estimated viscosity of the syrup of ηsyrest=28.0⋅10−3Pa⋅s, the resulting relationship between relaxation time and viscosity can thus be directly derived from Fig. 7g with τBηmix=ηmix⋅0.152s. The discrepancy between the viscosity estimates of syrup between the cases does not exceed 30%. The estimated hydrodynamic diameter equals dH=2τBkBT/πηmix3≈73.5nm. AC-susceptometry values at LF and HF (gray lines in Fig. 7a, b) were extracted, and their phases are provided as blue and red dash-dot lines in Fig. 7g together with phases of FMMD response for four IMTs. The solid line depicts the phase value at the peak value of the hL-scan, and the dashed line corresponds to the asymptotic value of the phase for low-frequency amplitude values. The FMMD phases for IMTs, A1,⋅, change equally and simultaneously for all viscosity values in range 0.742,1.505rad, spanning a total of 0.763rad with a standard deviation of 0.195rad3σ,N=8 for the cases of largest deviations. In AC-susceptometry measurements, the low-frequency (LF) phase changes relatively modestly in the range of 85.3⋅10−3rad. In contrast, the high-frequency (HF) component exhibits significantly higher sensitivity, ranging from 0.294rad to 1.495rad corresponding to a total variation of 1.201rad. Figure 8 illustrates the relationship between frequency mixing magnetic detection (FMMD) with hL-scan and AC susceptometry (ACS) with an offset component hDC exemplified with the “Co-Fe” sample. The figure highlights the comparability in measuring the effects of amplitude-driven phase modulation and its spectral consequences on both dynamic magnetometry techniques. In Fig. 8a, the average Brownian relaxation times (obtained as described above) are plotted as a function of the offset field hDC for three different viscosities. The solid lines represent fits to Eq. (26) with fitting parameters ξ and τ0 given in the Table inside Fig. 8a. A vertical violet line marks the region up to which the linear response regime is assumed to hold, and up to which the relaxation times remain constant. Beyond that limit, the field dependence of relaxation time must be considered (as discussed in the “Introduction”, Eq. (15)). A vertical red line corresponds to the field amplitude at which the FMMD response peaks are obtained from Fig. 7c. Comparing ACS and FMMD responses for sample “Co-Fe”: field dependency of Brownian relaxation time from ACS and formation of hook shape on FMMD response. a ACS-derived Brownian relaxation time as a function of offset field hDC for three different viscosities between (9.5⋅10−4,…,4⋅10−2)Pa⋅s. The red vertical line depicts the mean peak FMMD amplitude, 11.7±0.6mT/μ0. The violet line marks the offset field below which the relaxation time remains approximately constant. b FMMD response shown on the complex plane for the same samples. Red diamonds mark the FMMD signal at peak amplitude, and violet diamonds mark the field to which the linear response regime holds up to which relaxation time remains constant. The two inserts (entitled “ACMS”) show the ACS response at HF (left) and LF (right), with the complex plane shown for comparison. In Fig. 8b, the first IMT A1,2 derived from FMMD measurements of “Co-Fe” at various viscosities are plotted in the complex plane. Two points of interest are marked: The red diamond denotes the peak amplitude of the FMMD response, corresponding to the red line in Fig. 8a. The violet diamond identifies the field range up to which the linear regime holds (<2 mT) beyond which relaxation time becomes field dependent. Within the linear regime, the FMMD response follows a straight trajectory consistent with the Wiener model behavior, as depicted in row 4 of Fig. 1 and in Fig. 2c. However, beyond the violet diamond, the increasing field amplitude begins to modulate the relaxation time, resulting in the phase shift already predicted in Fig. 2c in our proposed model. This phase shift leads to a clockwise bend, or “hook,” in the complex trajectory of the FMMD signal response. The inserts in Fig. 8b further contextualize this behavior in the AC-susceptometry response at HF (left) and LF (right) with the following observations: The top-left inset shows the ACS response at high frequency (HF), which closely resembles the peak positions of the FMMD response (red diamonds). The top-left inset shows the ACS response at high frequency (HF), which closely resembles the peak positions of the FMMD response (red diamonds). The top-right inset shows the ACS response at low frequency (LF), where the ACS response lies on the real axis and remains independent for various viscosities. The top-right inset shows the ACS response at low frequency (LF), where the ACS response lies on the real axis and remains independent for various viscosities. The bottom-left inset provides a magnified view of the FMMD hook region, demonstrating the “hook” shape also for the smallest FMMD signals at the highest viscosities. The bottom-left inset provides a magnified view of the FMMD hook region, demonstrating the “hook” shape also for the smallest FMMD signals at the highest viscosities. Evaluation of magnetic moment resolution Figure 9 presents a comparative evaluation of the ability to reconstruct the magnetic moment distribution from dynamic dual-frequency excitation (via FMMD) vs. static magnetometry (via MPMS), as exemplified for samples with varying core diameters: “Syn-D 50 nm”, “Syn-D 70 nm”, and their mixtures. The numbers in their titles denote their hydrodynamic sizes provided by the manufacturer and do not reflect their magnetic core sizes, which determine their magnetic properties. The batches of these MNP samples were selected, as they demonstrated the narrowest non-overlapping magnetic moment distributions reconstructed from both FMMD and static magnetization (MPMS) measurements (see below) among the MNP types presented in Supplementary Fig. 1. Thus, these two types were suitable for experimental evaluation of resolution in magnetic moment domain. Their mixtures were prepared as described in the “Methods” section, and their magnetic moment distributions were reconstructed using MPMS and FMMD. Evaluation of magnetic moment resolution of FMMD. c Distribution of magnetic moments obtained from samples of Synomag-D 50 nm (#1) and Synomag-D 70 nm (#5) and their mixtures (#2,3,4), reconstructed from FMMD response A1,2, presented in (b). Dots: raw measurement. Solid lines: back-reconstructed signal. d The distribution of magnetic moments is reconstructed from static magnetization curves for the same samples. a contains corresponding static magnetization curves. Table (d) shows the saturation magnetization moments of the samples. Red lines in (c, d) denote the trustworthy reconstruction region of FMMD. Column (e) shows a quantile-quantile comparison between distributions. Red dashed lines denote the relative error that includes 95% of all distribution points. Figure 9a shows the static magnetization curves obtained for the samples via MPMS (cf. “Measurement of the static magnetization curve”), plotted as sample magnetization normalized to the maximum value versus the applied static DC magnetic field. The model back-reconstructed response (solid) overlaps with the experimental data (dots). The inset in Fig. 9d contains the estimated values of the saturated magnetic moment. The static curves of the original constituents, “Syn-D 70 nm” (#5) and “Syn-D 50 nm” (#1), appear distinct: the onset for “Syn-D 70 nm” (#5) takes place at smaller field values compared to “Syn-D 50 nm” (#1), due to a larger average magnetic moment. All their mixtures lie between the original curves, shifting proportionally to the volumetric ratio of the constituents. Figure 9b demonstrates the corresponding FMMD measurements in the complex plane (see “Measurement of intermodulation terms”). The initial constituents (#1) and (#5) have narrow “hook” widths following the Wiener model, whereas their 50:50 mixture (#3) demonstrates increased width. Such behavior corresponds to the case discussed in Fig. 2c, where a mixture of two samples that follow a simpler Wiener model yields a more complex amplitude-driven phase modulation. In Fig. 9c, the FMMD-based magnetic moment reconstructions are shown (cf. “Reconstruction of magnetic moment distribution from FMMD”). The red vertical lines denote the trustworthy resolution range of the FMMD method (cf. “Regularized reconstruction”) beyond which peaks cannot be reliably localized and must be considered as inaccessible background. The mixture samples (#2, #3, #4) are reconstructed as bimodally distributed without any prior knowledge of the original constituents, meaning that the FMMD measurements for #1 and #5 were not used in the reconstruction at all. The heights of the peaks in bimodal distributions (#2, #3, and #4) scale proportionally to the volume proportion of the corresponding constituent. The inset table in Fig. 9c shows the amplitude ratios of the reconstructed bimodal distributions, which can be directly compared with the volume ratios given in the legend. The position of the peaks of bimodally distributed samples (#2, #3, and #4) lies in the vicinity of the peaks of original constituents (#1 and #5). Constituent (#1) has a peak at 1.113aAm2, and the peaks of bimodal distributions lie in the range 1.187±0.199aAm2(±3σ,N=3), whereas the constituent (#5) has peak at 3.411aAm2 and the corresponding peaks of bimodal distributions lie in the range 3.455±0.0903aAm2(±3σ,N=3). Thus, relative error in peak determination equals 6.64% and 1.28% for constituents (#1) and (#5), respectively. The critical regularization parameter beyond which the mixture is considered as unresolvable was determined using the procedure “Evaluation of the system’s resolution for mixed particle systems.” and equals λcrit=0.3. The corresponding magnetic moment distribution reconstructions for MPMS-based measurements are shown in Fig. 9d. The initial constituents (#1 and #5) are resolved as distinctive peaks, with their positions overlapping those of the corresponding peaks in the FMMD-reconstructed distributions (Fig. 9c). The comparison between two types of reconstruction is conducted using per-quantile comparison (Q-Q plots) shown in Fig. 9e, obtained as described in “Distribution comparison”. The figure sequence corresponds to the order of the samples in the legend. The relative error δ that accommodates 95% of all distribution bins serves as a cumulative figure of merit of a coincidence between distributions, and is depicted with red dashed lines. Note that the relative error is used because of the logarithmic scaling of the magnetic moment distribution. For initial constituents “Syn-D 50 nm” and “Syn-D 70 nm”, these errors equal 53.4% and 46.2%, respectively. Additional cross-sample comparisons throughout each reconstruction method is provided in the supplementary materials in Supplementary Fig. 6 and Supplementary Fig. 7. From these comparisons, one can see that two distributions can be considered relatively alike if the limit relative error does not exceed δ≤100% and the curve crosses the diagonal line. Note that all Q-Q plots have a crossing with a diagonal, indicating a cogent degree of agreement in FMMD- and MPMS-derived magnetic moments. Furthermore, taking into account the comparison in Fig. 6l, we conclude that FMMD captures the correct order of average magnitude of magnetic moment evaluated in the FMMD-accessible region for all particle types used in this work except “ON 10 nm” which appears to have a small average magnetic moment lying outside of the FMMD-accessible range. The mixtures were reconstructed as unimodal distributions, unlike FMMD-based reconstruction, with the peak positioned between the initial peaks and shifting proportionally to the relative volumetric ratios of the constituents, consistent with the observation from the static magnetization curves. These magnetic moment distributions should not be interpreted literally in all cases. From a statistical physics perspective, they represent the partition function ZH,φ, where each magnetic-moment bin μi contributes to the corresponding energy level μ0μiH with weight Ci, For a magnetic colloid composed of several non-interacting particle types, each individually obeying the Langevin magnetization model, the magnetic-moment distribution interpretation naturally arises from the corresponding partition-function formulation. When additional energy levels appear, i.e., as a result of particle interactions or internal magnetic anisotropy, the validity of this interpretation deteriorates. The resolution analysis, conducted using the Backus-Gilbert method62 (see “Resolution analysis” in supplementary materials), for both FMMD and MPMS reconstructions with the used regularization parameters, yielded resolution factors of 2.68 (Supplementary Figs. 5) and 3.65 (Supplementary Fig. 4), respectively. These factors denote full-width-at-half-maximum (FWHM) in logarithmically scaled magnetic moment space: two monodisperse constituents differing in their magnetic moments by this factor will be resolved. Recall that the regularization parameter is fixed for both MPMS and FMMD reconstructions to λ=0.1, where it was chosen such that the factual FWHMSyn70nm≈1.9 of the narrowest constituent (“Syn-D 70 nm”, see Fig. 9c, d) is approximately the same. Simultaneously, this regularization level safely exceeds by an order of magnitude the noise-determined lower boundaries determined in Backus-Gilbert analysis (Supplementary Fig. 3c), which equal ~10−3 and ~10−4 for MPMS and FMMD, respectively, when 1% relative error noise level is targeted. Thus, the obtained results can be characterized as being rather overregularized than overfitted. The FMMD-based MMD reconstruction has a narrower trustworthy resolution range (see the resolution matrix in Supplementary Fig. 3a) but appears more stable to noise in the data than the MPMS-based one. The latter was obtained by analyzing the diagonality metric of the resolution matrix (Supplementary Fig. 3b) for both measurement methods under various additive noise conditions (Supplementary Fig. 3c). Stability is quantified as the lower bound on the regularization parameter, above which the resolution matrix’s diagonal elements smoothly depend on it. For comparability with other works17,18,63–66, a non-negative least-squares-based linear decomposition of the mixture signals over the initial constituents was performed for both MPMS and FMMD measurements, with results summarized in Supp. Tab. 1 in supplementary materials. The relative deviation between the expected volumetric fraction and the obtained ratio did not exceed 10% for both methods and appeared approximately identical for both methods. Finally, the samples (#1) and (#5) were used for calibration of coupling coefficients as described in “Calib. of MMD”. MPMS coupling coefficients obtained for (#1) and (#5) are equal to cmMPMS=1.334Am2 and cmMPMS=1.321Am2, respectively, demonstrating ~1% difference. Estimates of lower boundaries of FMMD coupling coefficients showed a higher deviation of ~29% with values cmFMMD=1.269Am2 and cmFMMD=1.647Am2 for (#1) and (#5), correspondingly. Discussion In this study, we proposed a method for enhancing the resolution of MNP magnetic moment estimation by directly assessing higher-order derivatives of the magnetization curve. The technique relies on the synchronous measurement of intermodulation spectral terms (IMTs) that emerge in the magnetization response under dual-frequency excitation. Although the direct assessment of higher-order derivatives through synchronous IMT measurement is well known in other fields67, to the best of our knowledge, this is the first time it has been applied to the non-destructive reconstruction of the magnetic moment of MNPs with enhanced resolution. Developed within a modulation theory framework using fundamental geometric assumptions, the polynomial equilibrium model introduced in this work provides an interpretational basis, allowing direct assessment of the magnetic moment of MNP from the raw FMMD signal. In this regard, our work extends the previous work of Ilg68 and Sun69, who also used polynomial models to explicitly interpret higher-order spectral terms. The former work remarks that the phase of MPS response is independent of excitation field amplitude, upon which our work adds interpretational range by addressing the spectral phase symmetries and their application to interpreting the FMMD signal. The latter work proposed linearly spaced amplitude scans for MPS to reconstruct the magnetic moment distribution, an approach we further developed here and provide supporting evidence that logarithmic field-amplitude sampling maps the FMMD raw signal to magnetic moment space, allowing its accurate reconstruction with increased resolution. Explicitly, we exploit the spectral symmetries for amplitude and phase modulation that arise from the interplay of finite magnetizability and relaxation processes and appear as corresponding components in the complex envelope of the measured FMMD signal. We exploit these symmetries experimentally and demonstrate that the shape of the magnetic response of MNP during FMMD, when plotted in the complex plane (coined “hook”-plot), contains directly interpretable input on whether the MNP sample is governed by an equilibrium behavior or else behaves like a more complex system governed by amplitude-driven phase modulations (cf. Figure 2c). In fact, the latter appears as curvature points in the complex plane (and simultaneously in the spectral domain), forming an asymmetry around the carrier frequency directly originating from the dual-frequency excitation field (as shown in Fig. 9b and Supplementary Fig. 1b). To capture the aforementioned signals, we introduced an FPGA-based measurement setup that enables reliable synchronous amplitude and phase demodulation of dual-frequency-induced intermodulation terms (IMT). To allow IMT measurement in a narrow band around HF, an interference-compensation technique was proposed in “Active cancellation of interference,” achieving a 79× reduction in interference amplitude. The necessary amplitude requirement for synchronous measurements is exceedingly fulfilled by at least three orders of magnitude over the noise floor (RMSvalue:~10−9VFMMD) for all samples, except “ON 10 nm”. This sample’s deviation toward a more substantial influence due to noise is attributed to its experimentally measured small magnetic moment in Supplementary Fig. 1d. This translates directly to the sample having the smallest mean particle size, which is also confirmed by simulations to diminish the overall FMMD signal generation20, thereby increasing susceptibility to noise. The fulfillment of the sufficient phase noise requirement is reflected by the maximum-likelihood-estimated Rice-Nakagami average Signal-to-Noise Ratio (SNR) parameter of all samples fulfilling the necessary amplitude requirement (s. Figure 6c), yielding at least 2 dB gain compared to the noise floor (SNRmdetune~1.5dB). Furthermore, the relative deviation in signal stability during long-term measurements for each individual MNP sample, cf. Figure 6d, does not exceed 10% for stock or even 3% for the normalized concentration of each sample. These levels confirm an acceptable degree of long-term stability for the measurement technique and for each commercial sample used. The percentage of relative change also falls within the range of deviation between measurements on identical MNP via magnetic particle spectroscopy (MPS) and imaging (MPI), reported by Paysen et al.70 to be 5% or lower, suggesting that our proposed method is comparable to these established techniques. Based on the phase symmetries of the polynomial equilibrium model described by Eq. (8), a method is proposed for identifying the physically relevant directions of real and imaginary axes on the complex plane of the FMMD response matching with ACS measurements, exemplified for sample “Co-Fe” (cf. Figure 7). The mutual correspondence between both techniques after phase correction is shown in Fig. 8b, where both the FMMD response and the ACS are probed at HF and LF, are plotted on the complex plane. Across samples with varying viscosity, the geometry of FMMD responses replicates the distribution of ACS responses at HF. The differences between the phases of higher-order IMTs in Fig. 7g appear insignificant, remaining generally congruent, consistent with the theoretical model Eq. (26). Reduction of Brownian relaxation time by more than one order of magnitude caused by viscosity alteration (cf. Figure 7f) is in accordance with theoretical and experimental predictions for single-frequency excitation71. Applied here to a dual-frequency excitation, the FMMD response scales and rotates without affecting the field dependence of the phase modulation, exactly as the latter part of Eq. (26) suggests. As a result, the field value corresponding to peak position (Fig. 7c) remains almost invariant with viscosity variations. Thus, the relative changes in FMMD-response’s phase can be attributed to the field dependence of relaxation time, while the average direction of the response on the complex plane corresponds to ACS HF susceptometry. The former is in accordance with single-frequency excitation, as shown by Yoshida and Enpuku71, while the latter allows for an estimation of the hydrodynamic size of the probed particles72. Indeed, by modeling the influence of varying viscosity, we have shown that the change in the relaxation behavior due to gradual immobilization of MNP shifts the peak position of the highest signal on the complex plane, but it does not change the nonlinear properties of the static magnetization curve. Thus, the EF amplitude corresponding to the peak remains unaffected, as shown by the red line in Fig. 7c. A similar physical model was used in the work of Dieckhoff73, who also analyzed AC susceptibilities under various Néel and Brownian relaxation times. These results coincide with those of Model61 in terms of signal loss and with those of Sun in terms of the Brownian relaxation time16. The latter work also shows that the reconstructed magnetic moment remains independent, whereas the reconstructed Brownian relaxation time varies. Finally, the work provided a comparison between the static magnetization curve obtained from MPMS and the dynamic magnetization behavior assessed via FMMD. Applying a distribution-weighted Langevin curve equilibrium model that combines static and dynamic magnetization curves enables comparison of the reconstruction performance of magnetic moment distributions from both measurements. Acknowledging the accessible range within which FMMD is able to trustworthily reconstruct the magnetic moment distribution, our results suggest that FMMD yields a magnetic moment distribution with improved resolution, as compared to MPMS (see Fig. 9c, d). This observation is further confirmed when comparing the magnetic moment reconstructions for samples of predetermined mixing ratios of known samples (Fig. 9). Here, FMMD demonstrates improved resolvability by allowing bimodal sample identification (which static MPMS fails to resolve). Furthermore, FMMD is able to quantify the mixing ratio: In fact, the 50:50 mixture was experimentally reconstructed as a mixture with ratio 52.3:47.7, the 80:20 mixture was reconstructed with 71.4:28.6 ratio and, finally, the 20:80 mixture was resolved with 28.7:71.3 ratio of peaks, yielding a maximum deviation of 8.7% from the expected mixing ratios. Currently, as reported in the literature, mixture analysis is conducted by decomposing the measured response into a linear combination of the responses of the initial constituents, as in the works of Rauwerdink17,18, Achtsnicht54, Pourshahidi74, and Viereck65, which require a priori assumptions. The relative concentration estimation errors reported range from 2.3% by Tu et al.18 to 10% by Viereck et al.65. For compatibility, we conducted a similar straightforward linear decomposition for both MPMS- and FMMD-measured mixtures. The worst-case error does not exceed 11% for the FMMD case and 10% for the MPMS case. We consider a priori widely accepted MPMS measurements to be more reliable and attribute the observed comparable method-wise errors to the possibility that some particles interacted, leading to a deviation from a pure linear combination. In the supplementary materials, we provide additional mixing experiments in which we mixed Ocean Nanotech 30 nm and 20 nm samples. These samples appear stable when kept separate, but completely sediment in ~1 h after mixing. On the reconstructed distributions of these samples, one can clearly observe the appearance of magnetic signals that would stem from new fictional particles with smaller magnetic moments than those appearing in the individual, non-mixed MNP samples. This effect can be attributed to the particles’ dipole-dipole interactions in clusters that create additional energy barriers, requiring a higher field amplitude for remagnetization and hence appearing as smaller particles in our measurements. This experiment demonstrates that linear decomposition cannot identify such a change in the mixture. In contrast to the conventional linear decomposition approach, our method can differentiate bimodal sample distributions without a priori assumptions or additional information about the constituent components of the sample mixture. In the FMMD-accessible range 0.311–8.09 aAm2, three-fold separated constituents potentially can accommodate three contrast lines. This approach can therefore facilitate the accurate characterization of heterogeneity in particle size, composition, and magnetic anisotropy, providing valuable information for synthesis and quality control, which is essential for applications in biosensing, magnetic hyperthermia, and magnetothermal stimulation. Furthermore, the method may allow for assessing well-characterized MNP systems for the onset of particle-particle interaction effects (as discussed and shown for the combination of 20 and 30 nm ON particles above) thus serving as a convenient means for rapid, non-destructive quality control and assurance of large(r) scale MNP production or individual characterization for research application as well as scale-up synthesis for commercialization processes. Further work should include the interpretation of the multimodal peaks appearing in the magnetic moment distributions of physically monodisperse samples. We assume that these peaks can be attributed to anisotropy barriers in the particles, depending on material, size, and shape50, potentially allowing us to estimate the effective anisotropy from FMMD measurements as well. Samples that act as described by the Wiener model, i.e., without amplitude-driven phase modulation, enable multicontrast dynamic information extraction, such as simultaneous measurement of viscosity and temperature, as already demonstrated by Draack et al. individually for viscosity75 and temperature76 in MPS. Additionally, assuming spherical particles, the magnetic moment can be easily converted to a mean particle core size, allowing magneto-physical particle analysis. In this way, the framework proposed in this work holds great promise for FMMD-based material characterization of MNP samples and meaningful advances for biomedical applications. We note, however, that exact requirements on particle systems are needed for such multi-dimensional analysis, whose concrete formulation remains to be carefully determined. Supplementary information Transparent Peer Review file Supplementary materials PDF Acknowledgements The authors thank all reviewers for their valuable suggestions, which significantly improved the paper. T.I.B. thanks Christoph Buchal and Anastasia Gerzhik for numerous fruitful discussions, for manuscript editing, and for their continuous support. The authors thank Norbert Wolters for his support in designing and fabricating the electronics; Dieter Lomparski for assistance in mechanical engineering; Ali Pourshahidi for help with sample preparation; Fritz Westphal for fruitful discussions and for providing the nanoparticle samples; and Oleg Petracic, Shibabrata Nandi, and Berthold Schmitz for assistance with MPMS and ACS measurements. Author contributions T.I.B., U.M.E., A.O., and H.J.K. conceived the study. T.I.B. developed the models and the software code used in the study. T.I.B., U.M.E., and H.J.K. contributed to validation. T.I.B. conducted research and performed experiments, visualization, and data presentation. H.J.K. and A.O. acquired funding and supervised effort on this work. T.I.B. and U.M.E. wrote the manuscript. All coauthors reviewed and edited the manuscript and provided approval of the final version. Peer review Peer review information Communications Engineering thanks Kai Wu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: [Anna Bakenecker] and [Rosamund Daw]. A peer review file is available. Funding This work was supported by the German Federal Ministry of Agriculture, Food and Regional Identity (BMLEH), project 281C406C21. The funder had no role in study design, data collection and analysis, the decision to publish, or the preparation of the manuscript. The content is solely the responsibility of the authors. Open Access funding enabled and organized by Projekt DEAL. Data availability The measurements of the magnetization curves, ac-susceptometry measurements, and dual-frequency-nonlinearity probing measurement data including the post-processing scripts that support the findings of this study are available in Jülich DATA repository with the identifier 10.26165/JUELICH-DATA/93REE677. Code availability The increase in resolution in magnetic moment space, central to this work, is demonstrated using a proposed algorithm for solving the inverse problem with zero-order regularization. The code implementing corresponding mathematical procedures described in the Methods sections “Regularized reconstruction”, “Reconstruction of magnetic moment distribution from FMMD”, and “Reconstruction of magnetic moment distribution from MPMS” is available in Jülich DATA repository with the identifier 10.26165/JUELICH-DATA/93REE677–82. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Change history 8/21/2026 In the original version of this article, the mathematical fonts were rendered incorrectly, leading to incorrect font sizes for subscripts, superscripts, parentheses, and other equation elements. These have now been corrected. Contributor Information Timur I. Bikulov, Email: timur.bikulov@rwth-aachen.de Hans-Joachim Krause, Email: h.-j.krause@fz-juelich.de. Supplementary information The online version contains supplementary material available at 10.1038/s44172-026-00713-5. References Associated Data Supplementary Materials Transparent Peer Review file Supplementary materials PDF Data Availability Statement The measurements of the magnetization curves, ac-susceptometry measurements, and dual-frequency-nonlinearity probing measurement data including the post-processing scripts that support the findings of this study are available in Jülich DATA repository with the identifier 10.26165/JUELICH-DATA/93REE677. The increase in resolution in magnetic moment space, central to this work, is demonstrated using a proposed algorithm for solving the inverse problem with zero-order regularization. The code implementing corresponding mathematical procedures described in the Methods sections “Regularized reconstruction”, “Reconstruction of magnetic moment distribution from FMMD”, and “Reconstruction of magnetic moment distribution from MPMS” is available in Jülich DATA repository with the identifier 10.26165/JUELICH-DATA/93REE677–82.