Moisture and measurement flaws may mask filler effects in epoxy nanocomposites

Epoxy nanocomposites loaded with two-dimensional fillers such as graphene nanoplates, graphene oxide and molybdenum disulfide are widely studied for electromagnetic applications, including electromagnetic interference shielding. A new open-access study published in Discover Polymers on 1 August 2026 reports a cautionary finding: when researchers prepared epoxy composites containing 0 to 5 weight percent of these fillers and measured their complex dielectric permittivity with a coaxial transmission/reflection method using a vector network analyzer, the extracted permittivity values came out substantially higher than typical literature benchmarks, and no systematic dependence on filler concentration appeared at all. Rather than presenting this as a discovery of extraordinary materials, the authors treat it as a warning about how easily measurement artifacts and environmental moisture can dominate the dielectric response of such composites.

The work, led by S. Bellucci and carried out with instrumentation at INFN–Laboratori Nazionali di Frascati, builds on an earlier investigation of the electrical conductivity and percolation behavior of the same class of epoxy composites filled with GNP, GO and MoS₂. In that previous study, direct current and low-frequency techniques were used. The new research extends the characterization to microwave frequencies, measuring the complex scattering parameters of annular disk samples inserted into a coaxial line over a nominal range of 0 to 20 GHz. The stated objective was not to establish percolation thresholds or demonstrate shielding performance, but to critically assess the reliability of broadband dielectric permittivity measurements in these materials and to elucidate the role of moisture absorption in masking filler-dependent effects.

The materials themselves were conventional. A commercial thermosetting epoxy resin was cured with an amine-based hardener, a formaldehyde-modified tetraethylenepentamine derivative, added at 8 weight percent relative to the resin. Three fillers were compared: electrically conductive graphene nanoplates with high aspect ratio, insulating and highly polar graphene oxide with oxygen functional groups, and semiconducting molybdenum disulfide, a layered material whose band gap increases as the number of layers decreases, from 1.2 electronvolts for the bulk to 1.9 electronvolts for a single layer. For each filler, composites were prepared at loadings of 0, 0.25, 0.5, 1, 2 and 5 weight percent, calculated with respect to the total weight of the cured composite. The same material batches as the earlier conductivity study were used to keep the two characterizations consistent.

Sample preparation emphasized dispersion and porosity control, both critical for dielectric measurements. Fillers were pre-mixed manually into the liquid resin and then processed in a planetary centrifugal vacuum mixer, which combines rotation and revolution under reduced pressure to disperse filler and remove entrapped air. GNP- and MoS₂-based mixtures received an additional probe-sonication step to break up agglomerates, while GO composites, which disperse more readily in epoxy, did not require extended sonication. After curing at room temperature followed by post-curing, samples were machined into annular disks with target inner and outer diameters of 3 and 7 millimeters; deviations beyond these tolerances were found to hinder insertion into the coaxial cell and degrade measurement repeatability. Notably, no dedicated drying or conditioning procedure was applied before measurement: samples were stored under laboratory ambient conditions, so potential moisture uptake was not suppressed.

The measurement chain used an Agilent/Keysight E5071C ENA vector network analyzer with a 20 GHz frequency option, calibrated with a full two-port procedure using an Agilent 85052D 3.5 mm calibration kit. Permittivity was extracted from the measured reflection and transmission coefficients through a standard transmission/reflection formalism, solving the resulting nonlinear system with a Newton-type root-finding algorithm implemented in Mathematica, initialized with physically reasonable values drawn from reference epoxy systems. The authors are candid about the limits of this procedure: the original acquisition files did not contain a complete record of all instrument settings, including IF bandwidth, source power, sweep points and averaging, and no independent reference dielectric measurement was performed during the campaign, preventing a complete metrological uncertainty budget.

The most striking technical result concerns frequency. Over the full 0 to 20 GHz range, the extracted permittivity spectra show pronounced oscillations, peaks and discontinuities above roughly 3 to 5 GHz for all filler types and loadings. The authors attribute these features not to intrinsic dielectric dispersion of the composites but to amplification of experimental uncertainties by the nonlinear inversion procedure: small phase errors, imperfect sample contact with the coaxial conductors, residual calibration errors and connector or fixture losses can produce large variations in retrieved permittivity, especially in weakly lossy dielectrics where the imaginary part of permittivity is small and noise-sensitive. Consistent with this, the instrument’s specified dynamic range drops to roughly 71 dB between 15 and 20 GHz at an IF bandwidth of 3 kHz. Quantitative analysis was therefore restricted to the band up to 5 GHz, where curves are smooth and reproducible, and the high-frequency data are retained only as documentation of the method’s practical limits.

Within that stable 0 to 5 GHz window, the outcome was unexpected in a revealing way. Neither the real nor the imaginary part of the complex permittivity showed any systematic dependence on filler type or concentration, in contrast to expectations from literature reports for similar composites, where GNP-filled epoxies approaching the percolation threshold typically show rising permittivity and conductivity driven by conductive pathway formation and interfacial polarization. As an additional consistency check, the imaginary part of the permittivity was converted into an effective AC conductivity, used only as an order-of-magnitude check because a full uncertainty propagation was unavailable. The resulting values remained compatible with weakly lossy, non-percolating composites and gave no indication of a continuous conductive network forming in any of the filled systems.

To explain the anomalously high permittivity, the researchers turned to a simplified effective-medium model treating the composite as a ternary system of epoxy matrix, filler and absorbed water, with the effective permittivity approximated by a linear combination of constituent contributions. Using representative literature values for the permittivities of epoxy resins, graphene-based fillers, molybdenum disulfide and water, the model indicated that a water content on the order of 10 weight percent would be sufficient to account for the experimentally observed permittivity enhancement. The authors stress that this figure is an indicative plausibility estimate, not a measured moisture concentration: no gravimetric water uptake measurement, controlled drying experiment or humidity-conditioned comparison was performed, so absorbed water cannot be identified quantitatively as the unique origin of the elevated values. Both the resin and the amine curing agent are known to be hygroscopic, and absorbed water, with its high permittivity and strong dipolar polarization, can substantially alter the microwave dielectric response of epoxy systems.

The study also acknowledges further limitations that keep its conclusions conservative. No SEM, TEM or optical microscopy characterization of the cured composites was performed, so filler dispersion, agglomeration and possible sedimentation before curing could not be independently assessed, and the authors do not claim that dispersion was optimal. For graphene oxide systems, additional uncertainty arises because reported GO permittivity values in the literature vary over several orders of magnitude depending on frequency range, degree of oxidation and measurement technique, making the model estimates for GO composites qualitative benchmarks rather than quantitative predictions. The agreement between experiment and the ternary model in the stable band is accordingly described as consistency in order of magnitude, not precise validation, and the absence of filler trends is interpreted as the combined result of measurement limitations, environmental contributions and possible microstructural heterogeneity.

The takeaway is explicitly metrological rather than performance-oriented. Under the conditions of this study, broadband dielectric extraction by the coaxial transmission/reflection method yielded physically meaningful, reproducible permittivity values only up to approximately 5 GHz; above that, the retrieved parameters were corrupted by oscillations and discontinuities consistent with inversion instability. The authors argue that moisture control, reference-material validation, microstructural characterization and proper uncertainty propagation are prerequisites before broadband permittivity data on epoxy nanocomposites can be trusted for quantitative materials design or EMI shielding assessment. For a field where reported permittivity values, percolation thresholds and frequency trends vary widely even for nominally similar epoxy/filler systems, the message is pointed: some of the scatter in the literature may reflect water in the samples and instability in the measurement chain rather than the fillers themselves.

Subject of Research: Dielectric permittivity measurement reliability in epoxy nanocomposites with two-dimensional fillers

Article Title: Critical assessment of dielectric permittivity measurements in epoxy nanocomposites based on two-dimensional fillers

Article References: Bellucci, S. (2026). Critical assessment of dielectric permittivity measurements in epoxy nanocomposites based on two-dimensional fillers. Discover Polymers, 3(1), Article 19. https://doi.org/10.1007/s44347-026-00056-6

Image Credits: AI Generated

DOI: 10.1007/s44347-026-00056-6

Keywords: epoxy nanocomposites, dielectric permittivity, graphene nanoplates, graphene oxide, molybdenum disulfide, moisture absorption, vector network analyzer, coaxial transmission/reflection, EMI shielding, microwave metrology, effective-medium model, percolation