Prediction of Nuclear Magnetic Resonance Surface Relaxivity for Shales Authors: Jiuhui Cheng, Shuitao Zhang, Xiaolong Song Published: 2026 ABSTRACT Surface relaxivity is a key parameter for the inversion of pore size distribution via nuclear magnetic resonance measurements. This paper aims to predict the surface relaxivity of shales based on other available experimental data, including chemical elements, minerals, and magnetic susceptibility. The partial least-squares regression method is conducted by mineralogy and chemical elements, and their root-mean-square errors are 0.01 μm/s and 0.05 μm/s, respectively. The regression coefficients show significant influence from several elements, such as iron and titanium. The effect of the magnetic characteristics of shales on the surface relaxivity is better illustrated by mineralogy, especially magnesium in clay minerals. This issue can be attributed to the magnetic susceptibility of shales. The results reveal a strong positive correlation between magnetic susceptibility and surface relaxivity. In summary, exploring the determinants of surface relaxivity can speed up and simplify the current calibration. Introduction Nuclear magnetic resonance (NMR) has been widely used to characterize fluid-saturated porous media, such as cement and rocks. , By using different pulse sequences, NMR has the capability to measure extensive information about rocks, including porosity, permeability, wettability, fluid type, and fluid diffusivity. − At present, the most widespread use of NMR is to obtain the pore size distribution of rocks. Nuclear magnetic resonance of fluid in rocks is governed by three relaxation mechanics: bulk relaxation, surface relaxation, and diffusion relaxation. Generally, the relaxation in rocks can be expressed by 1 1T2=1T2B+ρ2SV+Dγ2G2τe212 where T 2 denotes the relaxation time of fluid, T 2B is the bulk relaxation time of fluid, for instance, 1.7 s for bulk water, ρ2 represents the surface relaxivity, S/V is the surface-to-volume ratio of the pore space, D is the diffusion coefficient of fluid, γ is the gyromagnetic ratio of the spin fluid molecule, G is the magnetic field gradient, and τ e is the echo time. Low-viscosity fluids (e.g., water, low-salinity brine, and light oils) have longer bulk relaxation time, typically on the order of a few seconds. This second term in eq represents surface relaxation, which is closely associated with the pore size distribution of rocks. Considering the presence of a larger surface-to-volume ratio for pore space in rocks, fluid molecules will frequently hit the pore surface due to diffusion, resulting in stronger surface relaxation. The surface relaxation time of fluid inside rocks generally ranges from milliseconds to hundreds of milliseconds. Thus, the bulk relaxation can be ignored. The third term in the eq is due to the magnetic field gradient. The gradient magnetic field can be removed in some NMR tests, such as CPMG measurement. The internal magnetic field gradient induced by the rock itself is weak, and the effect of diffusion relaxation can be minimized by using a shorter echo time. As mentioned above, eq can be rewritten as 2 1T2=ρ2SV Accordingly, NMR characterization of the pore size distribution greatly relies on the surface relaxivity. The surface relaxivity depends on the interaction between the fluid and the solid. At present, researchers have shown that the strength of the interaction is governed by the paramagnetic substances on the solid surface. − Thus, it seems feasible to predict the surface relaxivity from the magnetic characteristics of the rock. Surface relaxivity is generally obtained by laboratory measurements, such as nitrogen adsorption, mercury intrusion porosimetry, and centrifugation methods. − Laboratory measurements require coring, a very expensive process. Logging generally provides formation information, including elements, minerals, and magnetic susceptibility, which can be used to correlate with surface relaxivity. To address this issue, the partial least-squares (PLS) method is introduced, which minimizes the interactions among mineral or elemental species. Thus, surface relaxivity can be predicted by the log information using the PLS regression method. This paper aims to predict surface relaxivity based on other available experimental data. The properties of the shale, selected as the sample, are characterized. Free induction decay was implemented to measure magnetization decay. The results show that the surface relaxivity can be predicted with small errors based on the PLS method. The effect of sample properties on the surface relaxivity is analyzed according to the regression coefficient. It is found that the surface relaxivity seems to be positively correlated with the magnetic susceptibility. Finally, the main conclusions are drawn. Methodology Characterization of Materials Five shale samples obtained from selected geographic locations in China were used in this study. The samples were first roughly crushed and then mechanically pulverized into a powder using an agate mortar. The sample powder was screened through a 320-mesh sieve to ensure a uniform particle size. The X-ray diffraction (XRD) analysis was performed on Rigaku Ultima IV using Cu–K alpha radiation (V: 40 kV and I: 40 mA) at ambient conditions. The scanning parameters were 5–70 degrees for 2θ, with a step size of 0.02 degrees and a scanning speed of 0.5 degrees per minute. Tables and present the mineralogical composition and contents of the shale samples. Shale samples contain clay minerals in varying amounts, and the main clay mineral is illite. In addition, a small amount of pyrite is present, which is related to surface relaxivity. The surface relaxivity of the samples has been measured by the spin echo method in previous studies. 1XRD Mineralogical Compositions (wt %) of the Shale Samples, and Data Were Previously Published. QuartzPotassium feldsparPlagioclaseDolomiteCalcitePyriteTotal claySampleSiO2 (wt %)KAlSi3O8 (wt %)NaAlSi3O8∼CaAl2Si2O8 (wt %)CaMg(CO3)2 (wt %)CaCO3 (wt %)FeS2 (wt %)- (wt %)S123.42.512.310.601.749.5S236.91.6610.671.836.1S340.71.36.64.622.61.922.2S427.80.64.112.837.91.315.5S555.80.51.612.516.90.712 XRD Mineralogical Compositions (wt %) of the Shale Samples, and Data Were Previously Published. 2Clay Minerals (wt %) and Surface Relaxivities of the Samples IlliteChloriteMixed illite/smectiteMass fraction of smectite crystal layersurface relaxivitySampleK2Al4(SiAl2)O20(OH)4 (wt %)Mg10Al2(Si6Al2)O20(OH)16 (wt %)- (wt %)- (wt %)- μm/sS1719.419.6102.59S258.623.817.6102.36S372.35.622.1101.67S498.11.9//1.46S594.25.8//1.34 Clay Minerals (wt %) and Surface Relaxivities of the Samples Energy-dispersive spectroscopy (EDS) measurements were conducted by scanning electron microscopy (SU8010; Hitachi High-Technologies Corporation). EDS can observe elements with atomic numbers greater than 6. Thus, hydrogen disappears in the elemental spectrum. The amounts of elements can be inferred from the relative peak intensities in the EDS. The elemental compositions of the samples are listed in Table . These samples mainly contain carbon, oxygen, and silicon elements, whereas sodium, magnesium, potassium, and titanium do not exceed 2 wt %. 3Chemical Element (wt %) of the Samples CONaMgAlSiSKCaTiFeSample(wt %)(wt %)(wt %)(wt %)(wt %)(wt %)(wt %)(wt %)(wt %)(wt %)(wt %)S123.6351.330.521.125.7812.220.341.920.690.32.14S229.148.580.371.083.7311.770.451.521.520.181.69S328.8348.670.170.542.4613.30.570.953.7300.77S415.9953.420.210.842.8814.160.481.389.310.171.14S518.725200.771.8219.480.370.815.0300.98 Chemical Element (wt %) of the Samples Partial Least-Squares Regression Method The PLS regression method was used to predict NMR surface relaxivity based on mineral and elemental data. This process was performed using the PLS package in the R language. PLS is a linear regression equation to predict the response variable matrix Y (NMR surface relaxivity in this study): 3 Y=AX where A is the regression coefficient matrix, and X is the variable matrix (mineralogy composition or element composition in this study). PLS regression is described in detail in previous publications. , Cross-validation uses the “leave-one-out” method in this PLS regression method, which is important to avoid overfitting. With cross-validation, it is able to assess how well the method predicts new samples that are not included in the original samples. Free Induction Decay (FID) Measurements Compared with CPMG measurement (T 2), the free induction decay curve is extremely susceptible to the influence of the internal magnetic field gradient of shale samples. Thus, extra samples were saturated with deionized water and prepared to conduct FID tests. This test records data per 0.5 μs until 30720 data points were obtained. To ensure a sufficiently high signal-to-noise ratio for the FID signal, 128 scans were adopted for each test. Generally, the dephasing length is related to the internal gradient magnetic field, which can be expressed as 4 l=χB0G where χ is the magnetic susceptibility contrast, and B 0 is the magnetic field strength, 0.05 T in this study. After obtaining the FID curves, the time domain information can be converted into frequency domain information to facilitate subsequent analysis Results and Discussions Figure presents the predicted surface relaxivity based on minerals and elements. The results illustrate the good capability of the PLS method to predict surface relaxivity. The root-mean-square error of prediction (RMSEP) for minerals and elements is 0.01 μm/s and 0.05 μm/s. The effect of minerals and elements on surface relaxivity is examined by the PLS regression coefficients. Figure shows the PLS regression coefficients based on minerals and elements. The absolute value of the PLS regression coefficient represents the strength of the correlation between the variable (minerals or elements in this study) and the surface relaxivity. A positive regression coefficient indicates that the mineral or element is positively correlated with surface relaxivity, while a negative regression coefficient illustrates the decrease in surface relaxivity. Predicted surface relaxivity vs measured surface relaxivity: (a) mineralogy and (b) chemical element. PLS regression coefficients for surface relaxivity: (a) mineralogy and (b) chemical element. Figure shows that the main elements negatively correlated with surface relaxivity are silicon, oxygen, calcium, and sulfur. The effects of silicon, oxygen, and calcium on surface relaxivity are in line with the theory. The surface relaxivity decreases as the amount of SiO2 increases relative to the paramagnetic substance, manifested as a negative regression coefficient for quartz. Similar results have been obtained through the experiment. Calcite (CaCO3) increases with the reduction of surface relaxivity, which is consistent with the reported results. Thus, calcium is negatively correlated with surface relaxivity. Sulfur should all come from pyrite (FeS2) according to the mineralogy composition data. Pyrite generally enhances surface relaxation. However, sulfur exhibits a negative correlation with surface relaxivity, suggesting the possibility of other trace sulfur minerals. Chemical elements positively correlated with surface relaxivity are carbon, sodium, magnesium, aluminum, potassium, titanium, and iron. Carbon was not analyzed due to the low sensitivity of the EDS technique for low-order elements. Sodium mainly occurs in plagioclase. However, many types of plagioclases may appear, such as anorthite and oligoclase. In addition, the magnetic susceptibility of plagioclase is weak, and its content is lower in this study. Thus, it is difficult to identify the effect of sodium according to the regression coefficient alone. Magnesium mainly comes from dolomite and chlorite. The magnetic susceptibility of chlorite is much higher than that of dolomite. Even if the content of chlorite is not high, it still causes to enhance surface relaxation, resulting in a higher surface relaxivity. Mineralogical composition analysis shows that aluminum exists in clay minerals and feldspars. The clay content is much greater than the feldspar content. The main clay minerals in these samples are illite, montmorillonite (Al4(Si4O10)2(OH)4•n H2O) and chlorite. Thus, a positive correlation appears between aluminum and potassium and the surface relaxivity. The oxidation state of an element has a significant influence on the surface relaxivity. The oxidation state of the elements can determine the magnetic susceptibility of minerals. Therefore, magnetic susceptibility calculations are performed using the mineralogical composition. Generally, the mineralogical composition is divided into two categories: paramagnetic and diamagnetic minerals. For instance, the surface relaxivity increases with the growth of paramagnetic minerals (e.g., clay and pyrite). Table presents the magnetic properties and magnetic susceptibility of minerals in this study. Obviously, the total magnetic susceptibility of each sample can be estimated based on the known mineralogical composition and its content. Figure shows that the surface relaxivity is positively correlated with magnetic susceptibility. Interestingly, magnetic susceptibility seems to determine the surface relaxivity. A larger surface relaxivity is observed in samples with higher magnetic susceptibility. The surface relaxivity of Middle Bakken and Three Forks formations are measured by the nitrogen adsorption method. Cheng et al. used the spin-echo method to measure the surface relaxivity of shale. The different fit coefficients may be attributed to the measurement method of surface relaxivity and the number of samples. Surface relaxation generally depends on the magnetic substances of the shale pore surface, which is associated with the magnetic susceptibility. Compared to other properties, the magnetic susceptibility of formations is easier to obtain through well logging. Correlation between the surface relaxivity and susceptibility. 4Magnetism and Magnetic Susceptibility of MineralogyMineralMagnetismDensityMass susceptibilityVolume susceptibility g/cm3 SI 10–6 m3/kgSI 10–6 Pyriteparamagnetic5.0100.31503Chloriteparamagnetic3.2000.0219 ∼ 0.484370 ∼ 1550Illiteparamagnetic2.7500.15412.5smectiteparamagnetic2.3500.14329Quartzdiamagnetic2.650–0.0062–16.42Potassium feldspardiamagnetic2.560–0.0058–14.85Plagioclasediamagnetic2.670–0.0041∼-0.0055–10.95∼-14.69Dolomitediamagnetic2.840–0.0134–38Calcitediamagnetic2.710–0.0051–13.8 Magnetism and Magnetic Susceptibility of Mineralogy To further demonstrate the relationship between surface relaxivity and magnetic susceptibility, FID measurements were conducted. Generally, the approximate upper limit of the internal gradient induced by shale samples in FID measurements can be written as 5 G≈(γD)1/2·(χB0)3/2 Combining eqs and , the dephasing length can be rewritten as 6 l2≈DγχB0 Obviously, the square of the dephasing length is inversely proportional to magnetic susceptibility. However, the dephasing length has no significant difference in FID curves. Thus, FID curves are transformed into frequency spectrum as shown in Figure , which is easy to give information about the dephasing length. The long dephasing length leads to the narrow width of frequency spectra, which can be obtained through Lorentzian model fitting. To highlight the curve characteristics, the frequency domain results only selected the range from −5000 to 5000 Hz. The shoulders can be attributed to the influence of noise, including impurities producing magnetic signals in the shale and noise from the low-field nuclear magnetic resonance instrument. Here, the width of the frequency spectra is replaced by parameter a, as shown in Figure b. This means the dephasing length is negatively related to parameter a. Assuming that the density of the shale sample is uniform, the magnetic susceptibility per unit volume is represented by parameter b. The related parameters and sample mass are given in Table . FID measurement: (a) FID signal decay curves for sample S1 saturated with water and (b) the transformed frequency spectrum. 5Magnetism and Magnetic Susceptibility of Mineralogy Mass a b Surface relaxivitySample(g)(Hz)(Hz/g)μm/sS131.2852718.682.59S233.9633259.582.36S322.33935716.021.67S430.77752016.921.46S511.71928023.891.34 Magnetism and Magnetic Susceptibility of Mineralogy According to the eq , the correlation between parameter b and the magnetic susceptibility constant χ can be expressed by 7 b−2∝χ Figure shows the surface relaxivity increases with b –2, and the R 2 = 0.98 illustrates good linear relationship. The measured results in Figure verify the evaluation results in Figure . All results reveal that the surface relaxivity depends on magnetic susceptibility. Thus, compared to elemental and mineral composition analysis, the third predicted method is easily established based on magnetic susceptibility. Correlation between the surface relaxivity and parameter b –2. Here are some limitations and improvements for the predicted method of surface relaxivity. First, the results depend on the input data of the samples: chemical elemental or mineralogy data and surface relaxivities. Thus, the measurement of surface relaxivity has a strong influence. Currently, conventional methods, such as NMR combined with pore size from the mercury intrusion method or S/V from the nitrogen adsorption method, are argumentative in obtaining surface relaxivity. The mercury intrusion method evaluates pore throats rather than pores measured by NMR. The nitrogen adsorption method requires dried samples, while NMR is suitable for fluid-saturated samples. For water-sensitive rocks, the surface relaxivity measurement has caused great controversy. Chemical elemental, mineralogy, and magnetic susceptibility logging data also lead to biased predictions. For instance, chemical elemental analysis is not sensitive to low-order atoms. Second, the prediction of the surface relaxivity also has requirements for the sample. Most importantly, the sample should be capable of NMR measurements to characterize the pore-size distribution. The relaxation of the fluid in the pores should satisfy the fast diffusion condition; otherwise, the relaxation time cannot characterize the pore structure. For instance, the pore distribution of carbonates generally cannot be measured by NMR due to their peculiar pore structure. Robinson et al. conducted experiments with cement and expressed caution about the relaxation time of the pore size. In addition, organic matter in some shale also contributes to the relaxation during the measurement. , Conclusions This paper deals with the prediction of the surface relaxivity of shales. The PLS method is used to predict the surface relaxivity with root-mean-square errors of 0.01 μm/s based on minerals and 0.05 μm/s based on chemical elements. Examining the regression coefficients can identify elements and minerals that affect surface relaxation values. The results show that surface relaxivity has a good linear relationship with magnetic susceptibility, which seems to be the intrinsic cause of both effects on the surface relaxivity. This method can be applied to evaluate the surface relaxivity in a wide range of porous materials and explore the mechanism in surface relaxation.