3.2. Effect of Fe Substitution on the Structure of α-Ni(OH)2 Nanosheets
Powder XRD was obtained to characterize the crystal structure and domain sizes of the materials and determine the effect of Fe substitution. The XRD pattern of the Fe-free Ni
1.00Fe
0.00 material (
Figure 3a) corresponds to a hydrated α-phase Ni(OH)
2 (ICDD# 022-0444) with three prominent peaks corresponding to the (001), (110), and (300) planes of α-Ni(OH)
2 and two minor peaks corresponding to the (002) and (111) planes [
41]. Within the crystal structure of α-Ni(OH)
2, the (001) diffraction represents nickel hydroxide layers organized along the c-axis of the unit cell perpendicular to the a/b-axes; the (002) planes correspond to half the distance between (001) planes. The prominence of the (001) and (002) peaks suggests that nanocrystals are oriented along the c-axis. The (110) planes are perpendicular to the (001) planes and arise from ordering of nickel and hydroxide atoms within the layers [
1].
The experimental XRD patterns were fit to determine crystallographic lattice parameters. α-Ni(OH)
2 is a hexagonal crystal structure with trigonal symmetry between the bond angles (α = β = 90°, and γ = 120°). The lattice parameters of the α-phase were determined for the unsubstituted and Fe-substituted samples. The c-lattice parameter determined from XRD analysis for unsubstituted α-Ni(OH)
2, Ni
1.00Fe
0.00, is 7.35 Å, and the a and b parameters are 5.37 Å (a = b within the unit cell) as shown in (
Table 1), which are consistent with values reported from Bode et al. for hexagonal α-Ni(OH)
2 [
41]. Variations between the interlayer d-spacing and the crystallographic constant c are due to the difference in calculating these parameters. The interlayer c-spacing was calculated using the Bragg equation for single point, 2θ
max, while the crystallographic constants a, b, and c were determined by fitting experimental data to a known diffraction pattern (ICDD# 022-0444) using a split pseudo Voigt function across all available Bragg-diffraction angles [
42]. The a, b, and c-parameters were within experimental accuracy of the analysis for Fe concentrations between 5–20%, indicating that iron incorporation into the α-Ni(OH)
2 occurred up to 20%. Fe concentration up to 20% did not distort the native crystal structure beyond the initial changes observed in Ni
0.95Fe
0.05. The changes in the lattice parameters with Fe substitution up to 20% are consistent with Fe substitution within the nickel hydroxide layers. The ionic radius of a 6-fold coordinated Fe
3+ (high spin) is 0.645 Å and is in a similar range as that of Ni
2+ (0.69 Å) [
43], and different interlayer interactions and structural disorder may also influence electron density and bond distances of atoms within the layers. We note the ionic radius of Fe
2+ (high spin) has a larger ionic radius of 0.78 Å.
When the reaction was substituted with 50% Fe in Ni
0.50Fe
0.50, there are several new diffraction planes that appear at 35.7°, 57.4°, 61.4°, and 63.1° 2θ (
Figure 3a), indicating that a second phase in addition to the α-Ni(OH)
2 phase is present. Based on the diffraction patterns, the secondary phase in the Ni
0.50Fe
0.50 materials is attributed to nickel ferrite (NiFe
2O
4; ICDD# 066-0778); however, further analysis is needed to determine the composition. We focused further analysis on the materials with up to 20% Fe substitution since within these samples the XRD patterns support that Fe is substituted within the α-Ni(OH)
2 lattice.
Fe substitution increases the interlayer (001) d-spacing from 7.43 Å in Ni
01.0Fe
0.00 up to 7.87 Å in Ni
0.90Fe
0.10 and 7.79 Å in Ni
0.80Fe
0.20 (
Figure 3b,
Table 1). The interlayer spacing of α-Ni(OH)
2 is affected by the reaction conditions; residual anions and molecules from solution phase reaction that are often incorporated into the structure [
5,
11]. The increase in the interlayer (001) d-spacing with an increase in Fe substitution is attributed to the increased concentration of SO
42− (as supported by EDS analysis discussed above) in the interlayer. The SO
42− anion has been shown to increase the interlayer (001) d-spacing of Al-substituted α-Ni(OH)
2 relative to samples containing NO
3− anions synthesized from either a metal-sulfate or metal-nitrate precursor [
44]. In contrast to the (001) diffraction plane, Fe substitution does not affect the d-spacing of the (110) planes to the same extent (
Table 1).
Additionally, the diffraction line profile of (001) reflections becomes broader and asymmetric. The (001) reflections arise only from layers of the structure normal to the c-axis. This lineshape is due to randomization of layers relative to one another along the c-axis. So as the Fe percentage increases, the nanosheets become more disordered and more entangled. This is consistent with the SEM images (
Figure 1) which show that not only do the nanosheet lateral dimensions decrease with Fe substitution, but they also become more disordered and “entangled” or more “interconnected”.
In addition to the (001) peak shifts with Fe concentration, the (001) peaks become broader as the Fe concentration increases. From Scherrer analysis of the XRD peak widths, unsubstituted α-Ni(OH)
2 exhibits a crystalline size of 14.3 nm for nanocrystal grains within in the (110) planes, notated as D
in-plane (in-plane of the nanosheets) and a crystalline size of 8.3 nm for domains defined by (001) planes, notated as D
through-plane (though the plane thickness of the nanosheets), as presented in
Table 1. The crystallite domain size determined from the (001) planes, D
through-plane, corresponds to the thickness of multiple stacked layers in the c-axis direction, and the distance determined from the (110) planes, D
in-plane, corresponds to the size of the crystalline domains within the plane of the nanosheets. Fe substitution decreases D
through-plane from 8.3 nm for Ni
1.00Fe
0.00 to 2.3 nm for Ni
0.80Fe
0.20. D
in-plane, the lateral size of the (110)-oriented nanocrystals, initially decreases from 14.3 nm for Ni
1.00Fe
0.00 to 9.1 nm for Ni
0.90Fe
0.10 and to 9.9 nm for Ni
0.80Fe
0.20. Further study is needed to determine the specific reason for smaller (110)-oriented nanocrystals (D
in-plane) with Fe substitution; however, we consider that in addition to the effect of Fe the different anion (sulfate) may also influence directional growth.
As the (001) planes are perpendicular to the (110) planes, we calculated the ratio of the (110) to (001)-oriented nanocrystals, R(D
in-plane/D
through-plane) (
Table 1). Note that this ratio of the nanocrystal grain sizes R does not depend on the choice of Scherrer constant. As Fe substitution increases, R(D
in-plane/D
through-plane) increases from 1.7 to 4.8. The high values of R(D
in-plane/D
through-plane) indicate distinctly different in-plane and through-plane nanocrystal sizes (shape anisotropy) and are consistent with the morphological structure of the material observed by SEM images (
Figure 1) and TEM images discussed below.
We used bright-field transmission electron microscopy (BF-TEM) to probe the structure of unsubstituted α-Ni(OH)
2 (Ni
1.00Fe
0.00) and the 5% Fe-substituted (Ni
0.95Fe
0.05) materials (
Figure 4). The lattice fringes of Ni
1.00Fe
0.00 (
Figure 4c) obtained from the planar edge region of the nanosheet (relative to the electron beam; boxed in red in
Figure 4b) show that an individual nanosheet is comprised of multiple nanocrystals (or crystallites) with lattice spacings of 2.69–2.72 Å, consistent with d-spacing of (110) planes of α-Ni(OH)
2 (2.68 Å) from XRD analysis (
Table 1). The nanosheets observed within the BF-TEM images have different darkness levels, which are influenced by the nanosheet thickness and aggregation. Unsubstituted α-Ni(OH)
2 nanosheets synthesized using a similar prep comprise two to three individual Ni(OH)
2 octahedra layers with an average thickness of 1.52 nm [
45], while Al-substituted Ni(OH)
2 nanosheets comprise stacks of 6–40 individual layers [
11]. TEM images (
Figure 4) show that the Ni
1.00Fe
0.00 and Ni
0.95Fe
0.05 are composed of nanosheets with large lateral dimensions which is consistent with XRD analysis that indicates the nanosheets organize and increase in length along the (110) plane relative to the (001) plane. For unsubstituted α-Ni(OH)
2 (Ni
01.0Fe
0.00), the lateral size of the crystallites, D
in-plane, (14.3 nm) determined from XRD analysis is 2–3 orders of magnitude smaller than the lateral dimensions of the nanosheets (~1000 nm) observed by electron microscopy; the nanosheets consist of multiple crystallites oriented without a preferred orientation within the nanosheet plate rather than a nanosheet being a single crystal. From the TEM images, as Fe is substituted for Ni, layers become more closely spaced and create more “ridges”. The topography of the 0% Fe sample shown on
Figure 4c is a lot flatter than the topography of the 5% Fe sample shown in
Figure 4f. These surface features can create local crystal anisotropy and can potentially increase the magnetic energy barriers for spins.
3.3. Magnetic Characterization of Fe-Substituted α-Ni(OH)2 Nanosheets
The temperature dependence of the ZFC and FC magnetization curves for all samples measured with a 100 Oe magnetic field are plotted in
Figure 5a and 5b, respectively. Qualitatively, all samples behave similarly. The ZFC magnetization is positive and rises sharply at temperatures below 20 K; it has a maximum at a characteristic temperature, T
max, and becomes negative as the temperature is lowered (
Figure 5a). The FC magnetization increases sharply below 20 K and saturates below 10 K (
Figure 5b). The irreversibility between the ZFC and FC measurements of magnetization originates from the interplay between the thermal energy and the energy barrier that hinders relaxation towards of the magnetization equilibrium and has been previously reported for nickel hydroxides [
20]. A peak or maximum in the ZFC magnetization is obtained when the measuring magnetic field is lower than the coercive field (H
c) and the coercive field is strongly temperature dependent [
46]. As we will show later, the coercivity at 10 K for the 5% and 10% Fe samples is 165 Oe and 164 Oe respectively, while the coercivity for the 0% Fe and 20% Fe is 6 Oe and 100 Oe respectively. So, the coincidence of the ZFC peaks at 10 K for the 5% and 10% Fe samples is directly related to the coincidence of the coercivity values. At temperatures above 20 K, the ZFC and FC magnetization curves coincide for all samples.
The inverse susceptibility (
) plots for all samples calculated from the ZFC magnetization in 100 Oe are shown in
Figure 5c. We analyze the inverse susceptibility for a magnetic field of 100 Oe to ensure that there are no artifacts due to magnetic flux trapped in the superconducting magnet or contributions from the capsule above 100 K where the sample magnetization is lower. The ZFC magnetization with the 100 Oe applied field can be considered as the net spontaneous magnetization. It is clear that there are two different kinds of temperature dependence. At high temperatures
increases linearly with temperature for all samples but is much larger for the 0% and 5% Fe samples (
Figure 5c). Below 50 K, we need to examine the
behavior for each sample separately. The magnetic susceptibility
of paramagnetic materials follows the Curie–Weiss law
above a critical temperature T
c.
C is the Curie constant and extrapolation of
vs.
T to a straight line cuts the temperature axis at
, the paramagnetic Curie point;
for ferromagnets and
for antiferromagnets or ferrimagnets. Below
, there is a long-range spin order even in the absence of an applied field so a spontaneous magnetization appears [
47].
Figure 6a shows
vs.
T for the sample with 0% Fe. The curve of
vs.
T is a straight line at high temperatures but becomes concave upward below ~70 K [
47]. Linear fits of
in the temperature range of 50–250 K give
. Since
is positive, this sample becomes ferromagnetic below ~30 K. The behavior of the sample with 5% Fe (
Figure 6b) is similar to the 0% Fe, and we extract
from the high-temperature linear fits. The Curie constant
C from the high-temperature fit of the inverse susceptibility is
for the Fe 0% sample and
for the 5% Fe sample. The uncertainty values in
C come from the exact temperature range we use for the linear fit. The Curie constant value that we derive here for the 0% Fe sample is in excellent agreement with the Curie constant previously reported by Rall and Seehra [
20].
The behavior of
with temperature is very different for the samples with 10% and 20% Fe shown in
Figure 6c,d, respectively. For temperatures above 125 K,
varies linearly with temperature according to the Curie–Weiss law. Linear extrapolations (we chose the 125–250 K range for consistency) give
and
for the 10% and 20% Fe samples, respectively. Since
is negative, they are either antiferromagnets or ferrimagnets. The Curie constant,
C, from the high-temperature fit of the inverse susceptibility, is
for the 10% Fe sample and
for the 20% Fe sample. Below ~100 K,
decreases rapidly with temperature and becomes concave downward over a large temperature range; it has a hyperbolic character. The Curie–Weiss law is obeyed only for temperatures well above the critical temperature (above 125 K), as discussed earlier. This
behavior is in sharp contrast with the ferromagnetic or antiferromagnetic cases, and for a simple crystalline system this behavior of
is clearly associated with ferrimagnetic ordering, indicating that the 10 and 20% Fe samples are ferrimagnetic [
48].
The inverse susceptibility
intersects the temperature axis at a critical temperature T
FN. We determine T
FN = 16 K and T
FN = 15 K for the 10% and 20% Fe samples, respectively, by a linear fit of
versus temperatures near the critical point (20–40 K) and finding the temperature for which
. In general, a ferrimagnetic system has two or more interacting sublattices, A and B, and the lattice of A ions is spontaneously magnetized in one direction, while the lattice of B ions is magnetized in the opposite or a different direction from the A direction. In contrast to an antiferromagnet, the magnitudes of the A and B sublattice magnetizations in a ferrimagnet are not equal, and a net spontaneous magnetization results. A and B can represent either different atomic species or the same ion types on sites of different symmetry. Although the interaction strengths within each lattice are not equal, both sublattices must have the same Curie temperature according to the Weiss theory of ferrimagnetism. Within this model, the net spontaneous magnetization can be zero at a temperature where the magnetizations of the two lattices are exactly balanced. This temperature is called the compensation point or compensation temperature,
To, and represents the temperature at which the opposing sublattice magnetizations are exactly balanced [
49]. This behavior is often called “magnetization reversal”. Although competing interactions between sublattices are necessary, they are inadequate by themselves in causing magnetization reversal; without any magnetic anisotropy, the magnetic moments would rotate to align with the magnetic field below the compensation temperature when the magnetization of one of the sublattices becomes dominant. So, in addition to competing interactions such as negative exchange coupling between two or more magnetization lattices or interfacial exchange coupling between FM and AFM lattices, high magnetic anisotropy is essential in observing magnetization reversal [
49]. For lower magnetic fields, the Zeeman energy is lower than the anisotropy energy, and the reversal of the net magnetization in the applied field direction will be prohibited by the magnetic anisotropy of the system.
The competing exchange couplings assumed in these samples are related to the substitution of Fe for Ni. The interlayer distance for the samples studied (see
Table 1) is constant for the investigated compositional range; the differences in the magnetic behavior can be correlated with variations in the chemical composition. For the two reported ferrimagnetic samples, the relative Ni-to-Fe concentrations in the powder are x = 0.1 and x = 0.2 (
Table 1). If we assume that the metal cations are statistically distributed across the layers with this average Ni to Fe ratio (Ni:Fe, 9:1, 8:2), then each Fe
3+ ion will be on the average surrounded by six Ni
2+ ions (Ni−OH−Fe pairs), whereas each Ni
2+ atom will be neighbored by at most one Fe
3+ and at least five Ni
2+ ions (Ni−OH−Ni and Ni−OH−Fe pairs). So, the competing exchange couplings assumed in these samples must be between Ni−OH−Fe pairs and Ni−OH−Ni pairs.
From the analysis of the temperature dependence of the magnetization, we conclude that α-Ni(OH)
2 is ferromagnetic and progressively becomes ferrimagnetic as Fe (up to 20%) is substituted for Ni. We further measured the FC and ZFC magnetization under higher applied fields. The FC and ZFC magnetization curves for applied fields of 0.005 T, 0.01 T, 0.05 T, 0.1 T, and 0.5 T are shown in
Figure 7. The ZFC magnetization curves measured under high magnetic fields as above do not represent the spontaneous magnetization anymore; they rather represent the ability of the system to respond to an external magnetic field at each temperature. The FC curve now represents the “frozen” total magnetic moment of the system. The FC and ZFC susceptibilities for all samples follow two interesting trends that we describe next.
First, the ZFC and FC curves coincide at high temperatures but separate below an “irreversibility” temperature, T
ir, which is dependent on the applied magnetic field. Below T
ir, the FC susceptibility increases as the temperature is lowered, has a plateau and saturates
. The ZFC susceptibility increases as the temperature is lowered, has a maximum at T
max and then decreases as the temperature is further lowered. For high (enough) magnetic fields, the FC and ZFC curves coincide. The irreversibility between the ZFC and FC measurements of magnetization originates from the interplay between the thermal energy and an energy barrier that hinders relaxation towards of the magnetization equilibrium and such irreversibility has been previously reported for nickel hydroxides [
20]. When a large magnetic field is applied, the FC and ZFC magnetizations can coincide. In that case, the system is in a state where interactions between spins are overpowered by the interaction of the moments with the external magnetic field, and the magnetic moments are oriented along the external magnetic field. The fact that the susceptibility of an antiferromagnet or ferrimagnet below the critical temperature depends on the size of the field used to measure it is due to crystal anisotropy. The second feature of the magnetization behavior as a function of temperature (
Figure 7) is that while M
FC is always positive, M
ZFC becomes negative below the compensation temperature,
To, which depends on the applied magnetic field.
The compensation temperature is plotted on
Figure 8a as a function of the applied magnetic field for all samples. As the magnetic field increases,
To is suppressed, and the ZFC magnetization becomes progressively “less negative”. For 0% Fe, an applied magnetic field of just below 200 Oe is sufficient to suppress the negative ZFC magnetization (
Figure 8a). For the 5, 10, and 20% Fe, the negative ZFC magnetization is not suppressed at low temperatures even under 1000 Oe, but
To decreases as the applied field increases. This is also the case for 10% and 20% Fe. Note that for a 0.5 T magnetic field, the ZFC magnetization does not go through zero anymore for any of the samples suggesting that the antiferromagnetic interactions between sublattices have been suppressed.
The ZFC magnetization at 2K is plotted on
Figure 8b as a function of the applied magnetic field. The ZFC Magnetization is negative for all samples for the 50 Oe and 100 Oe applied magnetic fields. The M
ZFC for the 0% Fe sample that shows ferromagnetic behavior has the smallest magnitude which is very close to zero. In contrast, the value of M
ZFC for the 5, 10 and 20% Fe samples has a larger magnitude and remains negative for up to a 1000 Oe applied magnetic field. Note that the magnitude of M
ZFC does not increase as a function of the magnetic field; rather, it is flat and increases sharply when a 5000 Oe magnetic field is applied which points to an intrinsic origin and not external factors such as a demagnetizing field. In our case, the negative magnetization can be understood as magnetic moments being pinned by magnetocrystalline anisotropy due to the multiple crystallographic domains. The pinning energy is higher than the energy imparted by the external magnetic field of 1000 Oe for the samples doped with Fe. Additionally, the M
ZFC at 5000 Oe is largest for the 0% Fe sample and decreases monotonically with Fe %, which reflects the effect of pinning of the moments; the magnetic anisotropy is smaller for the 0% Fe sample.
Figure 8c shows the temperature at which the ZFC magnetization has a maximum versus the applied magnetic field. As the applied magnetic field increases, the maximum of the ZFC magnetization decreases. In magnetic nanoparticles, the magnetic irreversibility is conventionally associated with the energy required for a particle to reorient its magnetic moment when a magnetic field is applied, overcoming a barrier due to shape, magnetoelasticity or crystalline anisotropy [
49]. Irreversibility between FC and ZFC also occurs in ferromagnetic and antiferromagnetic systems below the transition temperature and can originate from the magnetocrystalline anisotropy; it is more prominent in polycrystalline materials [
50].
Measurements of the sample magnetization at 2K as a function of the applied magnetic field H (M-H loops) are shown in
Figure 9. Hysteresis is observed for all samples. Notably, the initial magnetization at H = 0 is negative for the Fe substituted samples in agreement with the ZFC magnetization behavior discussed earlier (see
Figure 7). Measurements of each sample’s magnetization as a function of the applied magnetic field H at 4 K, 8 K, 10 K are shown in the insets of
Figure 9. Hysteresis is observed for all samples, and we extract the coercive field (or coercivity) (H
c), remnant magnetization (M
r), and saturation magnetization (M
s). The coercivity at 2 K clearly increases as the Fe content increases. While the M-H loop for 0% Fe shows typical ferromagnetic behavior at 2 K (
Figure 9a), the M-H loops for the 5%, 10%, and 20% Fe show some evidence of AF interactions (
Figure 9b–d): the loop has a more “square” shape, i.e., the remnant magnetization is close to the saturation magnetization. This also could point to a larger magnetocrystalline anisotropy for the higher Fe %.
Hc, Mr, and Ms are shown in
Figure 10 as a function of temperature. The coercive field (
Figure 10a) is highest at 2 K and decreases monotonically as the temperature increases for all samples. At 2 K, the coercivity is also highest for the sample with 20% Fe and lowest for the sample with 0% Fe as noted earlier. This trend is also followed at higher temperatures. This is consistent with the field required to suppress the negative M
ZFC at 2 K (
Figure 8b) and the field required to suppress the ZFC to FC irreversibility being higher as the Fe concentration increases. So, the field required to suppress the difference between the ZFC and FC susceptibilities is related to the magnitude and the temperature variation of the coercivity, H
C, which is a measure of the magnetic anisotropy.
Figure 10b shows that at 2K the remnant magnetization is highest for the 5% Fe and 10% Fe samples and lowest for the 0% Fe sample while the remnant magnetization for the 20% Fe sample is lower than the 5% and 10% Fe samples and is closer to the 0% Fe sample. This indicates that the magnetocrystalline anisotropy is stronger for the 5% and 10% Fe samples at 2 K. Finally, the saturation magnetization (taken at 9 T, full data is shown in
SI, Figure S7) shown in
Figure 10c is also decreasing as temperature decreases but is more sample-dependent than temperature-dependent.
This trend in saturation magnetization is clearer in
Figure 10d where we have plotted the saturation magnetization in Bohr magnetons per magnetic atom (Fe or Ni) as a function of the Fe %. The saturation magnetic moment we calculate for α-Ni(OH)
2 (0% Fe) is 1.8 μ
B and is in good agreement with a previous experimental report [
20]. The saturation magnetization decreases as Fe is substituted for Ni. This is consistent with the observed ferrimagnetic behavior and indicates that Fe and Ni form two crystal lattices which orient in opposite directions. The saturation magnetization decreases as the Fe concentration increases, which is consistent with the ferrimagnetic behavior of the high Fe concentration samples. The reduction in M
s can be explained by the presence of competing interaction between lattices and possibly the presence of a non-collinear spin structure below the critical temperature. The non-colinear spin structure is also apparent on the full H-range M vs. H loops (
Figure S7).
In the paramagnetic region, we can calculate the effective magnetization from the Curie constant
where
is Avogadro’s number,
is Boltzmann constant and
is the effective paramagnetic magnetic moment [
47]. The effective magnetic moment for 0% Fe is 2.9 µ
B which corresponds exactly to a high–spin state,
S, for Ni
2+ (
S = 1). The calculated effective moment for
S = 1 and spectroscopic splitting factor
g = 2 is 2.83 µ
B. This value is also in agreement with previous reports for α-Ni(OH)
2 [
20]. The effective magnetic moment only slightly increases from 2.9 µ
B for 0% Fe to 3 µ
B for 5% Fe. The magnetization for the 10 and 20% Fe samples are 7.1 µ
B and 7.3 µ
B, respectively (
Table 2). A substantial shift to higher effective magnetic moments occurs between 5 and 10% Fe, indicating that the effect of Fe substitution on the effective magnetic moment is not linear with % Fe. The effective magnetic moment is related to the spin and the spectroscopic splitting factor via the equation
[
47]. The effective paramagnetic magnetic moment in the 10 and 20% Fe samples corresponds to spin
S = 5/2, in line with the presence of high-spin Fe
3+ and perhaps an orbital contribution.
From our data and analysis, Fe substitution has a clear effect on the magnetization of α-Ni(OH)
2, and a summary of relevant magnetic parameters and ordering for the samples with different % Fe (up to 20%) is presented in
Table 2. The ferromagnetic behavior of α-Ni(OH)
2 and 5% Fe observed from our study is consistent with a prior study of α-Ni(OH)
2 [
20]. Compared to 5% Fe substitution, substitution at 10 and 20% Fe results in ferrimagnetism (rather than ferromagnetism), higher paramagnetic effective moments, and higher coercivity values. From our structural analysis, 10 and 20% Fe result in higher values of R(D
in-plane/D
through-plane). The correlation between coercivity and R(D
in-plane/D
through-plane) shown in
Figure 10e indicates that a higher ratio of in-plane/through-plane nanocrystalline grain size results in higher coercivity which is consistent with higher Fe % increasing the pinning of moments within the nanocrystallites. From prior studies of Ni(OH)
2 and related systems, the magnetism is affected by intraplanar exchange, interlayer exchange, and dipole interactions from interacting ligands [
20,
21,
51]. Morphology has also been shown to influence the magnetization of layered hydroxides [
18]. The oxidation state of Fe affects magnetization. In an aqueous solution, Fe
2+ can oxidize to Fe
3+ by reacting with dissolved O
2 [
52], the oxidation state of the Fe incorporated into the γ-NiFeOOH lattice (oxidized form of α-Ni(OH)
2) can also vary depending on the Ni:Fe substitution ratio [
26].
All the layered hydroxides of Ni obtained by intercalating different ligands order ferromagnetically with T
C around 16–18 K, even though there are large variations in the interplanar spacings. The appearance of ferromagnetism in α-Ni(OH)
2 as compared to the antiferromagnetism observed in β-Ni(OH)
2 has been attributed to the increase in the distance between Ni
2+ ions in adjacent nanosheets [
53]. In Ni-Fe-layered double hydroxides, the ferromagnetic critical temperature decreases with increasing interlayer separation [
54]. In our system, the transition from a ferromagnet to a ferrimagnet occurs by adding only 10% of Fe to the material. Such a dramatic change in the magnetic properties could be related to the low–dimensionality in the system, i.e., the effect of the surfaces and interfaces is stronger. A prior study of nickel iron hydroxides with 10 and 20% Fe substitution [
34] did not find ferrimagnetism and strong deviation from Curie–Weiss-type behavior at low temperatures as we observe within our 10 and 20% Fe materials. We note that this prior study used different precursors and synthetic conditions (coprecipitation under basic conditions) compared to that used within our study. The comparison of our result with this prior study shows that different magnetization occurs with the same Ni:Fe substitution ratios which demonstrates that structure and morphology play a critical role in the magnetization of nanostructured materials.
The sample with 50% Fe shows clearly different magnetization compared with the 0, 5, 10 and 20% Fe samples. The magnetic behavior of the sample with 50% Fe is shown in
Figure S8. The ZFC and FC magnetizations under 50 and 100 Oe magnetic fields do not overlap even at room temperature anymore, which indicates that the critical temperature for this sample is above room temperature. The different magnetic behavior of the material with 50% Fe compared to the materials with lower % Fe is attributed to the presence of nickel ferrite within the material as determined from XRD discussed earlier.