Possible mechanisms causing failure of Thellier palaeointensity experiments in some basalts

Abstract

The normally magnetized zone of the Jurassic Lesotho basalts, although providing apparently quite reliable palaeofield directions Kosterov & Perrin 1996), shows anomalous behaviour when studied in vacuum using the Thellier palaeointensity method: typically the slope of the natural remanent magnetization–thermoremanent magnetization (NRM–TRM) curves is very steep at intermediate temperatures (200 to 400–460°C). In order to elucidate the reasons for such an anomalous behaviour, six representative samples (from a total of 74 studied using this method) were subjected to a variety of analyses. These experiments indicate that the magnetic properties are dominated by pseudo-single-domain (PSD) magnetite grains some 1μm in size, resulting from high-temperature oxidation of titanomagnetite. Laboratory heatings in vacuum up to the Curie point do not change significantly the room-temperature hysteresis characteristics or the initial susceptibility k. Similarly, the k(T) curves in vacuum are (with a single exception) rather reproducible. Since the laboratory TRMs yield almost ideal NRM–TRM plots, the anomalous NRM–TRM plot is presumably due to some peculiarity of the natural TRM. The partial TRM (pTRM) acquisition capacity in the moderate temperature range (cooling from 200 to 20°C) is generally very strongly reduced after heating to 270°C, which indicates that some magnetic alteration has already occurred at these temperatures. Hysteresis measurements between room temperature and the Curie temperature T_c show that some small (less than 10 per cent) but significant irreversible changes in hysteresis characteristics also occur during heating. In particular, the coercive force H_c0 at room temperature is typically reduced after heating at a moderate temperature (175°C) but increases after treatments at 475°C and, more pronouncedly, at 580°C. The saturation magnetization J_s0 remains unchanged, except for a very small decrease (less than 5 per cent) occurring in some samples after the two latter treatments. These changes are most clearly seen on H_c(T)–J_s(T) bilogarithmic plots, which show that the moderate-temperature change in coercivity can extend up to 200–250°C. Thus hysteresis measurements as a function of temperature offer a promising tool for sample pre-selection for Thellier experiments. Alternating-field demagnetization and cycling of pTRMs at liquid-nitrogen temperature suggest that the blocking mechanism is largely multidomain-like near room temperature but becomes less so as the Curie point is approached. The main reason for the failure of the Thellier experiments is the loss of a fraction of the NRM (natural TRM) at temperatures apparently lower than the blocking temperatures in nature. It is suggested that this anomalous behaviour results from the reorganization of the domain structure of the PSD grains during heating. This transformation, which seems to be triggered by the coercivity decrease observed at very moderate temperatures, can reduce the NRM intensity without requiring any correlated pTRM acquisition.

1 Introduction

The strength of the Earth’s magnetic field in the historic and geological past is of considerable interest in geophysics. In their milestone paper, Thellier & Thellier (1959) outlined a method to determine the intensity of the ancient geomagnetic field from the characteristics of thermoremanent magnetization (TRM) produced during the cooling of natural volcanic rocks and archaeological artefacts. The Thellier method is still considered to be the most reliable of those proposed for this purpose. However, several conditions have to be obeyed to ensure the significance of the palaeointensity results:

(1) the primary remanent magnetization must be a TRM;

(2) the secondary components must be weak with respect to the primary component and must be removed at relatively low temperatures;

(3) the remanence carrier must be reasonably stable during heating in the laboratory;

(4) the independence and memory laws of the partial TRMs (Thellier 1938) have to be valid.

Condition (2) is easy to check and is fulfilled by a significant fraction of volcanic rocks. The Thellier method provides the most reliable way of checking the stability of the TRM (condition 3), but this condition is, however, only exceptionally observed to hold over the whole temperature range of the investigation. Condition (4) holds true for single-domain (SD) grains only, not for large grains Shashkanov & Metallova 1972). The magnetic carrier in the volcanic rocks selected for palaeointensity experiments is commonly a near-magnetite spinel formed from spinodal decomposition of a former Ti-rich titanomagnetite. The grain size of this spinel phase is generally larger than the single-domain/pseudo-single-domain (SD/PSD) threshold, which makes it possible that a significant number of volcanic rocks do not rigorously meet condition (4). It is commonly believed (Haggerty 1976) that this transformation occurs above 600°C. Thus, the primary remanence should be a TRM (condition 1).

The present work is an effort to understand the reason for the unusual thermal behaviour of the NRM of some of the early Jurassic Lesotho basalts. These rocks seem to be almost ideal recorders of palaeomagnetic field directions Kosterov & Perrin 1996). However, many of these lava flows yield quite unacceptable results when studied with the Thellier method. The objective of the work reported here was to investigate, through a variety of rock-magnetic studies, the reasons for this anomalous behaviour.

2 Palaeomagnetism, Opaque Mineralogy And Rock-Magnetic Characteristics Of Samples

2.1 Palaeomagnetism

The samples used in the present study were tholeiitic basalts from the Stormberg Formation, cropping out in Lesotho. The Lesotho basalts present a quite simple magnetostratigraphy with, from bottom to top, a reversed zone, a transitional zone and a well-developed normal zone (Van Zijl, Graham & Hales 1962; Marsh et al. 1997). The localization and palaeomagnetic characteristics of our entire collection of normal and reversed flows were described by Kosterov & Perrin (1996). We found that only a small fraction of these flows, all of reversed polarity, provided apparently reliable palaeointensity data (Kosterov et al. 1997). In contrast, all the flows from the normally magnetized Mafika–Lisui Pass section Kosterov & Perrin 1996) show, to various extents, a peculiar type of non-ideal behaviour in Thellier experiments. From a total of 74 samples from this section, studied using the Thellier method, we selected six representative samples for the present investigation.

Initially, in order to assess the samples’ suitability for the Thellier experiments, the following set of experiments was carried out. Viscosity indices Thellier & Thellier 1944; Prévot 1981) were determined from a two-week storage in the ambient field followed by another two-week storage in zero field. Palaeomagnetic directions for each sample were determined using either thermal (heating in air) or alternating-field (a.f.) demagnetization.

The determination of the characteristic-remanence (ChRM) directions from both thermal and a.f. cleaning (Fig. 1) was straightforward. The ChRM of the selected samples was easily isolated in the range 10–150mT (a.f. treatment) or 200–600°C (thermal cleaning), with mean angular deviations of less than 1° (Table 1). For all the samples from a single flow, the ChRM directions are very close to the mean. The secondary components were always weak compared to the primary component and yielded scattered directions that always differed from the direction of the present-day geomagnetic field at the sampling locality. The viscosity indices did not exceed 3 per cent, confirming that the viscous remanence is almost negligible.

2.2 Magnetic mineralogy

In order to identify the magnetic minerals, polished thin sections were examined with an optical microscope, and the temperature dependences of the initial magnetic susceptibility k and saturation magnetization above room temperature were measured under vacuum or in helium, respectively. In addition, low-temperature measurements of SIRM (saturation isothermal remanent magnetization) were carried out at the IRM (Institute for Rock Magnetism), Minneapolis, using a Quantum Design magnetic properties measurement system (MPMS).

Microscopic observations of polished thin sections indicated that in all samples the opaque mineral was mainly titanomagnetite, transformed by spinodal decomposition into Ti-poor titanomagnetite remnants delimited by exsolutions along (111) planes. This suggests that the dominant (and sometimes exclusive) magnetic mineral is near magnetite in composition. With only few exceptions, the ‘magnetite’ crystals were not affected by any subsequent lower-temperature alteration such as granulation. Similar observations hold true for the entire Lesotho basalt formation (Kosterov et al. 1997; Prévot et al. in preparation). The size of the original titanomagnetite crystals varies greatly from sample to sample. Samples 21 and 86 (the latter sample was drilled only 10cm above the bottom contact of a 3–6m thick flow) show elongated, skeletal titanomagnetite crystals, sometimes reaching several tens of microns in length, but not more than a few microns wide in their widest sections. In contrast, samples 47 and 244 (both drilled from flow interiors) contain euhedral crystals reaching a few hundreds of microns in size. The titanomagnetite crystals of sample 117 have intermediate characteristics. They are typically several tens of microns in size and vary from anhedral to euhedral shapes.

The exsolutions are ilmenite or, less commonly, metailmenite. In samples 21 and 86, the ilmenite is replaced by unidentified non-opaque phases. Titanohaematite exsolutions are sometimes present as a minor constituent (samples 117 and 244). No pseudo-brookite exsolutions were observed. Thus we have no specific evidence that the spinodal decomposition during cooling in nature stopped above the Curie temperature of magnetite.

In contrast to the titanomagnetite crystals, the size of the magnetite crystals resulting from spinodal decomposition seems rather constant. Their typical width varies from 0.5μm to 1μm from sample to sample. Their shape seems more equant when the magnetite was formed from skeletal crystals rather than from large, euhedral crystals. In the latter case, the length-to-width ratio of these magnetite rods can reach several units Davis & Evans 1976). These dimensions and shapes correspond to small PSD grains.

The temperature dependences of the initial susceptibility [k(T) curve] and the saturation magnetization [J_s(T) curve] of three representative samples are shown in Fig. 2. Sample 244 and, to a lesser extent, sample 47 (not shown) yield reversible curves with a single Curie point, indicating the presence of magnetite. In contrast, sample 86 yields an irreversible k(T) curve, with two inflection temperatures visible on both the heating and the cooling curves. This suggests, in agreement with the J_s(T) curve, that in addition to magnetite, a second spinel phase with a Curie point near 400°C is present. This latter phase, which is rather unstable upon heating, is presumably titanomagnetite (or titanomaghemite). Magnetite is the dominant phase in the other three samples. Sample 199 (and similarly samples 21 and 117) exhibits some indication [in the k(T) heating curve] that this second phase is also present here, but in a much smaller proportion than in sample 86.

In order to investigate further the magnetic mineralogy of the samples, we measured, in zero field, the thermal decay of the saturation remanence (SIRM) acquired at 15K in a d.c. magnetic field of 2.5T. All curves are characterized by a major SIRM loss between 90 and 130K (Fig. 3a). An examination of the derivatives of the SIRM_15K(T) curves (Fig. 3b) shows that the transition temperature, if defined as the temperature at which the rate of the SIRM decay is maximal, varies from 115K for sample 199 to 100K for samples 021 and 086, which exhibit less pronounced peaks. These temperatures are somewhat lower than the temperature of 121–122K reported for the Verwey transition of stoichiometric magnetite, indicating some degree of non-stoichiometry due either to vacancies or to substitution of minor elements for iron (Aragón et al. 1985; Kakol & Honig1989a; Aragón 1992). The broadness of the transition zone probably reflects some variance in chemical composition from one crystal to the other. An important conclusion of these experiments is that near-magnetite is present as a primary phase in all our samples.

The thermal demagnetization of SIRM_15K should also provide some indications about the grain size of the near-magnetite phase. For stoichiometric magnetite in the range 37–220nm, the fraction of SIRM surviving the warming-up to room temperature (SIRM memory) has been found to be grain-size-dependent (Özdemir, Dunlop & Moskowitz 1993). However, non-stoichiometry reduces or even suppresses the SIRM loss. For sample 199, which seems to contain almost-stoichiometric magnetite, the SIRM memory is nearly 40 per cent. According to the data of Özdemir et al. (1993), a very fine magnetic grain size (several tens of nanometres here) is expected. Qualitatively, a similar conclusion might be drawn for most of the other samples too. Compared to the microscopic observations and hysteresis characteristics (Section 2.3 below), this estimate seems too low by an order of magnitude, probably because the SIRM memory may depend also on other characteristics of the magnetic grains, such as the kind of anisotropy.

All five crystalline anisotropy constants of monoclinic magnetite are almost temperature-independent in the range 5–75K (Kakol & Honig1989b). Thus, any large SIRM decay in this temperature range is commonly attributed to the presence of superparamagnetic grains (e.g. Özdemir et al. 1993). According to Fig. ³3a, such grains represent a significant fraction of the whole magnetite population in all samples except sample 199.

2.3 Hysteresis characteristics

Measurements of hysteresis at room temperature (Table 2) were carried out using a laboratory-built translation inductometer at the Laboratoire de Géomagnétisme, Saint-Maur, France, in a maximum d.c. field equal to 800 mT. Sample 86 has the lowest coercive force (10.5 mT), samples 21, 47, 117 and 244 have an intermediate coercive force (16–18 mT), and that of sample 199 is 29.1 mT. The same tendency holds for the coercivity of the remanence, which is at its minimum (24.8 mT) for sample 86 and at its maximum (51.4 mT) for sample 199. This suggests that the phase with an intermediate Curie point, which is present in a considerable amount in sample 86, has a lower coercivity than the near-magnetite phase (see Fig. 2).

When plotted on a Day plot (Day, Fuller & Schmidt 1977) (Fig. 4), all samples fall into the pseudo-single-domain range, in agreement with the microscopic observations. It must be pointed out, however, that the SD/PSD limits shown in Fig. 4 refer to non-interacting particles, which is probably not the case here. As shown above, the magnetic grains are packed into host crystals which were formerly high-titanium titanomagnetite. Owing to the net demagnetizing field of the host grain as a whole, the J_rs/J_s ratio of the SD magnetite will be reduced from 0.5 to about 0.3, depending on the geometry of the magnetite grains Davis & Evans 1976). The average J_rs/J_s ratio of our samples being approximately 0.2 (Table 2), a PSD structure can be inferred from the hysteresis data, in good agreement with the microscopic observations. Rock samples having such hysteresis characteristics are commonly used in palaeointensity experiments (e.g. Prévot et al. 1985; Garnier et al. 1996). Note also that the hysteresis characteristics of the present samples are not significantly different from those of the samples from the Lesotho basalts which provide apparently reliable palaeointensity results (Kosterov et al. 1997).

3 Palaeointensity Experiments

3.1 Experimental procedure

Palaeointensity experiments were carried out using the Thellier method in its classic form Thellier & Thellier 1959), i.e. by double stepwise heatings in an arbitrarily positive/negative laboratory magnetic field, applied throughout the whole heating–cooling cycle. All heatings were performed in a vacuum better than 10⁻²mbar. The intensity of the applied field was 20.0±0.1μT. 17 or 18 temperature steps were used, in a range from 100 to 570–580°C. The temperature reproducibility between any two heatings to the same nominal temperature was normally within 2°C. In order to control for magnetic changes produced by heating, partial TRM (pTRM) checks were performed after the third heating step, and then after each subsequent step throughout the whole experiment. Also, the room-temperature magnetic susceptibility was measured after the second heating at each temperature step.

3.2 Thellier experiments on NRM

Examples of typical NRM–TRM plots, together with the corresponding orthogonal diagrams, are given in Fig. 5. Despite the low viscosity of the samples, the initial part of the NRM–TRM plots, corresponding to temperatures up to 200–250°C, is affected by a secondary magnetization. Thus, as usual, this part of the diagram cannot be used for palaeointensity determination. After this part of the plot, decrease of the NRM is observed, which is not matched by a corresponding pTRM acquisition. This decrease extends up to 400–460°C, depending on the sample. At this upper temperature, the NRM loss can be as large as nearly 50 per cent (Table 1 and Fig. 5). At higher temperatures, almost linear segments, extending at least up to 550°C and covering about 40 per cent of the total NRM intensity, are observed for most samples. Up to 522°C the pTRM checks are shifted to the left, which indicates a progressive decrease of pTRM acquisition capacity in the blocking-temperature range below 460°C, due to the successive heatings. For most samples, the last heating steps, in the vicinity of the Curie point, seem to be affected by an increase in pTRM acquisition capacity, as is commonly observed in most lavas.

None of these six samples yielded a convincing palaeointensity result. All the samples, except perhaps sample 244, would provide implausibly large palaeointensity estimates if the generally preferred, low temperature range (say 200 to 400–460°C) was used; in fact these strongly curved or even kinked NRM–TRM plots cannot be safely interpreted in terms of palaeointensity. Such an anomalous behaviour during the Thellier experiment could not have been foreseen, bearing in mind the rather encouraging palaeomagnetic and rock-magnetic properties of the samples.

3.3 Thellier experiments on laboratory TRM

Three samples (samples 47, 117 and 199) were selected for this study. All of them are characterized by a large, steep decrease in the NRM–TRM plot at intermediate temperatures (Fig. 5) and show only small changes in their hysteresis parameters after the heating experiments (see Figs 13 and 14 below). After the completion of the first Thellier experiment, a total TRM was created by cooling down from 600°C in a 20μT magnetic field. These laboratory TRMs were then subjected to a Thellier experiment under conditions similar to those of the first experiment, except that the laboratory field intensity was 30μT.

The second laboratory TRM (TRM₂) is equal to the first one (TRM₁) (once the scaling factor of 1.5 is taken into account) for sample 117, and is some 3 per cent higher for samples 47 and 199, revealing that only minor alteration continued during the second experiment. However, for all three samples the diagrams are slightly concave upwards (Fig. 6). It is interesting to estimate an apparent ‘palaeointensity’ from the ‘most linear’ segments of the TRM₁–TRM₂ diagrams, as is done commonly for natural TRM. If we take this approach, the results vary somewhat from sample to sample. For sample 47 the segment from 301 to 567°C yields the best ‘palaeointensity’ estimate of 19.2±0.2μT, only 0.2μT lower than the value obtained from the TRM₁/TRM₂ ratio. For sample 199, a fairly straight line can be fitted through all points except the last one, yielding an estimate of 19.2±0.1μT, again just 0.2μT lower than that obtained from the TRM₁/TRM₂ ratio. In contrast, the apparent palaeointensities from sample 117 may be as high as 25.2±0.5μT and as low as 18.5±0.2μT, depending on which points are used to fit a straight line. Thus, at most, the field palaeostrength can be overestimated by 25 per cent.

Despite the fact that we are inevitably dealing here with samples somewhat altered by previous heatings, the results of these Thellier experiments on artificial TRMs provide a fundamental constraint upon the interpretation of the non-ideal behaviour of these samples as observed during the NRM–TRM experiments: this behaviour is a property of the NRM, and is not observed for the laboratory TRM.

3.4 Unblocking versus blocking spectra

For the purpose of analysing our data, it is convenient to examine separately the blocking- and unblocking-temperature spectra. Fig. 7 compares, for samples 47, 117 and 199, the NRM unblocking spectrum (solid squares), the TRM blocking spectrum (open squares), and the unblocking and blocking spectra of TRM₁ (solid and open circles, respectively). The NRM and TRM unblocking spectra were normalized by the NRM and TRM intensities, respectively. All spectra have been reduced to values per one degree, in order to provide a common scale for the pTRM differences calculated from the variable temperature increments.

Examination of these spectra reveals that the NRM unblocking spectrum (solid squares in Fig. 7) is strongly dissimilar to the other spectra. Even for sample 244, which shows the best similarity of the spectra, there is still a small part between 250 and 330°C where NRM unblocking is more effective than pTRM blocking, resulting in a small but noticeable ‘kink’ in the NRM–TRM plot. Unlike the NRM, the unblocking curves of the laboratory-created TRM₁ closely follow the blocking curve. This emphasizes the uniqueness of the NRM as a specific magnetic state that cannot be replicated by laboratory TRM for our samples.

Three main hypotheses can be put forward to explain the failure of the Thellier experiments: (1) a difference between the blocking and unblocking temperatures inherent in the samples, due to the presence of multidomain grains (Levi 1977; Bol’shakov & Shcherbakova 1979; Worm et al. 1988; Shcherbakov, McClelland & Shcherbakova 1993; McClelland, Muxworthy & Thomas 1996; McClelland & Briden 1996); (2) destruction of parts of the remanence carriers or modification of their chemistry during heating in such a way that the low-T blocking temperatures are increased; and (3) irreversible physical changes during heating, resulting in the unblocking of the NRM at temperatures lower than the blocking temperature. The rock-magnetic experiments described below were carried out to try to identify the most plausible explanation.

4 Properties Of Partial Trms And Effect Of Heatings

4.1 Experimental procedure

After stepwise alternating-field demagnetization (up to 150 mT) of the NRM, an ‘initial’ pTRM was given to the samples by cooling from 200°C to room temperature in a 20μT magnetic field parallel to the Z direction. This pTRM [pTRM(200–20)] was then a.f. demagnetized in the same way as the NRM. A second ‘initial’ pTRM was given by cooling the samples from 270 to 200°C in a 20μT magnetic field parallel to the +X direction, followed by cooling to room temperature in zero field. After a.f. demagnetization of this pTRM [pTRM(270–200)], the first two-component control pTRM (pTRM1c) was acquired so that during cooling, the magnetic field was along the +X direction during the interval 270–200°C and then along the +Z direction during the interval 200–20°C (Fig. 8). This control pTRM was then a.f. demagnetized. After acquisition (and subsequent demagnetization) of ‘initial’ pTRMs in the intervals 330–270°C, 460–330°C and 520–460°C, a second two-component control pTRM (pTRM2c) was given during cooling in a 20μT magnetic field parallel to +X during the interval 520–460°C and then parallel to +Z during the interval 460–330°C, which was followed by cooling in zero field to room temperature. After acquisition and subsequent demagnetization of pTRMs in the intervals 560–520°C and 610–560°C, a third two-component control pTRM (pTRM3c) was given, following a procedure similar to that used for pTRM2c (Fig. 8).

All heatings were carried out in vacuum, under the same conditions as for the palaeointensity runs. The magnetic susceptibility was measured at room temperature after each pTRM acquisition. For samples 21, 47 and 199, sister specimens were subjected to the same procedure as the regular specimens but, in order to check the low-temperature (LT) memory of these remanences, these specimens were cycled to the liquid-nitrogen temperature in zero magnetic field prior to a.f. demagnetization.

4.2 ‘Initial’ pTRMs

Because it was not possible to demagnetize the samples completely using a 150mT alternating field (the highest value available in the laboratory), the pTRM intensity is defined here as the total amount of magnetization removed by a.f. demagnetization to 150 mT. This may introduce some error into the numerical values of the pTRMs but this error is quite small: the pTRM values obtained in this way agree well with those of the pTRMs measured for the sister specimens in a regular Thellier experiment.

The a.f. demagnetization curves of the ‘initial’ pTRMs, normalized by their initial values, are shown in Fig. 9. They all follow the trend of becoming harder with an increase of the blocking-temperature range, thus supporting the intuitive expectation (and the prediction of SD theory) that the stability of the pTRM increases with its blocking temperature. The shape of the demagnetization curves evolves from a typical multidomain-like curve to a more single-domain-like one, with a characteristic plateau at low alternating fields. The median destructive field increases considerably with the blocking temperature: it is equal to 5–8mT for pTRM(200–20) (15mT for sample 199) and reaches about 30mT for pTRM(T_c–560) (45mT for sample 199). Such a three-fold increase in unblocking field suggests that these pTRMs were not blocked by the same mechanism.

4.3 Control pTRMs

Owing to the experimental procedure used, six control pTRMs were available for the 200–20°C and 270–200°C intervals, three for the 460–330°C and 520–460°C intervals (see also Fig. 10), and only one for the 560–520°C and T_c–560°C intervals. The behaviour of the control pTRMs has much in common for all the samples studied, particularly regarding the pTRMs acquired in the two lowest-temperature intervals. The magnitude of the control pTRM(200–20), acquired after heating to 270°C, is reduced by a factor varying between 2 and 3 compared to the initial value. Similarly, the magnitude of the control pTRM(270–200), acquired after heating to 330°C, decreases by about 50 per cent. It is worth noting that even the first control pTRM(270–200) (acquired after heating to 270°C) has already decreased compared to its initial value. These results confirm the occurrence of irreversible changes in magnetic properties at very moderate temperatures.

Upon the subsequent heatings to progressively higher temperatures (up to 520°C), the pTRM acquisition capacity corresponding to the low-temperature intervals continues to decrease, although much less rapidly than before. A decrease of acquisition capacity also takes place for pTRM(460–330) after heating to 520°C; however, it is far less pronounced in its relative value than is the case for the low-temperature pTRMs. The decrease in pTRM acquisition capacity measured in this way matches quantitatively the diminution of pTRM as checked during the routine Thellier experiments.

Beyond 560°C, and especially at 610°C, another process starts that causes a significant increase of pTRM acquisition capacity and affects, more or less uniformly, the whole blocking-temperature spectra of the samples. The only exception is pTRM(610–560), which seems to decrease. However, a very small irreversible decrease (by just few degrees) of the Curie point produced by heating might also account for the observed decrease of the control pTRMs(610–560).

4 Low-temperature treatment of partial TRMs

Cycling to liquid-nitrogen temperature (77K) in zero magnetic field (LT treatment) is another way to gain insight into the blocking mechanism of the remanences carried by magnetite. In the course of an LT cycle, the sample passes through both the isotropic point, at which K₁ vanishes (around 130K), and the Verwey transition, from a cubic to a most probably monoclinic crystal structure (around 120K). It has long been known Kobayashi & Fuller 1968; Merrill 1970) that LT demagnetization affects primarily the fraction of remanence controlled by magnetocrystalline anisotropy, which is much more important, in relative measure, in multidomain grains. However, the multidomain remanence does not vanish fully after LT demagnetization, and the single-domain remanence does not remain unchanged Halgedahl & Jarrard 1995; Shcherbakova et al. 1996).

The effect of low-temperature treatment on the partial TRMs was investigated only for samples 21, 47 and 199, using specimens taken from locations directly adjacent to those used for the study of the regular pTRMs. After the treatment at liquid-nitrogen temperature in zero magnetic field, the low-temperature memory of the pTRM was subjected to the same a.f. demagnetization procedure as were the regular pTRMs. The LT memory is always at its minimum for pTRM(200–20), ranging from 50 to 60 per cent of the initial remanence, and it increases gradually with the temperature of the blocking interval of the pTRM (Fig. 11), approaching 90 per cent for the highest temperature range; this agrees with the recent results of Shcherbakova et al. (1996). However, in all our three samples the LT memory of pTRM(560–520) is 10 to 15 per cent lower than that of either pTRM(520–460) or pTRM(T_c–560). We have no explanation for this discontinuity. The whole data set suggests that almost half of the low-temperature remanence is carried by MD particles with pinning defects, whose coercivity is of magnetocrystalline origin.

The results of the a.f. demagnetization of the LT memory of the pTRM are plotted in Fig. 12, together with those for the original pTRMs. The low-temperature treatment removes preferentially the softest part of the remanence, in accordance with previous studies Kobayashi & Fuller 1968; Merrill 1970; Dunlop & Argyle 1991; Heider, Dunlop & Soffel 1992). However, for the low-temperature pTRMs, although their LT memory is more resistant to alternating-field demagnetization than the original pTRM, their demagnetization curves remain rather MD-shaped, in contrast with previous observations for either SIRM or total TRM (Merrill 1970; Heider et al. 1992). This suggests that even the part of the moderate-temperature pTRMs which resists LT cycling can be of MD origin, through internal stresses linked to crystal defects. In contrast, the high values of LT memory and strong resistance to alternating fields of the pTRMs acquired in the highest blocking-temperature intervals suggest an SD-like blocking process.

5 Effect Of Laboratory Heating On Hysteresis Characteristics

5.1 Room-temperature measurements

Hysteresis loops at room temperature were measured at Saint-Maur, using a maximum field of 800mT, on eight minicores a few millimetres in size cut from each sample. The first minicore was not heated; the other seven were first heated in vacuum to the various temperatures used for the experiments on partial pTRMs (Section 4). Some of the data (Fig. 13), particularly J_s and J_rs, show a significant scatter, which we attribute to variation of the ferrimagnetic oxide content between different minicores. Considering again the T₀–450°C interval where T_o is room temperature, it appears that heating in vacuum did not change the room-temperature hysteresis parameters by much, except for sample 86, which displays a significant decrease in coercivity. For several samples, the heatings to 560°C and 610°C resulted in significant or even drastic (sample 86) changes. On the Day plot (Fig. 14), the representative points shift towards the single-domain range. These changes are probably caused by the further development of ilmenite/magnetite intergrowths of much smaller size.

5.2 Measurements above room temperature

Measurements of hysteresis at high temperatures allow a more rigorous examination of the changes produced by heating. Furthermore, as we will see in the next section, these data allow us to investigate the origin of the coercivity. These measurements were carried out with a Princeton Measurements vibrating-sample magnetometer at the Institute for Rock Magnetism, Minneapolis. Two specimens from each sample were measured, using a maximum d.c. field of 1.0T. For the first set of specimens the complete hysteresis loops were traced at increasing temperatures from 25 to 610–630°C, with a 25°C increment below 450°C and a 10°C increment above. Because of the experimental set-up it was not possible to measure the coercivity of the remanence. For the second set of specimens, in order to distinguish between reversible and irreversible changes, measurements were carried out during heating, and also during cooling down to room temperature after reaching successively higher steps (175, 475, and 580 or 595°C). Also, after the 275°C step, the specimens were cooled to 175°C only. All measurements were performed under a helium atmosphere.

The temperature dependence of the saturation magnetization J_s(T) is shown in Fig. 2 for some selected samples from the second set of experiments. As pointed out above, and in agreement with the room-temperature measurements after heating in vacuum, the J_s(T) curves are basically reproducible. At most (samples 86, 117 and 199), a reduction of the order of 5 per cent is observed. In contrast, the coercive force (Fig. 15) exhibits marked irreversible changes. For all samples except sample 244, an irreversible decrease occurs at very moderate temperatures, which is documented by the H_c measurements made during cooling from 175 and 275°C. Note that in this temperature range, absolutely no changes in J_s occur (Fig. 2). This suggests that we are not dealing with chemical changes but, rather, with purely magnetic ones. After heating at 475°C and above a different process occurs: the coercivity exhibits an irreversible increase, accompanied for most samples by irreversible changes in J_s.

6 Origin Of Coercivity

The interrelations between the temperature dependences of hysteresis loop characteristics are an important source of information on the mechanisms of coercivity and hence on the remanence-blocking in a particular sample. The coercivity has its origin in the potential barriers that have to be overcome to enable the reversal of spontaneous magnetization in the case of single-domain grains or the motion of domain walls in the case of multidomain grains. In general, these barriers can be due to shape, magnetocrystalline or stress-induced anisotropy, the latter being governed in turn by magnetostriction. For single-domain grains, H_c∝J_s in the case of shape anisotropy, H_c∝K₁/J_s (where K₁ is the magnetocrystalline anisotropy constant) for magnetocrystalline anisotropy, and H_c∝λσ/J_s (where λ is the magnetostriction constant and σ is the internal stress) for stress-controlled anisotropy (Nagata 1961). In MD particles, the coercivity is due to crystal defects or inclusions. In this case, the coercivity is either of the form H_c∝K₁/J_s or H_c∝λσ/J_s (if internal stresses are present) (Hodych 1986). Since J_sK and λ for magnetite have very different temperature dependences, the origin of the coercivity of a magnetite-bearing sample can be inferred from the relationship between the temperature dependences of the coercive force and the saturation magnetization.

A convenient way to visualize this relationship is to plot the normalized coercive force against the normalized saturation magnetization on a bilogarithmic scale. The plot will then be a straight line if these two quantities are related through a single power law. The degree n of this power law can range from unity (coercivity due to pure shape anisotropy) to 6–9 (where magnetocrystalline anisotropy is dominant) Fletcher & O’Reilly 1974). A stress-controlled coercivity would exhibit intermediate values of n. Between 20 and 500°C, the magnetostriction constant λ has been found to vary as J2.31s for pure magnetite, while a larger power is found for Ti-substituted magnetites (,). The existing experimental data for magnetite-bearing rocks below room temperature (,, 1990, 1996), magnetite extracts ,) and synthetic magnetites ,; ) all give n values between 1 and 2, which are interpreted in favour of a mostly magnetostrictive origin of the anisotropy. The value of n seems to be almost independent of grain size, from the submicron range to several tens of microns. However, the data of Heider, Dunlop & Sugiura (1987) for hydrothermally recrystallized magnetites deviate somewhat from this trend, giving values of n between 2.5 and 4. This is readily explained by the low dislocation density in these crystals, which is hardly true for the exsolved magnetite grains typically present in basalts.

The bilogarithmic plots of coercive force versus saturation magnetization are shown in Fig. 16 for the first subset of specimens (measured during heating only). For all samples except sample 244, the initial part of the H_c–J_s plot, below 200°C or so, is characterized by a steeper slope than that beyond this temperature. This feature is at least partly due to the irreversible decrease in coercivity observed in this temperature range (Fig. 15), and the slope of this segment has therefore no significance in terms of anisotropy. On the other hand, for all samples, the slope of the curve is strongly and progressively reduced at high temperatures (beyond 500–530°C in general, but at only 300°C for sample 86). This feature correlates reasonably well with the irreversible increase in coercivity observed at high temperatures. Thus this high-temperature portion of the curve, which is not linear anyway, also has no significance regarding anisotropy.

Thus, only the linear, central part of the H_c–J_s curve, if present, can provide information about the origin of the anisotropy. The most reliable data are provided by sample 244, which is not affected by a noticeable decrease in coercivity at low temperatures (Fig. 15) and displays a linear dependence of the H_c–J_s plot extending from room temperature to 500°C. The coercive force thus follows a single power dependence on the saturation magnetization, with n≈1.2. No linear segment is observed for sample 86, and there are some doubts regarding sample 117, where the central part of the curve is slightly S-shaped. The other three samples yield n values between 0.83 (sample 21) and 1.26 (sample 199, Fig. 16). Note that the small irreversible increase in coercivity which is observed for all samples after heating to 475°C and beyond (Section 5) might bias n towards low values. Considering all our data, we conclude that the anisotropy is probably of mainly magnetostrictive origin. Magnetocrystalline anisotropy seems marginal.

The relationship between coercive force and saturation remanence has received little attention in the past. A common belief is that the coercive force should be directly proportional to the saturation remanence (e.g. Xu & Merrill 1990). However, for all our samples except sample 244, the H_c(T)–J_rs(T) curve (on a linear scale) corresponding to temperatures up to 400°C is clearly concave upwards rather than linear (Fig. 17). Since in the case of multidomain grains the saturation remanence and coercive force are related through the demagnetization factor N, the absence of linearity has to be ascribed to changes in N. This conclusion agrees with the results of the calculations of Xu & Merrill (1987), which show that for small multidomain grains N varies with the number of domains and approaches the ‘true’ multidomain limit of 4π/3 (for a sphere, in CGS units) only for grains with at least 10 domains. For grains with fewer than about 10 domains the demagnetization factor is subject to change with any rearrangement of the domain structure. Hence the non-linear relationship between the coercive force and the saturation remanence suggests that considerable changes in the domain structure of the magnetic mineral occur in our samples in the interval 20–400°C.

7 Discussion

Among the various mechanisms which might produce a non-ideal behaviour during Thellier experiments, two of them can be readily discarded. Any effect from secondary magnetizations is quite unlikely. The secondary overprints are fairly small and are removed at 250°C at the most, which is approximately the temperature at which the anomalous decrease of the NRM starts. We have no special reason to suppose that the primary remanence is not a TRM: as already mentioned, spinodal decomposition is commonly believed to occur above 600°C (Haggerty 1976). Thus, the magnetite in our samples is expected to carry a TRM. Note, however, that in the absence of pseudo-brookite in our rock samples, we have no specific data which might prove this general statement for certain.

As mentioned above, there are three main possible explanations of the behaviour of our samples:

(1) an intrinsic multidomain effect, involving no chemical or magnetic change;

(2) some irreversible chemical changes;

(3) some irreversible physical changes.

These three possibilities are discussed below in the light of our experimental data.

7.1 Intrinsic multidomain effect

Levi (1977) and McClelland et al. (1996) showed that magnetite grains with a size of several microns fail to yield the correct results in terms of palaeointensity, and always produce ‘NRM’–TRM plots that are concave upwards. Levi (1977) showed that the degree of concavity increases gradually with the mean grain size of the magnetite, which varies from about 0.1 to 2.7μm for his samples. For the largest size, the ratio of the slopes corresponding to the lowest- and highest-temperature parts of the NRM–TRM plots is about two. Thus, this intrinsic effect of grain size seems insufficient to explain the shape of the NRM–TRM plots of most of our samples.

This expectation is confirmed by the results of our TRM₁–TRM₂ experiments. These experiments were carried out on three samples (samples 47, 117 and 119) which are characterized by a large, steep decrease of the NRM–TRM plot at intermediate temperatures and show only small changes in the hysteresis parameters after heating. Even for sample 117, which shows the most pronounced concavity of the TRM₁–TRM₂ plot (Fig. 6), the ratio of the two slopes mentioned above is less than 1.4. This is much smaller than that observed on the NRM–TRM curve (Fig. 5). For the other samples (47 and 199), the discrepancy between the NRM–TRM and the TRM₁–TRM₂ curves is even more blatant.

This behaviour is in agreement with our other magnetic observations. The hysteresis parameters, resistance to alternating field and LT memory are all too large to be compatible with the presence of a significant fraction of large MD grains. At most, such grains could carry most of the low-temperature pTRMs, as discussed above. But these pTRMs contribute only a few per cent of the total TRM intensity, and thus cannot seriously affect the results of the Thellier experiments.

Another important conclusion to be drawn from these TRM₁–TRM₂ experiments is that this peculiar behaviour is specific to the natural TRM. Thus the observed changes are irreversible, and occur during the first laboratory heatings in the NRM–TRM experiments.

7.2 Irreversible chemical changes

These changes are most easily detected from the k(T) and, to a lesser extent, the J_s(T) experiments. Let us first recall that two magnetic minerals can be identified in our samples: (1) almost pure magnetite, which is present in all samples and is thermally stable (Fig. 2); and (2) a phase with a Curie point of about 400°C (probably titanomagnetite), which is unstable upon heating in vacuum to about 600°C (Fig. 2). This phase presents hysteresis properties suggesting a larger magnetic grain size than that of the magnetite. This second phase is important only in sample 86.

Chemical changes can explain the shape of the NRM–TRM plot if the main magnetic phase was strongly altered or partially destroyed at intermediate temperatures during the Thellier experiments. Figs 6 and 8 show that the temperature range of this process is best identified in samples 47, 86 and 199, where it extends from approximately 200 to 400°C. There is no magnetic evidence for any large chemical changes occurring in this temperature interval: the saturation magnetization measured at room temperature is not systematically modified (Fig. 14) and the room-temperature susceptibility either remains unchanged or decreases only slightly. In the second case, the decrease is not specific either to the 200–400°C temperature interval or to the samples which are most affected by the suspicious NRM decrease.

A more indirect but possibly more sensitive indication of chemical alteration might be the development of some CRM (chemical remanent magnetization) during heating. In this respect, it is of interest to note that the representative points of the ‘NRM’, drawn from palaeointensity measurements, do not exhibit as perfect an alignment on the orthogonal diagrams (Fig. 5) as that observed for conventional thermal demagnetization (heating in zero field, Fig. 1). Between 200 and 353°C (329°C for sample 199), the NRM directions are defined well by straight segments passing through the origin, and closely agree with the directions determined from thermal cleaning in zero field. Above this temperature, the ‘NRM’ inclinations (in sample coordinates) shift slightly towards the direction of the laboratory field applied during the first heating, which suggests that a small CRM may have been acquired. However, this process of chemical alteration, if real, does not start at sufficiently low temperatures to be somehow linked to the anomalous decrease in NRM, which begins at 200°C.

Thus we think that there is no evidence for a chemical alteration occurring between 200 and 400–460°C that is large enough to destroy the remanence in a proportion that can account for the large anomalous decrease in the NRM (Table 1) observed during the Thellier experiments.

7.3 Irreversible physical changes

Evidence for some physical alteration (i.e. alteration of magnetic properties) during heating at moderate temperatures is mainly provided by the coercivity data. The decrease observed at low temperatures seems to operate up to 200–250°C (Fig. 15). Annealing is known to decrease the coercive force of magnetite (Parry 1965; Lowrie & Fuller 1969; Smith & Merrill 1984), but the temperatures involved are much higher. We tentatively suggest that some decrease in internal stress is the cause of the coercivity decrease at low temperatures. The increase in coercivity, which generally starts between 275°C and 475°C, is an effect opposite to that of annealing. It can be due to some increase of internal stress due to the formation of microchemical heterogeneities, which, ultimately, will form new exsolutions. In agreement with this suggestion, we note that the hysteresis characteristics after the 610°C heating tend more towards those of SD grains; this is particularly pronounced for sample 86 (Fig. 14).

Bearing in mind all the experimental results, we are obliged to conclude that on heating our samples to moderate temperatures an irreversible process takes place that alters the samples’ magnetic properties, while the chemical composition of the remanence carriers remains virtually unchanged. There are essentially two magnetic changes: an irreversible decrease of coercive force at low temperatures and an anomalous loss of NRM between approximately 200 and 400–460°C.

In the absence of a consistent theory for pseudo-single-domain grains, the exact physical nature of this process is difficult to establish. Only a very crude and entirely qualitative model will be considered here. Considering the six samples we studied, we note that only one of them (sample 244) exhibits no decrease in coercivity at low temperatures. This sample also exhibits the smallest anomalous NRM decrease. Thus, the low-temperature decrease in coercivity seems to be at the origin of this anomalous behaviour. The coercivity of MD grains is controlled by crystal defects in two different ways. Defects are privileged sites for the nucleation of reversed domains and, subsequently, they control domain wall progression by pinning/unpinning. We suggest that the decrease in coercivity which occurs at low temperatures (up to 200–250°C) in our samples results in a decrease in the critical field needed for nucleation. If, as we assumed above, this coercivity decrease is due to a reduction in residual microstress, then the total free magnetic energy is modified and new local energy minima (LEM) can appear. Moreover, the increase in temperature itself can favour domain rearrangements into other LEM states (Enkin & Dunlop 1987; Enkin & Williams 1994). Thus we suggest that the decrease in coercivity and the increase in temperature result in a rearrangement of the magnetic domain configuration of some grains. This interpretation is supported by the fact that the demagnetizing-field coefficient N, as calculated from hysteresis data (Section 6), changes during heating at moderate temperatures, except for sample 244.

As a result of domain rearrangement, these grains would lose most, if not all, of their NRM. Obviously, the pTRM acquired during the subsequent cooling to room temperature would have absolutely no relation to the magnitude of the NRM lost during heating. If this reorganization occurred at moderate temperatures, the pTRM gain could be extremely small compared to the loss of the NRM (which is a total TRM). Such a process would result in an extreme steepness of the NRM–TRM curves, as observed for most of our samples at intermediate temperatures.

8 Conclusions

Drastic non-ideal behaviour during Thellier experiments can be observed in basalts. This anomalous behaviour is characterized by a large decrease in the NRM at intermediate temperatures which is not accompanied by a mirror increase in the TRM. This behaviour is observed in the absence of significant chemical changes due to the laboratory heatings. The presence of MD grains is not, in the present case, a plausible explanation for this anomalous behaviour.

We suggest that this behaviour is due to the transformation of the micromagnetic structure of PSD grains from a metastable configuration to a more stable one, as a result of an irreversible decrease in coercivity which occurs at relatively low temperatures (up to 200–250°C). This decrease is probably due to a reduction of internal stress during the laboratory heatings. This irreversible physical change can lower the nucleation field of some crystal defects, which, together with the moderate temperature increase, can favour the rearrangement of magnetic domains. For individual crystals, such a rearrangement occurring during heating would result in a large, if not total, NRM loss, while the pTRM acquired during the subsequent cooling could be quite small. The proposed process is entirely decoupled from the blocking/unblocking process: it can affect any fraction of the NRM regardless of its ‘true’ (i.e. caused by thermal fluctuations) unblocking temperature.

We have also shown here that the measurement of complete hysteresis loops as a function of temperature provides an independent and apparently very sensitive method for testing the magnetic stability of rock samples as a function of temperature. It seems to be of interest to use this type of investigation, as a complement to the classic tests of the Thellier palaeointensity method, in order to ensure better the validity of palaeostrength determinations.

Acknowledgements

This work was supported by CNRS-INSU under the DBT programme ‘Terre Profonde’ (contribution CNRS-INSU-DBT No. 82). A. Kosterov’s stay in Montpellier was made possible thanks to a PhD grant from the French Government. We thank James Marvin and Mike Jackson (Institute for Rock Magnetism, Minneapolis, USA) and Maxime LeGoff (Laboratoire de Géomagnétisme, Saint-Maur, France) for help with the hysteresis measurements. The manuscript has benefited from the reviews of Buffy McClelland and two anonymous referees. Funds for the IRM operation were provided by the Keck Foundation and the University of Minnesota.

References