Curie temperatures of synthetic titanomagnetites in the Fe-Ti-O system: Effects of composition, crystal chemistry, and thermomagnetic methods



[1] The present study is aimed at improving the calibration of the compositional dependence of the Curie temperature (TC) of titanomagnetite (Tmt) on the basis of temperature-dependent magnetic susceptibility (χ-T) curves measured on synthetic Tmts in the Fe-Ti-O system. In order to assess the possible influence of high-temperature cation vacancies onto the TC values, we have synthesized two types of assemblages in subsolidus conditions at 1 bar, 1100°C and 1300°C, under controlled oxygen fugacity conditions. Tmts synthesized in equilibrium with ilmenite-hematitess (Ilmss) are expected to have the highest vacancy concentrations, those in equilibrium with wüstite (Wus) the lowest. The composition and homogeneity of the synthetic Tmts were carefully checked with a scanning electron microscope (SEM) and an electron microprobe (EMP). TC was determined from χ-T curves using a kappabridge and, for comparison, from Ms-T curves measured with a variable field translation balance. Our data set shows systematically higher TC values for Tmt coexisting with Ilmss than for Tmt coexisting with Wus. Most χ-T curves are nonreversible, whereby the largest ΔTC (40 K) concern Tmt(+Ilmss) of intermediate compositions synthesized at 1300°C. Nonreversibility is interpreted as reflecting cation reordering in Tmt during the high-temperature χ-T measurements. TC values obtained from Ms-T curves are higher than those obtained from the χ-T curves, whereby the difference regularly increases (up to 40 K) with increasing Ti content, up to XUsp = 0.6. Our new calibration curves are suitable to retrieve Tmt compositions in basalts that were rapidly cooled and not oxidized by deuteric or hydrothermal fluids.

1. Introduction

[2] Measurements of the temperature dependence of the AC magnetic susceptibility are increasingly used to identify magnetic minerals, in particular members of the two solid solutions of iron-titanium-oxide minerals, titanomagnetite (Tmt) and ilmenite-hematite (Ilmss) which are the essential carriers of magnetism in basalts [e.g., Dunlop and Özdemir, 1997; Calvo et al., 2002; Miranda et al., 2002]. This method has proven very useful to estimate the chemical composition of titanomagnetite (solid solution between magnetite, Fe3O4, and ulvöspinel, FeTiO4), especially in case of crystals that are not measurable with the electron microprobe, because they are very small or skeletal or crowded with fine lamellae of ilmenite or of other spinel phases (see examples of Haggerty [1991]). In case of rocks that contain different generations of Tmt, the temperature-dependent susceptibility measurements can help estimate the corresponding different chemical compositions of this phase and reconstruct the crystallization and alteration history of the basaltic rock [e.g., Zhou et al., 2000; Kontny et al., 2003]. The magnetic susceptibility measurements have also the great advantage of being rapid and straightforward and, in principal, nondestructive. They can be performed on very small chunks of rock (a few mg in weight), i.e., without any time-consuming and costly preparation procedure.

[3] Estimates of the chemical composition of Tmt are based on the compositional dependence of the Curie temperature (TC) first established by Akimoto et al. [1957] and confirmed by a wealth of further experimental studies in the Fe-Ti-O system [e.g., Ozima and Larson, 1970; Bleil, 1973, 1976; Hauptman, 1974; Rahman and Parry, 1978; Cisowski, 1980; Deutsch et al., 1981; Tucker, 1981; Özdemir and O'Reilly, 1981; Moskowitz, 1987, 1993; Wanamaker and Moskowitz, 1994; Dunlop and Özdemir, 1997; Moskowitz et al., 1998; Özdemir, 2000]. All experimental results point to a negative correlation between Curie temperature and ulvöspinel content of the titanomagnetites (Figure 1). Second-order polynomial regression curves have been proposed to estimate the mineral composition from TC or vice versa (equations reported by Bleil [1976], Bleil and Petersen [1982, p.330], Clark [1997, p. 86], and Hunt et al. [1995]; see Figure 1). Because synthetic titanomagnetites with a few wt % Al2O3 or MgO display Curie temperatures lower by no more than 40°C [Richards et al., 1973; O'Donovan and O'Reilly, 1977; Özdemir and O'Reilly, 1978, 1981; Moskowitz, 1993], i.e., in rough agreement with the Al- and Mg-free ones, the regression curves appear also applicable to complex, nature-relevant Tmt compositions. Indeed, these curves have been commonly used in rock magnetic and paleomagnetic studies [e.g., Bücker et al., 1986; Gonzalez et al., 1997].

Figure 1.

Literature data on the variation of the Curie temperature (TC) in synthetic titanomagnetites in the Fe-Ti-O system as a function of composition (mole fraction of the ulvöspinel end-member, XUsp). The curve represents the polynomial fit TC = −150 XUsp2 − 580 XUsp + 851 [Bleil and Petersen, 1982, p 330].

[4] Nevertheless, even for synthetic Tmt in the Fe-Ti-O system, the available literature data on the Curie temperature scatter significantly, allowing only rough estimates of their compositions, with uncertainties around ± 0.1 for the mole fraction of the ulvöspinel end-member. This scattering partially reflects uncertainties concerning the chemical composition of the synthetic titanomagnetites used for calibration. In several studies it is not clear whether the samples were really single-phased and chemically homogeneous because they were not examined with microanalytical methods or because the authors were not aware of the possible compositional heterogeneity in single-phase polycrystalline samples [e.g., Akimoto et al., 1957; Hauptman, 1974; Rahman and Parry, 1978; Tucker, 1981; Moskowitz, 1993]. Only few studies have considered the possible influence of the high-temperature nonstoichiometry of Tmt on its Curie temperature [Hauptman, 1974; Rahman and Parry, 1978; Moskowitz, 1987; Wanamaker and Moskowitz, 1994]. Since the concentration of cation vacancies at a given Tmt composition decreases both with decreasing temperature and oxygen fugacity during synthesis [e.g., Aragon and McCallister, 1982; Senderov et al., 1993; Aggarwal and Dieckmann, 2002; Lattard et al., 2005], these parameters should also be considered in establishing calibration curves for TC. This point is important for the application to natural titanomagnetites. Curie temperatures derived from single-phase Tmt synthesized at 1275 to 1400°C, with unknown, but possibly high, vacancy concentrations are not directly comparable to those of titanomagnetites that crystallized in basalts at temperatures ≤1100°C with expectedly low vacancy concentrations.

[5] Another, not sufficiently understood factor to be considered for the determination of Tmt composition from TC is the influence of the cation distribution within the Tmt structure on the magnetic properties. There is considerable disagreement in the literature upon the cation distribution in Tmt and its temperature dependence [e.g., Akimoto, 1954; Néel, 1955; Chevallier et al., 1955; O'Reilly and Banerjee, 1965; Stephenson, 1969; Bleil, 1976; Trestman-Matts et al., 1983; Wechsler et al., 1984; O'Neill and Navrotsky, 1984] (see also review by Waychunas [1991]). In fact, it is obvious that the equilibrium cation distribution during high-temperature synthesis must be temperature-dependent, but the question is whether it can be frozen in during quenching and may be altered during the magnetic measurements at temperatures up to 700°C. Whereas earlier studies suggested distinct cation distributions in Tmt quenched from different temperatures [e.g., Akimoto et al., 1957; O'Reilly and Banerjee, 1965; Stephenson, 1969; Bleil, 1971, 1976], Wechsler et al. [1984] found no significant difference in the unit cell parameters or magnetization determined in Tmt samples quenched from temperatures of 930 to 1350°C or annealed at 800°C. Jensen and Shive [1973] stated that electron diffusion in Tmt is too rapid to preserve a high-temperature equilibrium distribution by quenching but that the distribution may well depend on the ambient temperature of measurement. O'Neill and Navrotsky [1984] proposed that differences in samples quenched from different high temperatures may have been complicated by imperfect control of stoichiometry. In the same line of reasoning, Wanamaker and Moskowitz [1994] suggested that long-range ordering in an intermediate titanomagnetite (60% of the ulvöspinel end-member, i.e., “TM60”) is a function of nonstoichiometry, with higher cation vacancies producing a more random cation distribution. They concluded that this effect may explain the differences among earlier cation distribution models.

[6] To our knowledge, an important methodological aspect has not been considered in previous studies that used TC values retrieved from χ-T curves to estimate the composition of titanomagnetites. The majority of the TC values used for calibration have been obtained from temperature-dependent saturation magnetization (Ms-T) curves, not from temperature-dependent susceptibility (χ-T) curves. Deutsch et al. [1981] have shown that for synthetic TM60 titanomagnetite, the Curie temperature obtained from Ms-T curves is close to that derived from χ-T curves and Rahman and Parry [1978] state that both types of values agree within ±5°C. However, we know of no further comparison between TC values retrieved from the two types of measurements.

[7] The main goal of the present study is to improve the calibration curve for the compositional dependence of the Curie temperature of titanomagnetite on the basis of temperature-dependent magnetic susceptibility curves measured on synthetic titanomagnetites in the Fe-Ti-O system. In order to assess the possible influence of high-temperature cation vacancies onto the magnetic properties, we have synthesized two types of polycrystalline Tmt-bearing assemblages in subsolidus conditions at 1300°C (Figure 2). As known from previous experimental studies [e.g., Taylor, 1964; Dieckmann, 1982; Aragon and McCallister, 1982; Senderov et al., 1993; Lattard, 1995; Aggarwal and Dieckmann, 2002] titanomagnetites in coexistence with ilmenite-hematite solid solution (Ilmss) are expected to have the highest possible vacancy concentrations at the temperature of interest, those in paragenesis with wüstite (Wus) should be devoid of vacancies. For comparison to previous studies, some single-phase Tmt samples (1300°C) were also considered. For more relevance to basalts, we have also examined Tmt-Ilmss assemblages synthesized at 1100°C. We have taken special care to synthesize samples in which the titanomagnetite is chemically homogeneous, both within the crystals and over the whole polycrystalline sample. In a few cases, chemical zonations or local chemical heterogeneities are present in the synthetic samples and we shall show in the following how these features influence the magnetic properties. We have also considered possible methodological artefacts by comparing TCs retrieved from temperature-dependent susceptibility versus saturation magnetization curves (χ-T versus Ms-T curves).

Figure 2.

Phase diagram showing the chemical composition of titanomagnetites and other coexisting Fe-Ti oxide phases synthesized in the present study at 1300°C under different oxygen fugacities. The oxygen fugacities are given as ΔNNO = log fO2 (sample) − log fO2 (Ni-NiO buffer). Single-phase fields are gray, two-phase fields white. In the latter fields, two tie lines at the same oxygen fugacity are exemplified by thin horizontal lines. Symbols point to the phase assemblages: circles for Tmt-Ilmss; triangles for Tmt-Wus; diamond for Tmt-Fe°; squares or rectangles for single-phase Tmt (rectangles point to compositional range in case of heterogeneous Tmt). Open symbols are for Tmt; solid symbols are for Wus or Ilmss.

2. Methods

2.1. Syntheses

[8] All syntheses were performed at 1100 or 1300°C (subsolidus conditions) in vertical quench furnaces under a variety of oxygen fugacities fixed either by CO/CO2 gas mixtures or by solid-state oxygen buffers (only for 1100°C synthesis). The temperature was measured before and after the runs with a type S (Pt-Pt90Rh10) thermocouple calibrated against the melting points of silver (960.8°C) and gold (1064.4°C). To ensure a complete equilibration of the synthesis products, runs lasted >24 hours at 1300°C, but up to 168 hours (7 days) at 1100°C.

[9] The starting mixtures were prepared from dried TiO2 (99.9%, Aldrich Chemical Comp. Inc.), dried Fe2O3 (99.9%, Alpha Products), as well as dried and reduced (under H2, 500°C) metallic iron (99+%, Heraeus) which were weight in stoichiometric proportions, ground together and mixed under acetone in an agate mortar and pressed to pellets of 200–300 mg (about 5 mm diameter, 4 to 6 mm in length). Metallic iron was employed only in starting mixtures for experiments performed with solid-state oxygen buffers. In this case, the original oxygen content of the sample should be close to that of the run product because of the restricted buffer capacity.

[10] For the solid-state buffer experiments, pellets of sample and buffer material (iron-wüstite, wüstite-magnetite, cobalt-cobalt oxide, fayalite-magnetite-quartz or nickel-nickel oxide) were inserted in silica-glass tubes, which were evacuated with a rotary vane pump to a vacuum in the order of 10−2 mbar prior to sealing. A silica-glass filler rod was used to minimize the internal volume of the ampoule and to separate the sample from the buffer. At the end of the experiments, the silica-glass ampoules were pulled out of the furnace and quenched into water, a procedure that lasted less than one minute. In all runs referred to in this paper all buffer phases were still present after termination of the experiments, i.e., the desired oxygen fugacity was maintained during the whole experiment.

[11] For the gas-mixing experiments, the sample pellets were placed on a grid of platinum wire. As sample and metal only share a very small surface and no melt ever occurs in the samples, negligible Fe loss to the wire, but maximum contact of sample with gas mixture can be achieved. High-purity CO (CO > 99.97 vol %) and CO2 (CO2 > 99.995 vol %) gases were mixed with electronic valves (Millipore) and allowed to flow from the bottom to the top of the furnace tube (inner diameter 4 cm) at a rate of 200 cm3/min. The oxygen fugacity was measured after the experiments with an yttria-stabilized zirconia sensor (SIRO2) with air as the reference. The sensor was calibrated at 1300°C against the Ni-NiO equilibrium [O'Neill and Pownceby, 1993]. We estimate the accuracy of the experimental fO2 values at about ±0.2 log unit at moderate to low oxygen fugacities (ΔNNO < +1; CO > 1 vol %), but up to about ±0.5 log unit at high fO2 (more details from Lattard et al. [2005]). All fO2 values given for the gas-mixing experiments in the following (Table 1) are those measured with the zirconia sensor. Since we also list the CO contents of the gas mixtures, the readers can easily retrieve the fO2 values from the tables of Deines et al. [1974] for comparison. Run products were generally drop-quenched into water at the bottom of the furnace, a procedure which ensures a very fast cooling (within 10 s). In a few cases, however, the gas flow was first turned off and the samples were pulled out of the furnace and quenched into water. The whole procedure usually took less than 1 minute. In the following it is referred to as “external quench.”

Table 1. Synthesis Conditions and Modal Compositions of Synthetic Titanomagnetite-Bearing Samples, Chemical Compositions, and Curie Temperatures
RunBulk Ti/(Ti+Fe), at. % Phases Tmt ChemistryMs-T Tc, K HeatTc From χ-T Curves,h K
Synthesis ConditionsTmt, vol %Ti/(Ti+Fe) (at.%)XUspgPeak Position1/χ MethodGrommé Method
T, °Ct, hoursBuffera or CO%blog fO2cΔNNOdMeane1σeMeanf1σfHeatCoolHeatCoolHeatCool
  • a

    Abbreviations of the buffer names: FMQ, fayalite-magnetite-quartz; IW, iron-wüstite; MW, magnetite-wüstite; NNO, Ni-NiO.

  • b

    Volume percent of CO in the CO-CO2 gas mixtures.

  • c

    The log fO2 values derived from emf measurements with a zirconia cell or from buffer values; log fO2 values for oxygen buffers from O'Neill [1987] (FMQ); O'Neill [1988] (IW, WM); O'Neill and Pownceby [1993] (NNO, Co/CoO).

  • d

    ΔNNO = log fO2(experimental) − log fO2(NNO buffer).

  • e

    Mean and standard deviation over three to six image analyses. If no value is given for 1σ only one image analysis was performed.

  • f

    Mean and standard deviation over ten single EMP analyses.

  • g

    XUsp = 3Ti/(Ti + Fe) (atomic proportion).

  • h

    Tc from χ-T curves from heating (heat) and cooling (cool) paths. Tc values obtained from χ-T curves measured with KLY 2 are in italic.

  • i

    Approximate Curie temperature estimated from complex χ-T curves or from those with rounded peaks (see section 5.1).

6F80x0.75b20.01300240.7−5.21.4Tmt+Ilm79(2)14.30(14)0.429 586620590628594636
6F69x1.531.01302241.5−5.90.8Tmt+Ilm23(11)18.36(14)0.551 483524490524i492558i
6F76x3.424.01299473.4−6.70.0Tmt+Ilm95 22.73(17)0.682398365389371395i372402
F72 (f) 627.61300225.8−7.1−0.4Tmt+Ilm93 25.32(18)0.760 302315311i325312328
M1300-831.01300248.2−7.7−1.0Tmt+Ilm66(6)27.77(17)0.833 235 237 238 
6F57x1843.313012418.0−8.3−1.6Tmt+Ilm30(6)30.20(13)0.906 187 187 193 
6F63x3437.013002834.0−9.1−2.4Tmt+Ilm78(3)32.09(10)0.963 145148147149i149150
6F57x3443.313002834.0−9.1−2.4Tmt+Ilm35(6)32.16(9)0.965 147 148 152 
6IT60x4940.013002449.0−9.7−3.0Tmt+Ilm45(4)33.60(14)1.008 128 129 132 
6F57x6643.312991866.0−10.2−3.5Tmt+Ilm48(3)34.39(17)1.032 109 109 114 
6F100x00.01300250.0−3.43.3Tmt100 0.00 0.000 865 866 869 
6F100x2.40.01300452.4−6.40.3Tmt100 0.00 0.000851864873865879868880
6F97x03.01300250.0−3.43.3Tmt100 2.65(5)0.080 824823832850i836850
6F72qx7128.012992571.0−10.4−3.1Tmt100 28.29(16)0.849 200 204i 205 
6F69x8131.013002581.0−10.9−4.2Tmt100 30.94(20)0.928 144 145 148 
6F69x81.531.012995181.5−10.9−4.2Tmt100 31.39(27)0.942 139 140 146 
8F68x8132.013002581.0−10.9−4.2Tmt100 31.95(11)0.959 128 130 135 
6F67x8133.013002581.0−10.9−4.2Tmt100 32.95(11)0.989 117 118 120 
6F97x103.013002610.0−7.7−1.1Tmt+Wus75 3.71(5)0.111 787792787i792i794i806i
6F97x113.013002811.0−7.8−1.2Tmt+Wus39 5.57(8)0.167751753760752i760i760i768i
6F97x123.013002312.0−7.9−1.3Tmt+Wus56(5)6.93(7)0.208 726729726i729i730i736i
6F92x148.013002314.0−8.1−1.5Tmt+Wus76 9.71(9)0.291670661668664i671i669675i
6F92x188.013005218.0−8.3−1.6Tmt+Wus46 13.38(10)0.401595585596586595i594600i
6IT90x1810.013005218.0−8.3−1.6Tmt+Wus67 13.42(6)0.403593582591586592i592597i
6F90x23.510.013002223.5−8.6−1.9Tmt+Wus60(5)15.85(12)0.476 524529524529528536
6F83ax3417.013002034.0−9.1−2.4Tmt+Wus75(5)21.10(13)0.633 388401397407400410
6F83ax4917.013002449.0−9.7−3.0Tmt+Wus62(9)23.69(12)0.711 319329324332326340i
6F83x6617.013002566.0−10.2−3.5Tmt+Wus42(11)26.79(12)0.804 239242240241254260
6F76x6623.513002566.0−10.2−3.5Tmt+Wus67(1)27.07(10)0.812 235 238i 248 
6F80x8120.013002581.0−10.9−4.2Tmt+Fe°81(4)30.17(25)0.905 158 158 162 
F76 (c) 323.51100721.3−7.71.2Tmt+Ilm52(6)13.87(25)0.416606593629593640i604688
3F69x1.2531.011011441.25−8.40.5Tmt+Ilm41(2)18.52(10)0.556 470506i474 476558
3F69Qe31.01098120NNO−8.90.0Tmt+Ilm36(1)20.14(15)0.604 439450443462i445468
F63 (f) 337.01100120FMQ−9.7−0.7Tmt+Ilm71(1)23.26(19)0.698368344352345373i354388
F69 (X) 331.01104725.5−9.8−0.9Tmt+Ilm85(3)24.69(21)0.741 303 303 307 
F72 (X) 328.01104725.5−9.8−0.9Tmt+Ilm48(3)24.87(18)0.746 300 303 303 
F63 (g) 337.0110093Co/CoO−10.4−1.5Tmt+Ilm56(2)26.36(25)0.791 260 263 264 
3P69WM31.01100144WM−11.0−2.0Tmt+Ilm66(5)27.54(12)0.826 217 218 226 
3F63x16.537.0110013516.5−10.9−2.0Tmt+Ilm47(2)28.52(18)0.856 206 208 209 
3F63x3037.011009630.0−11.5−2.6Tmt+Ilm82(2)29.81(26)0.894 170 172 173 
3P63IW37.01100144IW−13.3−4.4Tmt+Ilm66(2)32.38(14)0.971 122 122 123 
3IT60IW40.01100153IW−13.3−4.4Tmt+Ilm55(4)32.46(16)0.974 118 118 119 
M1100-1343.51100120IW−13.3−4.4Tmt+Ilm36(4)32.95(10)0.989 116 117 118 

[12] All run products were carefully characterized using optical microscopy, X-ray powder diffraction, BSE images from a scanning electron microscope (SEM), including image analysis to determine modal proportions, and chemical analyses with the electron microprobe (EMP).

2.2. Magnetic Measurements

[13] The AC magnetic susceptibility (χ) was measured as a function of temperature in the range between 80 and 970 K on small chunks (5 to 25 mg) of the pellets retrieved from all synthesis runs. We used both a KLY 2 and a KLY 4 Kappabridge, combined with a CS-2/CS-L furnace (AGICO; see details from Hrouda [1994]) operating at a low field of 300 A/m and a frequency of 870 Hz for KLY 4 and 920 Hz for KLY 2. The measurements were performed in two steps. First, the samples were cooled down to 80 K with liquid nitrogen and temperatures and susceptibilities were recorded during the subsequent warming up to 273 K. In a second step, the magnetic susceptibility was measured during a heating and cooling cycle from room temperature to 970 K and back to room temperature. The heating rate was 10 K/min. To avoid mineral reactions with atmospheric oxygen, the measurements were performed in a flowing argon atmosphere (110 ml/min). The temperature was measured with a Pt resistance thermometer that was placed within 1 mm distance to the sample. According to the manufacturer of the resistance thermometer (JUMO), the recorded temperature values are accurate within ±1 K at temperatures up to 423 K but within ±3 K in the range 423–973 K. The raw susceptibility data were corrected for the empty furnace and normalized to the susceptibility magnitude at 273 K. Repeated χ-T measurements with the same kappabridge on different chunks of the same sample yield TC from the heating curve that are all within ±5°, i.e., show a good reproducibility. At temperatures above room temperature, however, the TC retrieved with kappabridge KLY 4 are up to 12 K higher than those obtained with KLY 2. The reasons for this discrepancy are not clear, but in the following we shall rely essentially on the values measured with the KLY 4 Kappabridge. Practically all TC values above room temperature that are listed in Table 1 have been obtained with the latter apparatus (only 3 exceptions!). To avoid any confusion, the TC values retrieved with the KLY2 are in italics in Table 1.

[14] Thermomagnetic curves were obtained using a variable field translation balance (VFTB) at the Department for Earth and Environmental Sciences at the Ludwig-Maximilian Universität, Munich. Details on the method are given by Leonhardt [2006]. Prior to the thermomagnetic curves, hysteresis loops at room temperature were measured with the VFTB (maximum magnetic field H = 0.63 T) in order to determine the saturation fields. Ms-T curves were obtained by measuring the induced magnetization from room temperature to 970 K in argon atmosphere at saturation fields as determined from the hysteresis loops (see description of the method by Matzka et al. [2003]).

3. Synthesis Products

[15] All run products consist of polycrystalline, roughly equigranular aggregates, with grain sizes around 10 to 50 μm. The majority of these run products consist of assemblages of two coexisting oxide phases, either Tmt + Ilmss or Tmt + Wus (Figure 2), which have been synthesized either at 1300°C or at 1100°C. For comparison, we have also synthesized a few titanomagnetite single-phase samples (Figure 2). The run conditions, the modal composition of the samples and the chemical composition of the synthetic titanomagnetites are given in Table 1, together with the Curie temperatures determined from temperature-dependent magnetic susceptibility curves or from temperature-dependent saturation magnetization curves.

[16] With the exception of wüstite, which unmixes tiny titanomagnetite crystals during quenching [e.g., Simons and Woermann, 1978; Senderov et al., 1993], the drop-quenched run products show no sign of any chemical or textural changes during quenching. In all drop-quenched, two-phase run products, the relatively large titanomagnetite crystals are chemically homogeneous, within the crystals and over the whole sample [cf. also Lattard et al., 2005]. Only this type of samples has been used to recalibrate the compositional dependence of the Curie temperature of titanomagnetite.

[17] However, we wish to caution against some commonly used synthesis procedures which yield samples with inhomogeneous titanomagnetite. For instance, different titanomagnetite compositions appear in products of gas mixing runs that were not drop quenched but instead were pulled out of the furnace after the turning off of the CO gas and quenched into water outside the furnace (external quench). Through this procedure the still glowing samples (T > 800°C) are exposed to air for a few seconds. The outermost rim of these samples generally displays exsolution features in form of very fine lamellae or rims of Ilmss within or around the Tmt crystals (Figure 3c), or of Psbss lamellae in Ilmss host crystals. The composition of the Tmt crystals containing Ilmss-lamellae is iron richer than those in the central part of the sample pellet. Such features may appear in single-phase samples as well as in two-phase products. In any case, they are restricted to the outermost surface of the sample or to superficial cracks, i.e., to regions of the samples that were in contact with the surrounding gas during the external quench [Lattard et al., 2005]. These exsolutions can be understood as “oxy-exsolutions” in the sense of Buddington and Lindsley [1964]. The zones affected by oxy-exsolution represent only few percent of the total volume of the sample pellets, but (as we shall see in the following) they can significantly influence the magnetic properties of the samples. Such features were most probably present in some of the synthetic products used in other studies but have not been described because the authors did not (in older studies could not) examine their samples with the SEM.

Figure 3.

Effect of chemical heterogeneity on the magnetic properties of a synthetic polycrystalline titanomagnetite sample (run product 6F92x1.6; bulk XUsp = 0.24; synthesized through a single sintering at 1300°C in a CO/CO2 gas mixture). (a) BSE image of a polished section through part of the sample. (b) Ti/(Ti + Fe) values along the profile marked on Figure 3a, showing a pronounced chemical zonation with a plateau in the core region and increasing Ti/(Ti + Fe) values toward the outer surface. (c) BSE image of a small region near the surface, showing Ilmss lamellae and their Fe-rich Tmt-host crystals (Tmt3) in the outer 30 μm thick rim, resulting from oxy-exsolution during external quench (see text in sections 2.1 and 3). (d) χ-T curve in the temperature range 77–970 K (black, heating; gray, cooling). (e) Detail of the previous χ-T curve in the temperature range 640–840 K to better show the stepwise drop of the magnetic susceptibility with increasing temperature. The Curie temperatures represented by the vertical bars TC1 and TC2 have been estimated from rim and core compositions (Figure 3b), using polynom 1 in Table 3. The TC3 bar is deduced from approximate compositions of the near-surface, oxidized titanomagnetite (see Figure 3c). (f) Ms-T curve over the temperature range 300–970 K. Curie temperature (TC) retrieved with the method of Moskowitz [1981].

[18] The second type of chemically inhomogeneous titanomagnetites occur in some single-phase samples that were synthesized through gas mixing experiments from starting mixtures containing only Fe2O3 and TiO2 (without metallic iron). The sample pellets present outer shells (up to 200 μm thick) in which the Ti/Fe ratio increases toward the pellet surface (Figure 3b). Whether some samples used in previous magnetic studies may have presented such Ti-enriched rims is a matter of conjecture. It has been known for long that at least double sintering is necessary to minimize chemical heterogeneities in Tmt samples synthesized from Fe2O3-TiO2 mixtures [e.g., Bleil, 1971] and that “it is difficult to achieve a uniform distribution of Ti in the samples” [Kakol et al., 1991, p. 649]. In fact, careful examination of our double-sintered run products have in some cases revealed compositional heterogeneities. In any case, efficient checks of the compositional homogeneity of synthetic titanomagnetites are mandatory but have been reported only in a few papers [e.g., Moskowitz, 1987; Kakol et al., 1991; Wanamaker and Moskowitz, 1994].

4. Magnetic Susceptibility as a Function of Temperature (χ-T Curves)

[19] The χ-T curves (Figure 4) are dominated by the contribution of titanomagnetite, with distinct shapes as a function of the ulvöspinel content (XUsp). At XUsp > 0.9 the curves are flat with only a sharp “peak” (Figure 4a), the right flank of which corresponds to the sharp decrease to paramagnetic susceptiblity at temperatures above the Curie temperature (TC). With decreasing XUsp, the asymmetry of the peak increases (Figure 4b) and at XUsp < 0.7 the low-temperature flank becomes slightly convex and the susceptibility above room temperature increases while TC is shifted to higher temperatures (Figure 4c). Samples with end-member magnetite display a practically constant susceptibility between the Verwey transition temperature (∼120 K, transition between the cubic and the monoclinic structure) and the Curie temperature (∼850 K) (Figure 4d).

Figure 4.

Examples of χ-T curves (magnetic susceptibilities normalized to the room temperature values) for titanomagnetite-bearing assemblages synthesized in the Fe-Ti-O system at 1300°C, with decreasing XUsp from Figure 4a to 4d. Black curves were recorded during heating, gray curves during cooling. Note the peak or shoulder contributed by Ilmss (arrows) in Figures 4b and 4c. In Figure 4c, graphical estimates of the Curie temperature (TC) with the intersecting tangents method of Grommé et al. [1969] and the peak method. Samples IDs are (a) 6F63x34, (b) 6F72x4.4, (c) 6F72x1.5, and (d) 6F100x2.4. Synthesis conditions are given in Table 1.

[20] In samples containing Ilmss with 0.7 < XIlm < 0.85, the contribution of the rhombohedral phase to the χ-T curves is a small peak or a shoulder on the low-temperature flank of the Tmt peak (see arrows in Figures 4b and 4c). The corresponding Curie temperatures for Ilm-Hemss are in the range of 330 to 170 K and decrease with increasing XIlm, in accordance with the results of Ishikawa et al. [1985]. However, precise TC values can only be recorded from samples rich in Ilmss.

[21] The χ-T curves of samples with chemically inhomogeneous titanomagnetite do not exhibit the characteristic sharp decrease of the magnetic susceptibility at temperatures just above TC, but instead either a stepwise drop (Figures 3d and 3e) or a gradual decay of χ in the corresponding temperature range. The χ-T curves with steplike drops are typical of sample pellets that contain only titanomagnetite (single-phase assemblage) and display large Ti-rich outer rims (Figure 3b and section 3). The χ-T curves with a gradual χ decay around TC are registered either from Tmt single-phase samples with less pronounced Ti-rich rims or from run products with an oxidized surface due to external quench. The latter samples, which consist either of single-phase Tmt or of Tmt-Ilmss assemblages, contain few volume percent of Fe-richer titanomagnetite localized at the pellet surface (Figure 3c and section 3). In the following, however, only samples with chemically homogeneous Tmts will be considered.

[22] As can be seen from Figure 4, in many of our χ-T curves the branches above room temperature are nonreversible, i.e., the Curie temperatures are not the same on the heating and on the cooling paths. Nonreversiblity occur in nearly all samples with Tmt in the compositional range 0.1 < XUsp < 0.8, even if the Tmt are chemically homogeneous. The nonreversibility of the χ-T curves will be discussed in more detail in section 5.3.

5. Curie Temperatures Retrieved From χ-T Curves

5.1. Methods for Estimating the Curie Temperature From χ-T Curves

[23] The Curie temperature (TC) defines the transition from ferrimagnetic to paramagnetic ordering. In previous studies, TC determinations from χ-T curves have often been performed with the graphical “intersecting tangents method” of Grommé et al. [1969] where TC is defined by the intersection point of the tangent following the sharp drop of susceptibility with the one following the neighboring base line at higher temperature [e.g., Gonzalez et al., 1997; Harrison and Putnis, 1999a; Kontny et al., 2003; De Wall et al., 2004]. Hauptman [1974] proposed a projection of the χ drop to the zero line. However, Petrovsky and Kapicka [2005] have emphasized that the Curie temperature is not the temperature at which magnetic susceptibility vanishes but, instead, corresponds to a second-order phase transition with a critical point at which the magnetic susceptibility is theoretically infinite. These authors see two options for retrieving TC from χ-T curves. If the susceptibility strongly increases just below TC (often interpreted as Hopkinson peak), as shown in Figures 4a and 4b, the corresponding temperature can be taken as TC (“peak method”). Otherwise, TC can be seen as the lowest temperature at which paramagnetic behavior appears, i.e., at which the Curie-Weiss law holds: χ(T>Tc) = C/(TTC). Consequently, in a 1/χ versus T plot, TC is the lowest temperature end of a linear pattern over a significant temperature range (“1/χ method”). Petrovsky and Kapicka [2005] postulate that the application of the intersecting tangent method on χ-T curves overestimates the Curie temperatures. Indeed, our χ-T measurements with the KLY 4 kappabridge on the standard substances Ni, Fe3O4, Fe2O3 and Fe° show that Curie (or Néel) temperatures are distinctly overestimated with the intersecting tangent method, whereas the peak and the 1/χ methods yield values that are only a few degrees above the accepted literature values (Table 2).

Table 2. Comparison of Estimates of the Curie or Néel Temperatures Obtained With Different Graphical or Extrapolation Methods From χ-T or Ms-T Curves of a Few Standard Substancesa
SubstancesLiteratureEstimates From χ-TEstimates From Ms-T
  • a

    Temperature is given in kelvins. Literature values are from Hunt et al. [1995]). “Grommé” refers to the intersecting tangents method of Grommé et al. [1969], “Moskowitz” to the extrapolation method of Moskowitz [1981]. Ni, spans from a rod (Koch and Light); Fe3O4, single-phase polycrystalline magnetite synthesized from Fe2O3 powder at 1300°C, log fO2 = −6.4 (run product 6F100x2.4); Fe2O3, powder (Alpha Products); Fe, spans from a rod (Koch and Light).

Ni (99.998%)631628–632635636636636
Fe3O4 (6F100x2.4)848–858864865868851852
Fe2O3 (99.9%)948950–953951973952953
Fe° (99.998%)10431039–10431045105210501055

[24] TC values obtained from our samples with the different methods are listed in Table 1. We have not considered the “first derivative” approach (which identifies the peak in the first derivative of the χ-T curve) in our data set because this method yields practically the same results as the peak method. For further discussions we shall only use the values retrieved with the peak method. In case of peaks with a rounded shape, we have taken the onset of the sharp χ drop as TC. In such cases, the 1/χ-T curves do not show a distinct linear portion, i.e., do not allow a good TC estimate, because the transition from magnetically ordered to paramagnetic state does not occur at a discrete temperature, but over a (sometimes rather wide) temperature range (values followed by a superscript j in Table 1). For such samples our TC estimate with the peak method represents the temperature at which the paramagnetic behavior starts to dominate.

[25] In samples with heterogeneous titanomagnetite compositions, which show complex χ-T curves, the determination of TC is not straightforward. The temperature of the first significant decrease in χ appears to reflect the Ti-richest Tmt composition (Figure 3e). In case of a step-like drop at the ferrimagnetic to paramagnetic transition, further abrupt χ decreases may be correlated to some localized compositions (Figure 3), but such correlations are precarious. In case of χ-T curves with a smeared χ decrease, only the peak temperature seems to have some significance. The intersecting tangents method may yield different TC values, but their significance is dubious.

[26] The results listed in Table 1 concern samples with chemically homogeneous Tmts, which produce simple χ-T curves. The curves for Tmt in equilibrium with Wus, however, display step-like features, due to the fact that Wus exsolves Fe-rich Tmt upon quenching (see section 3). With the peak method the TC of the high-temperature Tmt is easily retrieved, but the 1/χ and the intersecting tangents methods yield uncertain results (values followed by a superscript j in Table 1).

5.2. Curie Temperatures as a Function of Tmt Composition, Synthesis Temperature, and Paragenesis

[27] In this section we address only the Curie temperatures that can be determined from the first heating branch of the χ-T curves. As mentioned in section 4, the subsequent cooling branch generally points to different TC values. This nonreversibility will be discussed in section 5.3.

[28] As known from previous studies (see Figure 1), the Curie temperatures of our synthetic titanomagnetites decrease with increasing ulvöspinel content (Figure 5a). Yet our data reveal a number of further interesting effects. For samples synthesized at 1300°C the TC values for Tmt in coexistence with Ilmss are all higher than those of Tmt in coexistence with Wus (Figure 5a). This ΔTC is in the order of 15 to 35 K (Figure 5b). At XUsp = 0.33 ΔTC culminates at 37 K and apparently decreases with increasing XUsp up to about 0.8. At high Ti contents (XUsp > 0.9), the comparison is difficult because Tmt can no longer coexist with wüstite in the Fe-Ti-O system (see Figure 2). Yet the trend persists for these Ti-rich compositions: Tmt synthesized in equilibrium with Ilmss do have higher TC than single-phase Tmt with the same Ti/Fe ratio. The TC values retrieved from Tmt-Ilmss assemblages synthesized at 1100°C are, at any given XUsp, intermediate between those of the two 1300°C series (Figure 5b). For each data group one can establish a second-order polynomial correlation (with excellent correlation coefficients) between the TC value and the ulvöspinel content of the titanomagnetites in the range 0.1 < XUsp < 0.8 (Table 3). At higher Ti contents, the slope of the regression line for Tmt in equilibrium with Ilmss is distinctly flatter (Figure 5a).

Figure 5.

Curie temperatures (TC) of synthetic titanomagnetites in the Fe-Ti-O system as a function of composition (XUsp). Estimates from χ-T curves with the peak method (see section 5.1). (a) Products of syntheses at 1300°C with a second-order regression curve for Tmt in equilibrium with Wus (dashed line). For Tmt in equilibrium with Ilmss, there are two regression curves: at XUsp < 0.8 a second-order curve (thick black curve) and at XUsp > 0.8 a linear regression (thin black line). Equations of the regression curves are given in Table 3. (b) Plot of ΔTC versus XUsp, whereby ΔTC = TC (sample) − TC (regression value for Wus-bearing sample at the same XUsp). The data for XUsp > 0.8 are not represented because the extrapolation of the regression curve for Tmt (+Wus) is not adequate in this XUsp range.

Table 3. Coefficients and Correlation Coefficients of the Regression Polynoms Obtained for the Correlations Between Curie Temperature and Composition of Synthetic Titanomagnetites in the Fe-Ti-O Systema
PolynomSynthetic T, KCompositional RangeCoexisting OxideabcuvwR2
  • a

    Polynoms 1 to 5 are based on TC values retrieved from χ-T curves, polynom 6 is based on TC values retrieved from Ms-T curves. All temperatures are in kelvins. Polynoms of the form TC = aXUsp2 + bXUsp + c and XUsp = uTC2 + vTC + w. Read −4E–7 as −4 × 10−7.

115730.1 < Xusp < 0.8IlmSS−218.195−608.713887.008−4E-7−0.0007941.0333570.999
215730.8 < Xusp < 1.0IlmSS −628.406755.979 −0.0015791.2011580.993
315730.1 < Xusp < 0.9Wus−154.449−650.283865.549−3E-7−0.0009531.0533920.999
413730.4 < Xusp < 0.8IlmSS−222.712−629.299892.454−3E-7−0.0008561.0291150.999
513730.8 < Xusp < 1.0IlmSS −662.873766.905 −0.0014941.1546730.990
615730.1 < Xusp < 0.8IlmSS−400.062−414.955866.621−1E-6−0.0002040.9150240.999

[29] The most possible explanation for the differences between the Curie temperatures of titanomagnetites synthesized in equilibrium with different phases (Ilmss versus Wus), is to assume that the two sets of Tmt have different concentrations of cation vacancies. Tmt synthesized in equilibrium with wüstite should be devoid of cation vacancies and may even contain cation interstitials. In contrast, Tmt in equilibrium with Ilmss are known to present significant concentrations of cation vacancies at 1300°C, but somewhat lower concentrations at 1100°C [e.g., Aragon and McCallister, 1982; Senderov et al., 1993; Aggarwal and Dieckmann, 2002].

[30] In principal, our results confirm those of Hauptman [1974] and Rahman and Parry [1978], who reported a strong increase of the Curie temperatures (of up to 60 K) with increasing vacancy concentrations for two single titanomagnetite compositions (XUsp = 0.6 and 0.3) equilibrated at 1275 or 1300°C. Yet our results show somewhat smaller TC increases (Figure 5). This discrepancy is most probably related to the fact that Hauptman [1974] and Rahman and Parry [1978] oversaw small amounts of Ilmss in their allegedly single-phase titanomagnetite samples. According to the results of Lattard et al. [2005], the four highest oxygen fugacities reported by Hauptman [1974] for the equilibration of a Tmt with XUsp = 0.6 at 1275°C are outside the stability field of single-phase Tmt but instead fall in the binary Tmt-Ilmss field (see Figure 2). Consequently, the corresponding samples of Hauptman [1974] probably contained some percent of Ilmss and the coexisting Tmt had a lower XUsp and hence a higher TC than expected. Some of the χ-T curves presented by Hauptman [1974] give support to our hypothesis. They have a step-like form (curves 5 and 6 in Hauptman's Figure 6) which suggest juxtaposed compositions of titanomagnetite. Concerning the data of Rahman and Parry [1978], the highest fO2 presented for Tmt with XUsp = 0.3 also falls in the binary Tmt-Ilmss field delineated at 1300°C by Lattard et al. [2005]. Moreover, Rahman and Parry [1978, p. 233] report that their thermomagnetic curves “in some cases suggest the occurrence of a higher Curie point transition” whereby the “maximum amount of this higher Curie point component …was about 5% by volume.”

[31] In conclusion, our data set definitely shows a range of Curie temperatures for titanomagnetites with the same Ti/Fe ratio, whereby the lowest TC is registered for Tmt coexisting with Wus, i.e., the Tmt with the lowest possible vacancy concentration, and the highest TC for Tmt coexisting with Ilmss, i.e., those with the highest possible vacancy concentration.

5.3. Nonreversibility of the χ-T Curves

5.3.1. Experimental Results

[32] Nearly all titanomagnetites with TC values above room temperature present χ-T curves with a conspicuous nonreversibility of the Curie temperature. In most samples the strong change in magnetic susceptiblity marking the Curie temperature is shifted to higher temperature in the χ-T pattern recorded on the cooling path compared to that of the preceding heating path (Figure 4).

[33] The difference (ΔTC) between the two Curie temperatures obtained from the cooling and from the heating branch, respectively, is dependent not only upon the Tmt composition, but also upon the synthesis temperature and the coexisting oxide phase. Tmts coexisting with Wus have the lowest ΔTC, with nearly reversible χ-T curves for Fe-rich compositions, but a slight nonreversiblity (ΔTC < 15 K) for compositions in the XUsp range 0.4 to 0.8 (Figure 6). Tmts in equilibrium with Ilmss at 1300°C present the strongest nonreversibility, which culminates at XUsp ≈ 0.5, with ΔTC up to 40 K. For one Fe-rich composition (XUsp = 0.169), however, the TC value on the cooling branch is lower than that on the heating branch (ΔTC = −6 K). Tmts synthesized in equilibrium with Ilmss at 1100°C have much lower ΔTC than those synthesized at 1300°C (Figure 6).

Figure 6.

Difference (ΔTC) between Curie temperatures obtained from the cooling and from the heating branch of χ-T curves as a function of titanomagnetite composition (XUsp).

[34] Further tests show that the nonreversiblity is also influenced by the maximum heating temperature during the χ-T measurements. For example, in case of a Tmt with XUsp = 0.7 synthesized at 1300°C in equilibrium with Ilmss, the χ-T curve is reversible if the maximum heating temperature does not exceed 500 K. The nonreversibility increases with increasing maximum heating temperature and reaches a plateau value for maximum heating temperatures above 700 K (Figure 7a). In case of a Tmt with practically the same XUsp, but synthesized in equilibrium with Wus, nonreversibility appears only if the heating temperature exceeded 700 K (Figure 7b). Repeated heating-cooling cycles from room temperature to 970 K lead to converging heating and cooling branches, i.e., the nonreversibility disappears and the Curie temperature stabilizes at a value near that of the original TC on the cooling branch (Figure 8).

Figure 7.

Curie temperatures of titanomagnetites with XUsp ≈ 0.7, retrieved with the peak method from the heating (solid symbols) and from the cooling branches (open symbols) of χ-T curves as a function of the maximum heating temperature during the χ-T measurements. (a) Tmt synthesized at 1300°C in equilibrium with Ilmss (sample 6F72x4.4). (b) Tmt synthesized at 1300°C in equilibrium with Wus (sample 6F83ax49).

Figure 8.

The χ-T curves recorded during three succeeding heating-cooling cycles (maximum heating temperature 970 K) from the same sample piece (Tmt + Ilmss, XUsp = 0.55; Sample 6F69x1.5). Black curves are for heating; gray curves are for cooling.

[35] The nonreversible behavior of the χ-T curves may be related to changes in the chemical composition, either related to oxidation or to unmixing of Tmt, or in the degree of cation order of titanomagnetite. We shall examine these possibilities in sections

5.3.2. Is the Nonreversibility due to Oxidation?

[36] The increase of the Curie temperature on the cooling branch could point to oxidation during the χ-T measurements yielding either titanomaghemite or a Fe-richer Tmt with concomitant formation of Ilmss lamellae (oxy-exsolution). In our samples, the formation of titanomaghemite can be ruled out because this phase is known to develop only from very fine grained Tmt (typically ball milled) oxidized in air at temperatures below 573 K and to quickly breakdown at higher temperatures [e.g., Moskowitz, 1981]. In contrast, our χ-T measurements have been performed on relatively coarse-grained assemblages under flowing Ar atmosphere and the curves were reversible when the maximum heating temperature did not exceed 573 K (Figure 7).

[37] Systematic SEM examination of all samples measured under flowing Ar have shown no typical oxy-exsolution “trellis” texture, with very fine lamellae of Ilmss in the {111} planes of magnetite-rich Tmt [e.g., Buddington and Lindsley, 1964; Haggerty, 1991]. EMP analyses have also revealed no change in the chemical composition of the Tmts before and after the χ-T measurements. The only exception was a sample, which was measured three times, i.e., with three consecutive heating-cooling cycles from room temperature to 970 K (χ-T curves in Figure 8). After the third cycle, the sample presented oxy-exsolution trellis textures in the near-surface region (rim of about 30 to 40 μm in depth).

[38] In contrast, single χ-T measurements performed in air do cause near-surface oxidation, resulting in trellis textures in the outer rim of the sample. This has been observed on one Tmt-Ilmss sample (XUsp = 0.73, XIlm = 0.82). The cooling branch of the corresponding χ-T curve is distinctly different from those retrieved from our usual measurements in flowing Ar. Not only is the χ drop shifted to higher temperatures, but the curve also presents a long flat tail ending at temperatures around 840 K, which points to small amounts of magnetite-rich Tmt. Indeed, oxidation of such a Fe-Ti oxide assemblage at temperatures below 900 K should produce Fe-rich Tmt [e.g., Buddington and Lindsley, 1964].

[39] To summarize, we have no indication for oxidation of our samples during single χ-T measurements under Ar flow.

5.3.3. Is the Nonreversibility due to Subsolvus Unmixing of Titanomagnetite?

[40] In principal, the nonreversibility of the χ-T curves could also be related to unmixing of the original high-temperature Tmts into intergrowths of a Fe-rich and a Fe-poorer spinel, caused by subsolvus exsolution during the χ-T measurements. Harrison and Putnis [1996] have documented such processes in the magnetite-spinel solid solution. The situation for titanomagnetite is, however, somewhat different, because the consolute point of the miscibility gap probably lies below 870 K or even 770 K [Vincent et al., 1957; Lindsley, 1981; Price, 1981; Lindsley, 1991], i.e., at lower temperatures than the maximum heating temperature during our χ-T measurements (970 K). Therefore unmixing of Tmt during these measurements appears unlikely.

[41] In any case, the observed nonreversiblity of single-phase end-member magnetite (Fe3O4) cannot be related to unmixing (Run product 6F100x2.4 in Table 1 and Figures 4 and 6). In a Fe-rich Tmt, for instance with XUsp = 0.169, unmixing would be expected to produce partial exsolution of small amounts of a Ti-richer Tmt, whereas the composition of the host Tmt would become somewhat Fe-richer. Consequently, the cooling branch of the χ-T curve should be shifted toward higher temperatures, but the contrary happens (see Figure 6). For Ti-rich titanomagnetites, unmixing would be expected to produce partial exsolution of small amounts of a Fe-richer Tmt, whereas the composition of the host Tmt would become slightly Ti-richer. Consequently, we would expect cooling branches of the χ-T curves with steps or gradual decays, such as those depicted in Figure 3 for inhomogeneous Tmts. However, such features do not occur in case of the samples listed in Table 1 (see Figure 4). The cooling branches of the χ-T curves may be somewhat smoother around TC (Figures 4b, 4c, and 8), but otherwise they closely resemble the heating branches and do not show indication for the appearance of different Tmt compositions during the χ-T measurements.

[42] Very fine grained Tmt exsolution lamellae could be expected to have single domain behavior and display a pronounced Hopkinson peak on the cooling branch of the χ-T curve [e.g., Harrison and Putnis, 1996, Figure1b], but no such peak appeared in the cooling branches of our χ-T curves. In total, we do not see any indication for subsolvus unmixing of Tmt during the χ-T measurements.

5.3.4. Cation Distribution and Nonconvergent Ordering

[43] Since changes in Tmt compositions, in relation to oxidation or unmixing, do not appear to be responsible for the nonreversibility of the χ-T curves, only changes in nonconvergent cation order can be further considered. We present here first a brief outline of the theory about temperature-dependent cation distribution in titanomagnetite and the kinetics of cation reordering.

[44] The stoichiometric end-members of the titanomagnetite solid solution, magnetite and ulvöspinel, are known to have the inverse spinel cation distribution at room temperature. Their structural formula can be written as (Fe3+)tet [Fe3+ Fe2+]oct O4 for the magnetite end-member and (Fe2+)tet [Fe2+ Ti4+]oct O4 for the ulvöspinel end-member. At elevated temperatures, ulvöspinel is known to keep its inverse configuration [e.g., Wechsler et al., 1984], whereas in magnetite cations become progressively disordered over both tetrahedral and octahedral sites (nonconvergent disordering) with increasing temperature [e.g., Wu and Mason, 1981; Wiβmann et al., 1998]. The structural formula for a high-temperature magnetite with a random cation distribution is (Fe0.673+ Fe0.332+)tet [Fe1.333+ Fe0.672+]octO4.

[45] On the basis of thermoelectric coefficient measurements and of the thermodynamic study of O'Neill and Navrotsky [1984], Trestman-Matts et al. [1983] have developed a cation distribution model for the titanomagnetite solid solution as a function of equilibration temperature. Titanium is assumed to occupy only octahedral sites, in accordance with magnetic and neutron diffraction data [e.g., O'Reilly and Banerjee, 1965; Stephenson, 1969; Ishikawa et al., 1971; Wechsler et al., 1984]. Calculations with the model of Trestman-Matts et al. [1983] show, as expected, that in all Tmt at equilibrium nonconvergent disordering progresses with increasing temperature, i.e., the Fe2+ content in octahedral sites decreases and the one in tetrahedral sites increases (Figure 9a). On the basis of the model of Stephenson [1972] these changes translate into a decrease of TC with increasing equilibrium temperature (Figure 9b). For a Tmt with XUsp = 0.7, for example, the Curie temperature is predicted to be 342 K if the cation distribution equilibrated at 1573 K, but 371 K if the cation distribution could equilibrate at 293 K (Figure 9b). With decreasing XUsp down to 0.4, the effect of the equilibrium temperature on the cation distribution become stronger, hence the differences in TC steadily increase. At XUsp < 0.4, however, the trend reverses (Figure 9c).

Figure 9.

Relationships between Curie temperature and modeled cation distributions in Tmt. (a) Temperature-dependent cation distribution at equilibrium for XUsp = 0.7 calculated with the model of Trestman-Matts et al. [1983]. Thin vertical bars mark the temperatures at which perfect inverse and random distributions are predicted; short vertical bars mark the Curie temperatures extracted from the heating and cooling branches of the χ-T curves. (b) Curie temperatures calculated with the model of Stephenson [1972] for the equilibrium cation distributions depicted in Figure 9a. (c) Differences (ΔTC) between the calculated TC values for cation distributions at equilibrium at 1573 K (1300°C) and those for equilibrium at lower temperatures for various Tmt compositions. For 0.7 > XUsp > 0.3, ΔTC continuously increases with decreasing XUsp (solid curves). At lower XUsp the trend reverses (dashed curves). The black bar represents the equilibrium temperature corresponding to the quenched-in cation degree of order after synthesis at 1573 K, the gray bar represents the approximate relaxation temperature during the χ-T measurements. The thick parts on the curves are the ranges of TC values recorded from the heating and cooling branches of the χ-T curves.

[46] Yet the kinetics of cation ordering in Tmt plays a dominant role in the interpretation of our results. On the basis of the observation that unit cell parameters and/or magnetic properties of synthetic Tmts quenched from different temperatures are identical [e.g., O'Donovan and O'Reilly, 1980; Wechsler et al., 1984], the cation distribution established in titanomagnetite at high synthesis temperatures is reputed unquenchable. Indeed, since reequilibration requires only the transfer of an electron between Fe cations in adjacent tetrahedral and octahedral sites, its kinetics must be very rapid [Jensen and Shive, 1973]. However, as discussed in detail by Harrison and Putnis [1999b], not only the quench temperature but also the quench rate determines the quenched-in degree of order. For quench rates usual in experimental studies, Harrison and Putnis [1999b] suggest for Tmt that the closure temperature, i.e., the quench temperature above which the quenched-in degree of order remains constant, is less than 873 K. The quenched-in degree of order should correspond to an equilibrium order at still lower temperature, in the order of 750 to 800 K [see Harrison and Putnis, 1999b, Figure 16]. This would hold for all our synthetic samples.

[47] During the χ-T measurements, which involve heating and cooling up to 970 K at a rate of 10 K/min, rapid changes in cation ordering may appear above a characteristic relaxation temperature and the ordering state may even reach its equilibrium value, which will be metastably kept on cooling to room temperature [cf. Harrison and Putnis, 1996]. Whether such changes in cation ordering lead to a shift of the Curie temperature on the cooling branch of the χ-T curve or not, depends on the relative position of the Curie temperature and the relaxation temperature.

[48] Let us first consider a titanomagnetite synthesized in equilibrium with Ilmss with XUsp = 0.7 and a nonreversible TCTC ≈ 20 K). The χ-T measurements performed with different maximum heating temperatures suggest a relaxation temperature of about 600 K (Figure 7a). During the first heating of a χ-T measurement, the TC value registered at 354 K (value interpolated from our experimental data; see polynom 1 in Table 3), i.e., well below the relaxation temperature, must represent the quenched-in degree of order. Upon heating and subsequent cooling above the relaxation temperature at a rate of 10 K/min, the cation distribution might reach the equilibrium values depicted in Figure 9a. The higher TC value recorded during cooling (about 374 K) probably reflects the equilibrium degree of order established at the relaxation temperature (about 600 K) and preserved during further cooling. The model of Stephenson [1972] predicts a higher TC value on the cooling branch (Figures 9b and 9c), but the difference is much lower than the nonreversiblity observed in our χ-T measurements (only 3 to 4 K compared to the observed 20 K; Figures 9c and 6). Harrison and Putnis [1999b] have also noted that Stephenson's model strongly underestimates the nonreversiblity. Yet this model is useful to explain the increase in nonreversibility with decreasing XUsp down to 0.4. As shown in Figure 9c, the differences between the TC values corresponding to the equilibrium order at about 600 K (relaxation temperature) and those corresponding to the degree of order at about 750–800 K (i.e., quenched-in orders from syntheses) significantly increase with decreasing ulvöspinel content, at least down to XUsp = 0.4 (Figure 9c).

[49] For Tmts with XUsp ≤ 0.4 the situation changes because the Curie temperature is higher than the relaxation temperature during the χ-T measurements (Figure 9c). Above the relaxation temperature, the measured TC values mirror the equilibrium degree of order and the χ-T curves are more or less reversible. A particular case is represented by a sample with XUsp = 0.2, which shows a lower Tc value on the cooling than on the heating path (Figure 6). Such a behavior has been reported by Harrison and Putnis [1999b] (see their Figure 19) for a spinel with the same mole fraction of the magnetite end-member in the Fe3O4–MgAl2O4 solid solution. Their interpretation is that for this composition TC falls within the temperature range of the ordering hysteresis.

[50] As shown in Figure 6, Tmts synthesized in equilibrium with Wus over the whole compositional range have χ-T curves with only slight nonreversibility. This conspicuous difference to the behavior of Tmts coexisting with Ilmss is most probably related to the different vacancy concentrations in the two series of Tmt. As already discussed in section 5.2, Tmts coexisting with Wus must have the lowest vacancy concentration for any given Ti/Fe ratio and T-fO2 synthesis conditions. Consequently they are expected to only reluctantly change their cation distribution during the χ-T measurements. Indeed, the relaxation temperature for a Tmt with XUsp = 0.73 synthesized in equilibrium with Wus appears to be about 200 K higher than that for a Tmt with the same XUsp synthesized in equilibrium with Ilmss (Figures 7a and 7b).

[51] On the same line of reasoning, the smaller nonreversiblity of Tmt synthesized in equilibrium with Ilmss at 1100°C (compared with 1300°C; Figure 6) can be related to the inverse correlation between vacancy concentration and synthesis temperature [Aggarwal and Dieckmann, 2002].

6. Curie Temperatures Retrieved From Ms-T Curves

[52] Since the majority of the literature data on the compositional dependence of TC of titanomagnetite have been obtained from temperature-dependent saturation magnetization (Ms-T) curves, we have also measured, for comparison, the thermomagnetic curves of several of our Tmt-bearing synthetic samples.

6.1. Methods for Estimating the Curie Temperature From Ms-T Curves

[53] Theoretically, TC is the temperature at which the spontaneous magnetic alignment vanishes and paramagnetic behavior dominates. In practice, TC has often been estimated with the graphical “intersecting tangents method” of Grommé et al. [1969]. Moskowitz [1981] has shown that this method yields too high TC values in case of complex, irregular Ms-T curves and has proposed to extrapolate the curves in the temperature range from TC-100 to TC with a square root function. Although our samples generally exhibit regular, continuous Ms-T curves, we have used the latter method in the version implemented by Leonhardt [2006]. The corresponding TC values are listed in Table 1. Measurements on the standard substances Ni, Fe2O3 and Fe yield TC values a few degrees above the accepted literature values [cf. Hunt et al., 1995]; those on a synthetic Fe3O4 are within the range of literature values (Table 2).

6.2. Nonreversibility of the Ms-T Curves

[54] Just like the χ-T curves, most Ms-T curves are nonreversible, because their cooling branch is shifted to higher temperatures compared to their heating branch. For pure magnetite the Ms-T curve is reversible, but with increasing XUsp the shift continuously increases (Table 1). In contrast to the χ-T curves, the cooling and heating branches of the strongly nonreversible Ms-T curves are not parallel. Instead, the cooling branches present flat “tails” toward high temperatures (Figure 10) and all cut the zero line at T between 800 and 850 K. The latter feature points to a small amount of magnetite-rich Tmt, which results from partial oxidation at the surface of the samples during the Ms-T measurements. Indeed, BSE images of all samples after the Ms-T measurements reveal trellis oxy-exsolution textures in the near-surface regions, whereas no sign of oxidation is recognizable on the samples prior to the Ms-T measurements. If the maximum heating temperature during the measurements did not exceed 770 K, no oxy-exsolution could be detected and the cooling branch of the Ms-T did not show any high-temperature tail. Apparently, oxidation becomes effective only at temperatures above about 800 K and is thus discernible only on the cooling branch of the Ms-T curves. One may wonder why surface oxidation occurred although the measurements were performed in Ar atmosphere. Apparently, the glass wool placed around the sample to keep it in place within the VFTB hinders the flow of Ar.

Figure 10.

Comparison of temperature-dependent magnetic susceptibility (χ-T) and magnetization curves (Ms-T and J-T) measured on two chunks of the same titanomagnetite sample (Tmt + Ilmss; sample 6F72x1.5). Ms-T curve measured with a field of 5060 Oe (0.63 T), J-T curve with a field of 40 Oe, χ-T curve with 3.8 Oe. Black curves for the heating branches; gray curves for cooling branches.

6.3. Comparison With the TC Values Retrieved From χ-T Curves

[55] The TC values obtained from the heating branches of the Ms-T curves are listed in Table 1. Most of them are higher than those obtained from the heating branch of the χ-T curves, whereby the difference ΔTC = TC(Ms) − TC(χ) regularly and strongly increases (up to nearly 40 K) with increasing Ti content of the Tmt, at least up to XUsp = 0.6 (Figure 11). At higher Ti contents no comparison is possible because the VFTB does not allow measurements below room temperature. In contrast to the ΔTC calculated from the nonreversible χ-T curves (Figure 6), the present ΔTC (Figure 11) does not so strongly vary with the synthesis temperature or with the synthetic assemblage (+Ilmss or +Wus). This suggests that this ΔTC is not significantly influenced by the concentration of cation vacancies.

Figure 11.

Difference between the Curie temperatures estimated from Ms-T and χ-T curves as a function of titanomagnetite composition (XUsp).

[56] The large ΔTC values cannot be explained by differences in the temperature records in the two apparatuses. Measurements on standard substances yielded similar TC values with the χ-T and the Ms-T methods with a maximum ΔTC around 10 K, but no systematic change with temperature (see Table 2).

[57] A direct comparison of the Ms-T and χ-T curves for a sample with a Tmt of intermediate composition, together with a thermomagnetic curve (J-T) measured with a relatively low field intensity (40 Oe, instead of H = 5060 Oe, i.e., 0.63 T, for saturation) makes clear why these curves yield so different TC values (Figure 10). In the χ-T curve, the initiation of the sharp, linear drop in susceptibility (see arrow in Figure 10b) is taken to define the Curie temperature. A similar linear drop occurs in the J-T curve and could be considered to point to the same TC value. The concave tails of both curves at higher temperatures are not taken into account for the TC determination. In contrast, the heating branch of the Ms-T curve gradually changes from a convex to a concave form with increasing temperature and the extrapolation to the Ms = 0 value [after Moskowitz, 1981] includes part of the concave tail or the curve, yielding a higher TC value. Yet it is not clear what the concave tails of the heating branches of the Ms-T curves represent. As discussed in section 6.2., they do not appear to reflect oxidized portions of the samples.

[58] At present, we can only keep in mind that the temperature-dependent measurements of the saturation magnetization and of the magnetic susceptibility yield different TC values for titanomagnetite, with increasing divergence with increasing titanium content up to XUsp values of about 0.7. This discrepancy might be related to the vastly different field intensities applied for the Ms-T versus the J-T and χ-T measurements (high DC field for Ms-T measurements, low AC field for χ-T measurements), which might influence cation reordering.

7. Comparison With Literature Data

[59] As shown in Figure 5, our results constrain a much narrower range of Curie temperatures at any given XUsp than the literature data compiled in Figure 1. Part of the scattering in the literature data is most probably related to chemical inhomogeneities in the titanomagnetite samples, in particular due to surface oxidation during quenching or to the formation of Ti-richer outer zones during gas mixture experiments (see section 3.). Some alleged single-phase samples may have contained small amounts of Ilmss so that the true Tmt composition was richer in Fe than assumed (see section 5.2.). Such features, which can only be detected through careful examination with the SEM, were probably overlooked in several former studies, in particular in those performed in the sixties and the seventies of the last century.

[60] As discussed in section 5.2., the other essential reason for the scatter of the literature data may be seen in the use of single-phase titanomagnetites with unknown, and possibly variable vacancy concentrations. Our results strongly suggest a positive correlation between TC and the vacancy concentration at given Ti/Fe value, in principle agreement with Hauptman [1974] and Rahman and Parry [1978]. However, the span in the TC values at any given Ti/Fe does not exceed 40 K (Figure 5), i.e., is distinctly smaller than proposed in the former studies.

[61] The third factor that can explain discrepancies between our data and those of the literature is the type of measurement employed to retrieve TC, i.e., either Ms-T or χ-T measurements. As discussed in section 6.3., Ms-T measurements yield higher TC values for all titanomagnetite compositions in the range 0.2 < XUsp < 0.8 (Figure 11). At 0.6 < XUsp < 0.8 our TC values obtained from Ms-T curves perfectly match those of Bleil [1973] (Figure 12), which are also based on saturation magnetization measurements. At lower XUsp, however, the results of Bleil [1973] are at lower TC values than ours (Figure 12). Our values were measured on Tmts synthesized with Ilmss at 1300°C, those of Bleil [1973] concern single-phase Tmts synthesized either at 1300° or 1100°C. Considering that Bleil [1973] employed lower fO2 conditions than ours for his syntheses at XUsp < 0.6, his single-phase Tmts most probably had lower vacancy concentrations than our Tmts (+Ilmss) and, consequently, lower TC values.

Figure 12.

Compositional dependence of the Curie temperature of synthetic titanomagnetites in the Fe-Ti-O system. Comparison of the regression curves obtained from samples synthesized at 1300°C or 1100°C in the present study with the data points of Bleil [1973]. The dash-dotted curve is based on Ms-T measurements; all other curves are based on χ-T measurements.

8. Summary and Implications for Natural Titanomagnetites

[62] On the basis of our experimental results in the Fe-Ti-O system, we have shown that there are well-defined inverse correlations between the Curie temperature of titanomagnetites and their ulvöspinel end-member content (Table 3), with significant differences between the TC values obtained from Tmt synthesized at 1300°C either in equilibrium with Ilmss (max. concentration of cation vacancies) or with Wus (min. concentration of cation vacancies). The TC values for Tmt synthesized at 1100°C in equilibrium with Ilmss, which plot between the two 1300°C series (Figure 12), represent a good proxy for basaltic parageneses that were rapidly cooled and not oxidized by deuteric or hydrothermal fluids, i.e., essentially for Tmt in submarine pillows and lavas flows. Our calibration curve for 1100°C could represent an upper limit for the TC values, because natural Tmts may have lower vacancy concentrations, in particular if they reequilibrated at lower temperatures in rocks devoid of Ilmss. Nonmagnetic elements such as Mg or Al in natural Tmts are also expected to lower the Curie temperature [e.g., Richards et al., 1973; O'Donovan and O'Reilly, 1977; Özdemir and O'Reilly, 1978, 1981; Moskowitz, 1993]. Preliminary own results on Mg- and Al-bearing titanomagnetites, however, confirm that in concentrations relevant to basalts the lowering effect is only slight. Furthermore, natural Tmts were most likely not as rapidly quenched as in the laboratory, that is, their cation distributions may be more ordered than those of our synthetic samples, resulting in reversible χ-T curves with slightly higher TC values corresponding to those derived from our cooling branches. Indeed, χ-T curves of natural Tmts in submarine basalts, are commonly reversible, if measured in Ar flow [e.g., Kontny et al., 2003; Vahle and Kontny, 2005].

[63] In total, our calibration curve for Tmts synthesized at 1100°C in equilibrium with Ilmss should yield good estimates of the Tmt composition from Curie temperatures in the range 250 to 600 K. For higher TC values, we recommend the regression curve obtained for Tmts synthesized at 1300°C in equilibrium with Ilmss (Table 3), because we have no data from samples synthesized at 1100°C with Xusp < 0.4.

[64] Since we have shown that Ms-T measurements yield higher TC values than χ-T measurements, for Tmt compositions with XUsp > 0.2, we recommend not to use calibration curves based on Ms-T measurements [e.g., Bleil, 1976; Bleil and Petersen, 1982, p.330; Clark, 1997; Hunt et al., 1995] to estimate Tmt compositions from TC retrieved from χ-T measurements. These calibration curves yield overestimates of the ulvöspinel content of titanomagnetites in the range 0.5 < XUsp < 1.0. With the equation given by Bleil and Petersen [1982], the overestimate is of about 0.05, i.e., 5 mol% of the ulvöspinel content (Figure 12). Admittedly, this is not a very strong effect.

[65] In fact, the main uncertainty in estimating the composition of natural Tmts from their Curie temperature stems from the method used to retrieve TC from the χ-T curves (peak versus intersecting tangents method), as already mentioned by Petrovsky and Kapicka [2005]. The χ-T curves reported for Tmt in submarine pillow basalts commonly show a gradual decay of χ around TC, which reflects the chemical inhomogeneities within and among the titanomagnetite crystals [e.g., Kontny et al., 2003; Vahle and Kontny, 2005]. As an example, we may quote sample SR0800-0.9 [Vahle and Kontny, 2005] from a massive basaltic pillow (HSDP-2 drill hole in Hawaii), which yields a χ-T curve with a peak at 301 K and a gradual drop of susceptibility, ending at 833 K. With our regression curve for Tmt synthesized in equilibrium with Ilmss at 1100°C (polynom 4 in Table 3), we obtain a XUsp value of 0.74 from the peak TC, which should represent the Ti-richest and dominating Tmt composition in the sample. EMP analyses yield, in excellent agreement, a maximum XUsp of 0.75 (C. Vahle, personal communication, 2006) (XUsp value calculated after Stormer [1983]). Vahle and Kontny [2005] have estimated two TC values with the intersecting tangents method, one major value at 383 K, and a minor value at 486 K (see their Table 1 and Figure 3a), which would translate into XUsp values of 0.66 and 0.54 with our equation. It is clear that these values represent only rough compositional “means” which must not occur in any crystal. In fact, EMP analyses of relatively large Tmt crystals give XUsp values ranging between 0.62 and 0.75. Titanomagnetite crystals that were too small to be analyzed with the EMP are Fe richer.

[66] On the whole, the peak value in a χ-T curve of a titanomagnetite-bearing basalt can yield a reliable estimate of the Tmt composition by using an appropriate regression equation (see Table 3), if all titanomagnetite crystals have very similar compositions. If titanomagnetite is chemically inhomogeneous (i.e., zoned crystals and/or crystals of different compositions over the sample) the peak value from the χ-T curve will give a good estimate of the Ti-richest composition. In contrast, TC values retrieved with the intersecting tangents method give only approximate mean compositions of the titanomagnetite in the sample.


[67] The Deutsche Forschungsgemeinschaft (DFG) funded this research through grants La 1164/5-1 and 5-2 within the framework of the International Continental Drilling Program (ICDP). We thank Carsten Vahle (Heidelberg) for help in the rock magnetic laboratory and for discussions, Hans-Peter Meyer for maintenance of the EMP and the SEM, and Ilse Glass and Aleksander Varichev for help at the SEM. Philipp Antrett, Ramona Langner, and Alexandra Vackiner assisted in the laboratories at different stages of the study. We are indebted to David Krása, Jürgen Matzka, and Nikolai Petersen (Munich) for fruitful discussions and assistance during the saturation magnetization measurements. We thank Eduard Petrovsky (Prag) and, especially, Richard Harrison (Cambridge) for their thorough and very helpful reviews and Wyn Williams for his editorial work.