Introduction

In the past two decades, remarkable technical advances have been made in pressure cell technology allowing researchers to carry out high-pressure investigations in the fields of chemistry, biochemistry, earth and planetary sciences and condensed matter physics. High pressure is used both in the laboratory and on an industrial scale to produce, for example, artificial diamonds, new superconductors and new forms of matter1,2. Pressure is also used to drive materials into new electronic states. Under high pressure, some materials become superconductors, others undergo magnetic phase transitions, and others undergo metal–insulator phase transitions3,4. In magnetic pyrochlore oxides, pressure has been shown to freeze the spin-liquid ground state of Tb2Ti2O7 (ref. 5). Pressure is therefore an important weapon in a researcher's arsenal for exploring phase space.

The canonical spin ices, Ho2Ti2O7 and Dy2Ti2O7, with magnetic Ho or Dy ions6,7,8,9, are part of the pyrochlore family of oxides of general formula A2B2O7 (ref. 10). They have face-centred cubic (fcc) lattice constants of afcc≈10.1 Å. These materials are not very compressible, and studies with high physical pressure have not revealed any significant modification of the spin ice properties11. Significantly reduced lattice constants can be obtained in principle by replacing Ti4+ with a smaller B ion, such as Ge4+, but it is found that at ambient pressure the pyrochlore structure is only stable, if the ratio of the ionic radii, ρ=rA/rB, is less than 1.55 (ref. 10). Dy2Ge2O7 and Ho2Ge2O7, with ρ≈1.8, adopt a tetragonal structure, when synthesized under ambient pressure12. The range of stability of the cubic pyrochlore form can be increased using a high pressure, high temperature synthesis, which extends the regime of stability beyond ρ≈1.8 (ref. 13). The cubic pyrochlore form of Dy2Ge2O7 prepared in this way has a lattice constant of 9.9290 Å, equivalent to a canonical spin ice under enormous physical pressure11.

The microscopic 'dipolar spin ice' Hamiltonian of Ho2Ti2O7 and Dy2Ti2O7 includes complex dipolar and superexchange interactions14,15. However, to a good approximation, it may be represented by a much simpler spin Hamiltonian that is equivalent to the original spin ice model6,7. In this description9,14, there are three parameters: the lattice constant afcc, the rare earth magnetic moment μ, and an effective near-neighbour exchange parameter Jeff. The equilibrium statistical mechanics then maps onto the statistical mechanics of idealized water ice, such that the low-temperature magnetic state is equivalent to the proton disordered state of water ice H2O (refs 6,7), and shares with it the Pauling configurational entropy8. The spin ice state is thus equivalent to pure H2O, and its excitations are equivalent to the ionic defects H3O+ and OH (refs 6,7,16,17). The success of this 'near-neighbour spin ice' description may be attributed to the almost perfect self-screening of the dipole–dipole interaction between rare earth moments in the effective ground state14.

Although dipole–dipole interactions may be ignored in the spin ice ground state, in ref. 17 it was shown that inclusion of the dipole–dipole interaction in the excited states causes the 'ionic defects' of spin ice to behave as magnetic charges that interact via the magnetic Coulomb law. The description of these defects as magnetic monopoles was firmly established in ref. 16 by approximating the microscopic spin Hamiltonian Ho2Ti2O7 and Dy2Ti2O7 to a 'dumbbell model', where finite dipoles replace spins. The dumbbell model approximately restores dipolar corrections that are integrated out in the near-neighbour description, but still retain three parameters16. These are the monopole 'contact distance', (the lattice constant of the diamond lattice inhabited by the monopoles), the elementary monopole charge Q=2μ/a (refs 16,17) and the monopole 'self-energy' ν, that replaces Jeff (ref. 16). In ref. 18, the self energy ν was equated with a monopole chemical potential in the grand canonical ensemble.

The monopole system is a magnetic Coulomb gas of deconfined monopoles and antimonopoles with overall charge neutrality, which closely approximates a magnetic electrolyte ('magnetolyte') in the grand canonical ensemble18,19,20,21,22,23,24. Accordingly, experiments on the canonical spin ices reveal strong evidence of the standard field response of such a system, the Wien effect19,23,24, as well as of the applicability of Debye-Hückel theory in zero applied field21. In the magnetolyte description of spin ice, the scale of length may be set by the contact distance a and the scale of energy may be set by the Coulomb energy at contact, μ0Q2/4πa. Different spin ices—that is, different triplets {Q, a, ν}—should have identical monopole interaction potentials if energies and lengths are scaled by the above characteristic quantities—so called 'corresponding states' behaviour. Thus, the zero-field magnetolyte properties should be fully controlled by the dimensionless temperature T*(=4πkBTa/μ0Q2) and the dimensionless monopole density per lattice site x (ca3, where c is the concentration). These two parameters, T* and x, map out a phase space which, as mentioned above, is expected to be surprisingly rich.

The xT* phase behaviour of spin ice has not been determined in detail, but by analogy with electrolyte models25,26,27, we would expect a gradual transition from a weakly correlated magnetolyte at relatively large T*/x to a strongly correlated magnetolyte at small T*/x (Fig. 1). However, in a given spin ice, T* and x cannot be independently varied as x is a function of T* and the chemical potential ν (Methods). Thus, any one spin ice material maps out a single trajectory in the space of x and T*, and existing spin ices are found to be firmly in the weakly correlated regime (Fig. 1). In this regime, the fraction of bound monopole-antimonopole pairs is sufficiently small that it may be neglected for most purposes (the field response being an exception: see ref. 23). Considering, for example, Dy2Ti2O7, the chemical potential, ν, is found to be −4.35 K (ref. 28), which consists (in magnitude) of half the energy required to create a (+−) contact pair, ɛpair/kB≈5.7 K, plus half the energy required to unbind the pair, μ0Q2/4πakB≈3 K. As the dipole (+−) pair is much higher in energy than the individual (+ or −) charges, the pairing tendency is weak. In contrast, it can be seen that ν≈3 K is the chemical potential that puts monopole–antimonopole pairs or (hetero-)dimers at the same free energy as free monopoles, and thus marks the boundary between the weakly and strongly correlated regimes (Fig. 1).

Figure 1: Creating strongly correlated magnetic monopoles in spin ice.
figure 1

A corresponding states diagram for spin ice in terms of reduced temperature T* and monopole density, x. A given spin ice material maps out a single trajectory but the canonical spin ices such as Dy2Ti2O7, Ho2Ti2O7 lie in the weakly correlated regime (green), rather than in the strongly correlated regime (blue). By high-pressure synthesis, we have created a new spin ice Dy2Ge2O7 that lies on the boundary of strong and weak correlation (red line), and hence has significant monopole dimerization at all measured temperatures (we have also created Ho2Ge2O7 but this lies in the weakly correlated regime). The strongly correlated regime (lower right) has monopole correlations beyond simple pairing, potentially leading to a gas-liquid transition or charge ordering.

The only way to experimentally approach the strongly correlated regime in Figure 1 is therefore to change the chemical potential ν by changing the energy of pair creation. Fortunately, the latter depends in large part on exchange constants that vary with distance16. Thus a change of lattice constant from 10.1 to 9.93 Å, as achieved by substituting Ti for Ge in Dy2Ti2O7 (Fig. 2), may be sufficient to radically alter the chemical potential. We have discovered that this is indeed the case, and that Dy2Ge2O7, which has a much smaller lattice constant than any spin ice hitherto investigated, lies almost precisely on the boundary of the strongly and weakly correlated regimes in Figure 1. This means that it has significant monopole dimerization at all measured temperatures.

Figure 2: Proof that Dy2Ge2O7 (open) and Ho2Ge2O7 (solid) have the characteristic Pauling zero point entropy of spin ice and water ice.
figure 2

The experimental molar entropy, found by integrating the magnetic specific heat Cm divided by temperature T, when referenced to its high temperature value of 2Rln(2) per diamond lattice site, reveals a zero temperature component of Rln(3/2) equal to the Pauling value. The inset shows the Rietveld refined x-ray powder diffraction pattern of the cubic pyrochlore phase of Dy2Ge2O7 with a lattice parameter of 9.9290(5) Å. Errors in the data are smaller than the symbols and represent ±1σ.

It should be noted that the chemical potential and phase behaviour discussed here is not the same as those discussed in refs 16,22 in connection with a field-induced phase transition29. Thus the authors of ref. 16 argue that the magnetic field in that case favours ordering of positive and negative monopoles on different sublattices, leading to the phase transition, and use the terminology 'staggered chemical potential' to describe this. Experimental evidence in support of this scenario is presented in ref. 22. However, the chemical potential we refer to is very different in that it tunes the number density of monopoles without favouring any local ordered arrangement, and is thus equivalent to the chemical potential of ions in an electrolyte (which is not true of the 'staggered chemical potential' of ref. 16).

Results

Pauling entropy

Phase pure cubic Dy2Ge2O7 and Ho2Ge2O7 were prepared and characterized as described in the Methods. Here we describe our results in detail for the Dy compound only, and simply note that we have performed a similar characterization of the Ho compound, which proved to be less interesting in the present context, as it has a more typical chemical potential (Fig. 1). The magnetic entropy determined by integrating the specific heat divided by temperature (cm/T) is shown in Figure 2, where the Pauling residual entropy expected for spin ice8 is extremely well reproduced. Magnetometry measurements on Dy2Ge2O7 showed this material to have a very similar magnetic moment to Dy2Ti2O7. Incurring negligible error, we henceforth assume that the magnetic moment per Dy is equal in the two materials (9.87 μB)15.

Debye-Hückel theory

In Figure 3, we show the measured cm/T plotted against temperature and fitted to Debye-Hückel theory with monopole chemical potential ν=(3.35±0.05) K. The method we use has been developed to extend Debye-Hückel theory to a good approximation into the high-temperature regime; when applied to Dy2Ti2O7 this method gives a similarly good fit to cm(T)/T with a chemical potential of the expected magnitude. It is based on mapping the system to a lattice gas with site exclusion. The lattice gas is considered to have a temperature-varying chemical potential equal to the sum of the true chemical potential as defined in ref. 28 and the standard Debye-Hückel Coulombic correction to the chemical potential30. Without the latter correction, the predicted specific heat (dotted line in Fig. 3) describes the experimental data well in the limit of high and low temperature, highlighting that this approach is a robust method of deriving an experimental estimate of the monopole chemical potential that is not significantly biased by the limitations of Debye-Hückel theory. In passing, we note that the origin of the approximate collapse of the experimental data and Debye-Hückel calculation onto the ideal lattice gas model at high temperature has a different origin to that at low temperature. In the latter case, the monopole gas is sufficiently dilute that interactions can be neglected, whereas in the former case, a dense and interacting monopole gas reproduces apparent ideal gas behaviour, as the result strong Coulombic screening.

Figure 3: Measured heat capacity per mole of Dy2Ge2O7 at zero field compared with theoretical models.
figure 3

Main figure: the modified Debye-Hückel theory (Black line), with monopole chemical potential ν=3.35(5) K the only adjustable parameter, gives an excellent description of the experimental magnetic heat capacity of Dy2Ge2O7 (points). Dashed black line shows the heat capacity of an ideal lattice gas with onsite exclusion with the same chemical potential. This model describes the data well at low temperatures where Coulomb interactions may be neglected, and at high temperatures where the interactions are strongly screened. Inset: the effect of varying chemical potential from −3.35 K (appropriate to Dy2Ge2O7 (DGO)) to −4.35 K (appropriate to Dy2Ti2O7 (DTO)).

As a further test for consistency, we may use our fitted value of ν to derive a value of the effective near-neighbour coupling constant, Jeff, according to the relationship discussed in ref. 16, and then compare this with a Jeff estimated from the temperature of the specific heat maximum, as discussed in ref. 31. The result is Jeff=(0.62±0.1) K, (0.60±0.1) K, respectively, estimates that are equal, within experimental error. For Dy2Ti2O7, the corresponding value is Jeff≈1.1 K, which is roughly twice as large. The large difference is accounted for by a more negative (antiferromagnetic) exchange contribution to the spin–spin interaction in Dy2Ge2O7 that opposes the positive (ferromagnetic) dipolar coupling, that is almost the same in the two compounds.

Bjerrum pairing

The measured chemical potential, −3.35 K, puts Dy2Ge2O7 in a regime where monopole dimerization should be very significant. To confirm this, we have used the classic theory of Bjerrum32, who separated the contribution of closely spaced charges out of Debye-Hückel theory and regarded these as distinct chemical entities to be considered alongside the free charges. In Figure 4, we present the energy u(T) of the system found by integrating the specific heat. The result of Debye-Hückel+Bjerrum theory, as described in the Methods, is shown to give an excellent description of the low-temperature data, without the addition of any new fitting parameters. This result is not in contradiction with the fit of Figure 3, as the modified Debye-Hückel theory described in the Methods, including the short-distance contribution of charge interactions, naturally incorporates the effect of Bjerrum pairs, to an excellent approximation. The monopoles of Dy2Ge2O7 are found to be about 50% dimerized at all measured temperatures.

Figure 4: The experimental energy u(T) of Dy2Ge2O7 and the significance of the Bjerrum correction.
figure 4

Main figure: red circles are experimental data points. Black line is the predicted energy incorporating both free monopoles and (±) monopole dimers, using the chemical potential estimated from the data of Figure 3. Black dashed line is the Debye-Hückel theory with a short range cutoff of two lattice spacings, which thus neglects charge dimers. However, our modified Debye-Hückel method fits the data just as well as Bjerrum's method, as it accounts for the dimers to a good approximation (Fig. 3). Inset: upper curve to lower curve are the theoretical energy for Dy2Ge2O7 (with and without Bjerrum pairs) and Dy2Ti2O7 (with and without Bjerrum pairs), respectively. The Bjerrum correction is much more important for Dy2Ge2O7 than for Dy2Ti2O7.

Discussion

We may put these results in the context of the corresponding states diagram (essentially the phase diagram of T* versus x) for the restricted primitive model electrolyte, a basic model of electrolyte behaviour. In the case of a continuum electrolyte, there are three significant boundaries on this diagram marking, respectively, the onset of significant dimerization, the conductance minimum, and phase separation (see, for example, Fig. 1 in ref. 25). In Dy2Ge2O7, we have reached the first of these boundaries for spin ice. To reach the other boundaries would require us to find a spin ice material with |ν|3.3 K. However, a lattice Coulomb gas-like spin ice may show yet more complex phase behaviour in this limit, including charge-ordered phases26,27. In fact, the ultimate limit of tuning the monopole chemical potential to ν3 K has already been identified through numerical studies on the dipolar spin ice model14. In monopole language, this structure consists of the ordering of 'double charges' ±2Q to give a magnetic structure with '4 spins in/4 spins out' on alternate tetrahedra. This structure (also known as the FeF3 structure) becomes stable at ν=2.4 K, Jeff=0.2 K. In the unchartered region between ν≈3.3 K and ν≈2.4 K, we would expect to find a great deal of interesting physics associated with increasing monopole correlations and the gradual appearance of double charges. Our results illustrate that this region should be accessible to experiment, as we have shown how high-pressure methods afford the opportunity of dramatically altering the chemical potential of magnetic monopoles in spin ice, to the degree where new aspects of monopole physics can be revealed.

Methods

Sample preparation and characterization

Batches of up to 50 mg of the pyrochlore dysprosium germanate, Dy2Ge2O7, were made in a Walker-type, multi-anvil press. Stoichiometric amounts of Dy2O3 and GeO2 were ground thoroughly, wrapped in gold foil, compressed to 7 GPa and heated to 1,000 °C. Rietveld refinement of the X-ray powder diffraction pattern confirmed the face-centred cubic space group, (Fd-3m, No. 227) and the absence of any tetragonal pyrogermanate. The room temperature lattice parameter was determined to be 9.9290(5) Å, (Fig. 2 (inset)). Temperature- and field-dependent magnetization measurements confirmed a rare earth magnetic moment of 10 μB and a Curie–Weiss constant of 0.0 K. The heat capacity was measured using a thermal relaxation method between 0.34 and 25 K. The lattice contribution was subtracted from the measured specific heat to reveal the magnetic contribution. The energy u(T) was found by numerical integration of the measured specific heat.

Heat capacity analysis

To calculate the specific heat, the energy per diamond lattice site, u, was written u=|ν|x, where x is the number of monopoles per lattice site: . Here νDH is the standard Debye-Hückel correction to the chemical potential (related to the electrochemical activity coefficient): |νDH|/kBT=lT/(lD+a), where lT=μ0Q2/(8πkBT), is the Bjerrum length, is the Debye length, and Vd is the volume per diamond lattice site (for a detailed discussion of these quantities, see ref. 23). Using these equations x and lD were determined self consistently, and then the specific heat was found by differentiating u.

To incorporate charge dimers, we considered them as near-neighbour pairs, which is appropriate on a lattice26. Their chemical potential is νd=2|ν|−μ0Q2/(4πa), giving the number of pairs per diamond lattice site: xB≈2exp−|νd|/(kBT). The Debye-Hückel correction was modified to avoid double counting these pairs as follows: νDH/(kBT)→lT/(lD+2a). The energy was then calculated as: u=|νd|xB+|ν|x, which was compared with the measured u(T).

These methods have been comprehensively tested and shown to provide a robust analysis of specific heat data on spin ice materials. They were used to estimate the curves in Figures 1,3 and 4. In Figure 1, the chemical potentials used are: 3.35 K (DyGe), 4.35 K (DyTi), 5.5 K (HoGe) and 5.8 K (HoTi) in an obvious notation; for the Ho materials, these are only rough estimates, owing to the difficulty of accurately isolating the electronic specific heat from the nuclear component9.

Additional information

How to cite this article: Zhou H. D. et al. High pressure route to generate magnetic monopole dimers in spin ice. Nat. Commun. 2:478 doi: 10.1038/ncomms1483 (2011).