A Unified Approach to Filling- and Bandwidth-Control Mott Transitions

The nature of the system in which quantum fluctuations and electron correlations play essential roles is one of the main subjects in condensed matter physics. When the kinetic energy and the Coulomb repulsion compete severely, the ground state of many-body electron systems can be highly nontrivial. Metal-insulator transitions driven by the electron correlation postulated by Mott [1] provide a typical example of such a nontrivial behavior.

In the transition to the Mott insulator, it is known that there exist two different basic routes to control the competition of the interaction and kinetic energies. One is the control by bandwidth (relative to Coulomb repulsion) and the other is filling (averaged number of carrier electrons in a unit cell). Controls by these two parameters can be found in a lot of examples in real materials including transition metal compound, organic materials and 3He systems. In spite of plenty of the experimental results, phase diagrams of the Mott insulator and metals have not been fully elucidated in microscopic theoretical descriptions.

The Hubbard model with nearest- and next-nearest-neighbor transfers is a minimal model to describe the essence of the Mott insulator and metals with their transitions on the bandwidth- and filling-control routes.  The filling-control Mott transition (FCMT) was studied at zero temperature by the quantum Monte Carlo method in the Hubbard model on a square lattice [2]. The transition shows a continuous character with a singular divergence of the compressibility and critical divergence of the antiferromagnetic correlation length.  The bandwidth-control Mott transition (BCMT) was studied also at zero temperature by the path integral renormalization group method [3]. In contrast to the FCMT, the BCMT shows a first-order transition, although a naive expectation is that the continuous FCMT anticipates also the continuous BCMT.  This contrast is essentially consistent with the trend of the experimental observations cited above.

Mott has originally proposed a first-order BCMT because of the role of the long-ranged part of the Coulomb interaction [1]. However, the numerical result shows that the first-order transition takes place even with the onsite interaction only. In this circumstance, it has been greatly desired to clarify basic properties of the BCMT and the FCMT in a unified way to further elucidate the contrast.

In this study, we have developed the grand-canonical path-integral renormalization group method [4] which is useful to calculate chemical potential dependence of physical quantities. The path-integral renormalization group method has been originally developed in the canonical ensemble. This algorithm does not suffer from negative-sign problem and can be applied to any lattice structure with any boundary conditions.  Extension of this algorithm to the grand-canonical framework makes it possible to study the correlated electron systems whose control parameters are filling, bandwidth and lattice structure in a unified way. In the grand-canonical path-integral renormalization group calculations, metal-insulator transitions are carefully examined with finite-size scalings and extrapolations to the thermodynamic limit.

By using the newly developed method, the ground-state phase diagram of the Hubbard model with next-nearest-neighbor transfer scaled by nearest-neighbor transfer t’/t =-0.2 on the square lattice has been determined in the plane of chemical potential m and Coulomb interaction U [4](Figure 1). The Mott-insulator phase is drawn as the pink area and the metallic phase is drawn as the yellow area. The remarkable result is that the V-shaped Mott insulator phase appears. At the corner of the V-shaped metal-insulator boundary, the BCMT occurs and at the edges except the corner, the FCMT occurs. We have confirmed that the continuous character of the FCMT with diverging charge compressibility and the first-order BCMT with a jump of the double occupancy.

To analyze the relation between the shape of the phase boundary and the order of the metal-insulator transition, we have analytically derived a general relation of the slope of the metal-insulator transition line in the m-U phase diagram and physical quantities [4]: In the case of the first-order metal-insulator transition, dU/dm is expressed by the ratio of the jumps in the filling and the double occupancy Dn/DD. In the case of the continuous transition, dU/dm is expressed by the ratio of the compressibility and dn/dU in the metallic phase. These relations can be regarded as generalizations of Clausius-Clapeyron and Ehrenfest equations in the first- and second-order transitions at finite temperature, respectively, to the quantum phase transitions.

These relations support that the V-shaped phase boundary is resulted from the first-order BCMT coexisting with the continuous FCMT with diverging compressibility. Namely, it is shown that the V-shaped metal-insulator transition line together with the first-order BCMT is not compatible with the presence of the first-order FCMT near the BCMT.

When V-shaped Mott insulator phase appears, the charge gap Dc opens at U=Uc and shows marked linear dependence on U for U>Uc, namely, Dc~ U-Uc. It is noted that the linear opening of the Mott gap has been actually observed in the perovskite compounds, R1-xCaxTiO3 [5], which is consistent with our results.

We have also shown the possible and impossible shapes of the metal-insulator boundary with first-order and continuous transitions by using the thermodynamic relations derived above: The U-shaped structure of the insulator phase with the first-order BCMT coexisting with the first-order FCMT can be realized in contrast with the V-shaped case. On the other hand, in the case of the U-shaped insulator phase (for example, the case with an essential singular form of the charge gap as Dc ~ exp[at/(U-Uc)]), if the first-order BCMT exists at the corner, the first-order transition cannot be retained at the FCMT, which is classified to the same class as the V-shaped case.

 

References

[1] N. F. Mott and R. Peierls: Proc. Phys. Soc. London. A49 72 (1937).

[2] N. Furukawa and M. Imada: J. Phys. Soc. Jpn. 61 3331 (1992).

[3] T. Kashima and M. Imada: J. Phys. Soc. Jpn. 70 3052 (2001).

[4] S. Watanabe and M. Imada: J. Phys. Soc. Jpn. 73 1251 (2004).

[5] T. Katsufuji, Y. Okimoto and Y. Tokura: Phys. Rev. Lett. 75 3497 (1995).

 

Authors

S. Watanabe and M. Imada

Fig. 1:

Ground-state phase diagram in the plane of chemical potential m and Coulomb interaction U for nearest-neighbor transfer t=1.0 and next-nearest-neighbor transfer t'=-0.2 on the square lattice in the thermodynamic limit. Open circles represent the metal-insulator boundary extrapolated to the thermodynamic limit by using the data of N=4×4, 6×6, 8×8, 10×10 lattice systems. The solid black lines represent the least-square fit of metal-insulator boundary for U=4.0, 5.0, 6.0, 7.0 and 8.0. The V-shaped Mott insulator phase is drawn as pink area and the metallic phase is drawn as the yellow area. The corner of the V-shaped metal-insulator boundary is the bandwidth-control Mott transition point and the edges are the filling-control Mott transition points. The red diamonds represent chemical potentials for half filling for U=0.0 and U=2.0 in the thermodynamic limit. By connecting those points to the bandwidth-control Mott transition point, the blue-dashed line represents the half-filled density, which is the bandwidth-control route in the metallic phase.