Calculating the electronic structure of TiSe2 presents a challenge: standard density-functional approximations can misrepresent the relative positions of titanium d and selenium p states, while conventional G0W0 results depend on the electronic structure used as their starting point. A quasi-self-consistent GW approach offers a way to address that sensitivity. Applied to TiSe2, it predicts a semimetallic high-temperature phase and provides a route to selecting a hybrid functional for studying the low-temperature charge density wave (CDW).
Why the starting point matters for TiSe2
TiSe2 is a layered material that develops a CDW below approximately 200 K. The transition is associated with a structural superlattice, but the relative roles of electron–phonon coupling and excitonic effects remain debated. Understanding the high-temperature electronic bands is important to evaluating these competing explanations.
Common local and semilocal density-functional approximations, including LDA and GGA, do not provide a complete account of TiSe2’s electronic and vibrational properties. In particular, they can place the Ti d levels inaccurately and overstate the overlap between titanium d and selenium p bands. Hybrid functionals improve the description by including a fraction of Hartree–Fock exchange, but their results depend on the selected exchange fraction and, in some cases, the range of the Coulomb interaction.
GW offers a physically motivated alternative: it describes exchange using a screened Coulomb interaction. Yet the widely used perturbative G0W0 method calculates corrections from an initial, or starting-point, electronic structure. If that starting point is a poor representation of the material, the resulting band structure may be unreliable. This concern is especially relevant for TiSe2, where exchange has a substantial influence on the orbitals and band overlap.
How quasi-self-consistent GW is constructed
Rather than carry out a fully self-consistent GW calculation, the method uses the random phase approximation (RPA) and an optimized effective potential to construct a more suitable starting point. The RPA provides a framework for calculating the screened interaction, while the optimized effective potential supplies a local potential for generating the electronic states used in the calculation.
The approach approximates the local RPA potential with a local hybrid potential. It then adjusts the hybrid’s fraction of exact exchange so that the hybrid and GW corrections agree for the difference between the highest occupied and lowest unoccupied states. In this way, the exchange fraction is determined through a consistency condition, rather than selected solely by fitting to a conventional semiconductor test set.
This is a quasi-self-consistent procedure, not a full solution of the GW equations. Its purpose is to improve the starting point and account for changes in both eigenvalues and orbitals, while avoiding the higher computational demands of fully self-consistent calculations. Tests on argon, cubic boron nitride, and scandium nitride show smaller effects from this procedure than in the titanium dichalcogenides, where the starting point has a stronger influence.
Results for the high-temperature phase
For TiSe2, G0W0 calculations based on different starting potentials produce markedly different band overlaps. Starting from PBE gives a gap in the calculated spectrum, whereas using an exchange-rich potential changes the result substantially and produces a much larger band overlap. This variation illustrates why an unadjusted perturbative calculation can give a misleading picture of the material.
With the optimized local hybrid starting point, the calculation instead finds a small band overlap of about 0.1 eV. The authors characterize the high-temperature phase as semimetallic. This result is consistent with transport measurements and falls within the range of several angle-resolved photoemission spectroscopy (ARPES) measurements discussed in the study, though it differs from earlier interpretations that reported a gap.
The calculated band dispersions also compare favorably with the hybrid-functional results. Some differences remain, including in the position of a relatively flat band at lower energy, and the calculations do not include spin–orbit coupling. Those considerations matter when comparing computed bands directly with ARPES measurements.
Using the optimized hybrid functional to study the CDW
The exchange fraction selected through the quasi-self-consistent procedure also defines a hybrid functional that can be applied to the low-temperature CDW structure. This extends the practical value of the method: calculations of the CDW distortion and associated lattice behavior can be performed without applying the more demanding GW treatment to the distorted structure.
For the RPA-optimized hybrid functional, the calculated energy change as the atoms distort follows closely the results of earlier hybrid-functional work. The earlier study found that exchange was important to reproducing the CDW distortion and related phonon behavior. The agreement therefore supports the use of a theoretically motivated exchange fraction, rather than relying only on parameter fitting.
By contrast, a different hybrid optimization procedure yields a larger exchange fraction and a substantially larger band gap. In the reported comparison, that functional reduces the calculated energy gain from the CDW distortion. The result underscores that not every route to choosing a hybrid parameter is equally suitable for capturing both the high-temperature bands and the low-temperature structural instability.
Practical implications
The study shows that GW calculations for TiSe2 require careful attention to the starting electronic structure. For this material, treating G0W0 as a simple correction to a standard PBE or LDA calculation is not sufficient to establish a robust band picture. A quasi-self-consistent approach using a local hybrid potential offers a way to improve that starting point and obtain a small semimetallic band overlap.
For researchers modeling TiSe2, the practical takeaway is to evaluate how exchange affects the orbitals as well as the eigenvalues, and to distinguish between different procedures for selecting hybrid-functional parameters. The RPA-based method connects the band-structure calculation to a hybrid functional that can also be used to investigate the CDW phase. It does not settle the broader debate over the transition mechanism, but it provides a consistent computational framework for examining the electronic structure and lattice distortion together.