Wannier Function Spread: Localization Insights

• Wannier function spread is a measure of the real-space localization of orbitals derived from Bloch eigenstates, crucial for electronic-structure analysis.
• It decomposes into gauge-invariant and gauge-dependent components, guiding the search for maximally localized Wannier functions.
• Gradient-based minimization algorithms optimize the spread, enhancing tight-binding modeling, chemical bonding insights, and transport property analysis.

Wannier function spread is the quantitative measure of the real-space localization of Wannier orbitals constructed from Bloch eigenstates or energy eigenstates. It plays a central role in ab initio electronic-structure theory, tight-binding model construction, and the analysis of chemical bonding, polarization, and transport properties in both crystalline and noncrystalline systems. The spread functional serves as the target for numerical minimization in the search for maximally localized Wannier functions (MLWFs) and underpins several modern algorithms, theoretical developments, and experimental probes within quantum materials science.

1. Definition and Decomposition of the Spread Functional

The total spread Ω for a set of Wannier functions $\{w_n\}$ associated with an orthonormal basis of an isolated N-band subspace is defined as the sum of second central moments:

$$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$

where $\langle w_n | r^2 | w_n \rangle$ is the expectation value of $r^2$ (quadratic spatial moment), and $\langle w_n | r | w_n \rangle$ defines the center of each Wannier function.

For periodic systems, $\Omega$ admits a decomposition (Ri et al., 2014, Sakuma, 2013, Damle et al., 2018):

$\Omega = \Omega_I + \Omega_D + \Omega_{OD}$

• $\Omega_I$ (gauge-invariant): Depends only on the occupied subspace, i.e., independent of the particular gauge choice.
• $\Omega_D$ (diagonal gauge-dependent): Measures the spatial fluctuation of individual Wannier centers.
• $\Omega_{OD}$ (off-diagonal gauge-dependent): Quantifies the inter-Wannier overlap away from the home cell.

These quantities can be explicitly connected to k-space expressions via overlaps of the cell-periodic parts of Bloch functions on a discrete mesh:

$$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$0

and the spread in practice is sampled using such overlaps, facilitating the use of efficient minimization schemes (Damle et al., 2018, Sakuma, 2013).

2. Properties, Interpretation, and Gauge Structure

Wannier function spread is intimately tied to the spatial decay and physical locality of the basis. For a given (isolated) set of bands, $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$1 is minimized with respect to the $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$2 gauge transformation acting on the Bloch states, subject to smoothness and symmetry constraints as needed (Ri et al., 2014).

The gauge-invariant part $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$3 is directly related to the valence-band quantum metric tensor $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$4—the Brillouin zone average of the quantum geometric tensor:

$$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$5

where $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$6 (Cárdenas-Castillo et al., 2024). $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$7 is thus a band subspace characteristic and cannot be reduced by gauge transformations; the minimization of $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$8 effectively reduces the gauge-dependent remainder $$\Omega = \sum_{n=1}^N \Bigl( \langle w_n | r^2 | w_n \rangle - |\langle w_n | r | w_n \rangle|^2 \Bigr)$$9 to zero in the ideal limit (for MLWFs), yielding the most localized real-space representation (Ri et al., 2014, Modugno et al., 2011).

Time-reversal symmetry guarantees that the MLWFs can always be taken as real-valued functions, and the strict minimum of the spread is always achieved by a real gauge choice (Ri et al., 2014).

3. Minimization Algorithms and Computational Strategies

Spread minimization is carried out by optimizing a set of gauge transformation matrices $\langle w_n | r^2 | w_n \rangle$0 (isolated bands) or generalized block structures (entangled bands), typically using gradient-based methods such as steepest descent, conjugate-gradient, or quasi-Newton solvers on appropriate matrix manifolds (Damle et al., 2018, Tillack et al., 5 Feb 2025).

A canonical numerical iteration for unitary $\langle w_n | r^2 | w_n \rangle$1 minimizes Ω under the constraint $\langle w_n | r^2 | w_n \rangle$2, updating via the anti-Hermitian gradient $\langle w_n | r^2 | w_n \rangle$3:

$\langle w_n | r^2 | w_n \rangle$4

with $\langle w_n | r^2 | w_n \rangle$5 and step size $\langle w_n | r^2 | w_n \rangle$6 (Zhu et al., 2015). For entangled bands or high-throughput workflows, recent developments recast spread minimization as optimization over semi-unitary “projection” matrices, with an exact expression for the gradient allowing robust and automated solutions (Tillack et al., 5 Feb 2025).

Novel definitions of spread under periodic boundary conditions—such as the density-convolution spread—eschew ill-defined “centers,” providing manifestly gauge-continuous, thermodynamically consistent, and computationally efficient objectives (Li et al., 2023).

For disordered and nonperiodic systems, the spread is still defined as the sum of real-space second moments, with the minimization performed over banded unitary transformations constructed from the energy eigenstates; the approach generalizes directly to higher dimensions (Zhu et al., 2015).

Table 1: Spread Minimization Workflows (select approaches)

Approach	Gauge Structure	Spread Functional
Marzari–Vanderbilt (isolated bands)	$\langle w_n | r^2 | w_n \rangle$7 at each $\langle w_n | r^2 | w_n \rangle$8	$\langle w_n | r^2 | w_n \rangle$9 with k-space overlaps
Variational MLWF for entangled bands	Partial isometry	$r^2$0 + subspace constraint
Optimized Projection Functions (OPF)	Semi-unitary X-matrix	$r^2$1 via SVD+gradient
Density-Convolution (DC/TDC)	N/A (center-free)	$r^2$2 via integrals/FFTs

4. Extensions: Interactions, Symmetry, and Physical Probes

In the presence of symmetry constraints, the spread minimization can be augmented with explicit site or point-group symmetry, yielding symmetry-adapted Wannier functions (SAWFs). This proceeds by restricting permissible gauge transformations to those commuting with the action of the little group at each $r^2$3-point, which may result in local or global minima of Ω that do not coincide with the unconstrained global minimizer (Sakuma, 2013).

Interactions (e.g. in optical lattices) renormalize the Wannier spread due to density-density correlations. Self-consistent minimization in the presence of on-site repulsion broadens the Wannier functions, as quantified by an increase in the second central moment $r^2$4 (Zhu et al., 2015). This broadening directly affects model parameters such as the Hubbard $r^2$5 and $r^2$6, and can be experimentally detected by shifts in atomic clock transition frequencies.

The gauge-invariant part of the spread $r^2$7 is experimentally accessible via optical sum rules, linking the Brillouin-zone average quantum metric to frequency-integrated optical conductivity or absorbance. In three dimensions,

$r^2$8

and in two dimensions, via the absorbance divided by frequency, both enabling direct extraction of $r^2$9 from broadband optical data (Cárdenas-Castillo et al., 2024).

5. Technical Variants and Practical Considerations

Several techniques have been introduced to address limitations of the standard spread minimization:

• Variance-penalized spread functionals minimize not just the mean spread but also the variance among individual WFs, suppressing pathologies such as a single poorly localized orbital in large or entangled subspaces (Fontana et al., 2021).
• Automatable workflows exploit robust initial guesses via SCDM or projection function techniques, and adaptive selection of the number of bands for localization, enabling high-throughput Wannierization (Tillack et al., 5 Feb 2025, Fontana et al., 2021).
• Alternative spread definitions, such as the truncated cosine (TDC) surrogate, retain spectral accuracy, gauge continuity, and substantially expedite convergence (10×–70× reduction in iterations) versus the Marzari–Vanderbilt (MV) spread (Li et al., 2023).
• In special cases (e.g. 1D double-well lattices), analytic gauge fixing—phase alignment and SU(2) mixing—can drive both diagonal and off-diagonal spread components to zero; the optimal spread for composite bands is then set by $\langle w_n | r | w_n \rangle$0 alone (Modugno et al., 2011).

6. Illustrative Results: Materials and Model Systems

The spread functional has been systematically benchmarked across a wide array of systems. For example, in 3D semiconductors (Si, Ge), experimental integration of the dielectric function yields $\langle w_n | r | w_n \rangle$1 values between 68–83 Å$\langle w_n | r | w_n \rangle$2 (Cárdenas-Castillo et al., 2024). In complex 2D materials, $\langle w_n | r | w_n \rangle$3 ranges from 0.64 to 0.80 as a function of chemical identity. For strongly disordered 1D wells, the spread saturates at $\langle w_n | r | w_n \rangle$4–$\langle w_n | r | w_n \rangle$5 for moderate disorder strength after full minimization, with the final Wannier functions exponentially localized over ∼10–15 wells (Zhu et al., 2015).

Numerical optimization consistently demonstrates that state-of-the-art variational solvers and center-free spread functionals yield lower or equal spreads with robust convergence compared to legacy schemes, particularly in systems with entangled bands or large basis sets (Damle et al., 2018, Li et al., 2023, Tillack et al., 5 Feb 2025).

7. Theoretical and Experimental Significance

Wannier function spread is a fundamental descriptor of electron localization, insulator/metal crossover, and the ability to construct accurate tight-binding models. The rigorous definition, gauge structure, algorithms for minimization, and connection to quantum geometry and optical response make the Wannier spread a central tool for both theoretical and experimental condensed matter physics. Ongoing developments—convergence acceleration, symmetry adaptation, interaction effects, and experimental extraction of $\langle w_n | r | w_n \rangle$6—continue to expand the range of systems and phenomena accessible to Wannier-based analyses (Cárdenas-Castillo et al., 2024, Li et al., 2023, Fontana et al., 2021, Zhu et al., 2015).