Marzari–Vanderbilt Spread Functional

• Marzari–Vanderbilt Spread Functional is a variational measure that quantifies the second moment (spread) of Wannier functions, distinguishing trivial and nontrivial topological phases.
• Its minimization via gauge optimization yields maximally localized Wannier functions, which are essential for constructing accurate tight-binding and Hubbard models.
• The approach integrates real-space and momentum-space representations to decompose gauge-invariant and gauge-dependent contributions, thereby diagnosing topological obstructions.

The Marzari–Vanderbilt (MV) spread functional is a fundamental variational measure of Wannier function localization in periodic systems, introduced to quantify and optimize the spatial extent of composite Wannier functions constructed from isolated or entangled energy bands. The MV functional provides both a quantitative criterion for maximal localization and a theoretical tool to expose the interplay between band topology, gauge structure, and real-space electron localization. Its minimization leads to maximally localized Wannier functions (MLWFs), which are central to electronic structure theory, quantum transport, and the construction of effective tight-binding and Hubbard models.

1. Formal Definition and Gauge Structure

Let $\{w_n(x)\}_{n=1}^m$ be an orthonormal set of composite Wannier functions spanning the occupied subspace of a gapped periodic or magnetic Hamiltonian $H$. The MV spread functional is defined as the total second central moment (“spread”) of the Wannier orbitals about their centers: $$\Omega = \sum_{n=1}^m \left[ \langle w_n, |x|^2 w_n \rangle - |\langle w_n, x w_n \rangle|^2 \right]$$ Alternatively, for a normalized state $\Psi$ in the $m$-band subspace,

$$\Omega = \langle \Psi, x^2 \Psi \rangle - |\langle \Psi, x \Psi \rangle|^2$$

The minimization of $\Omega$ over all possible choices of orthonormal Wannier functions, related by $k$-dependent unitary “gauges,” yields the MLWFs. The functional can be decomposed into a gauge-invariant part $\Omega_{\mathrm{I}}$, reflecting intrinsic band geometric properties, and a gauge-dependent remainder $\tilde{\Omega}$, which can be minimized by appropriate gauge transformations (Monaco et al., 2016, Monaco et al., 2016, Panati et al., 2011).

2. Real-Space and Momentum-Space Representations

Real-Space Representation: Each term $\langle w_n, |x|^2 w_n \rangle$ evaluates the mean-squared position of $w_n(x)$, while $|\langle w_n, x w_n \rangle|^2$ subtracts the squared mean—yielding the variance. The total spread is the sum of variances over all occupied Wannier functions.

Momentum-Space (Bloch) Representation: In $k$-space, the Wannier functions are constructed from Bloch functions by unitary mixing: $$w_n(x) = \frac{1}{|B|} \int_{B} e^{ik\cdot x} v_{n,k}(x) dk$$ The MV functional becomes: $$\Omega = \int_{\mathbb{T}^d} dk \;\mathrm{Tr}\left[ P(k)\,|i\nabla_k|^2\,P(k) \right] - \sum_{j=1}^d \left| \int_{\mathbb{T}^d} dk\;\mathrm{Tr} \left[ P(k)\,i\partial_{k_j}\,P(k) \right]\right|^2$$ where $P(k)$ is the projector onto the occupied space at each crystal momentum $k$. The gauge-invariant part of $\Omega$ is directly related to the quantum metric, the real part of the quantum geometric tensor (Monaco et al., 2016, Verma et al., 2021).

3. Decomposition: Gauge-Invariant and Gauge-Dependent Components

The MV spread can be written as: $$\Omega = \Omega_{\mathrm{I}} + \Omega_{\mathrm{D}} + \Omega_{\mathrm{OD}}$$

• $\Omega_{\mathrm{I}}$: Gauge-invariant; function of the quantum metric $g_{ij}(k)$ averaged over the Brillouin zone.
• $\Omega_{\mathrm{D}}$: Diagonal gauge-dependent part, associated with the Wannier centers and intra-orbital spread.
• $\Omega_{\mathrm{OD}}$: Off-diagonal gauge-dependent part, from inter-orbital (band-mixing) terms.

A maximally localized gauge can always minimize $\Omega_{\mathrm{D}}$ and $\Omega_{\mathrm{OD}}$ to zero in topologically trivial phases, leaving $\Omega = \Omega_{\mathrm{I}}$ as the minimum attainable spread (Wang et al., 2014, Modugno et al., 2011).

4. Topology and the Localization Dichotomy

A central result is the localization dichotomy in dimensions $d \leq 3$ (Monaco et al., 2016, Monaco et al., 2016, Panati et al., 2011):

• Trivial Topology (All Chern Numbers Vanish): Existence of a globally analytic, periodic Bloch frame. One can construct exponentially localized composite Wannier functions with finite spread $\Omega < +\infty$. The occupied subspace allows MLWFs with controlled spatial extent, and $\Omega$ is finite.
• Nontrivial Topology (Some Chern Number Nonzero): No choice of gauge yields exponentially localized Wannier functions. For any Wannier basis, the second moment $\int |x|^2 |w_n(x)|^2 dx$ diverges and $\Omega = +\infty$. This is a direct manifestation of topological obstructions: the Bloch bundle cannot be globally trivialized, and the quantum Hall effect or Chern number enforces delocalization in at least one spatial direction.

This dichotomy is tightly connected to the geometry of the underlying vector bundle and is physically realized in Chern insulators and quantum Hall systems, where the failure of exponential localization is tied to quantized transverse (Hall) conductivity (Monaco et al., 2016, Verma et al., 2021).

5. Variational Principle and Minimization Algorithms

Minimization of $\Omega$ over all unitary gauges $U^k$ is achieved by a two-stage process:

1. Initial Projection: Construction of trial localized orbitals to generate a good starting gauge.
2. Iterative Gauge Optimization: Nonlinear conjugate-gradient minimization of $\Omega[U]$ with respect to $U^k$, using the explicit gradient with respect to gauge variations. Discretized versions on $\mathbf{k}$-meshes utilize overlap matrices $$M_{mn}^{(\mathbf{k},\mathbf{b})} = \langle u_{m,\mathbf{k}} | u_{n,\mathbf{k}+\mathbf{b}} \rangle$$ for practical implementation (Mustafa et al., 2015, Wang et al., 2014).

Variants such as the optimized projection functions (OPF) and selectively localized Wannier functions (SLWF) further enable robust minimization, with or without explicit initial guess functions and for selected subspaces (Mustafa et al., 2015, Wang et al., 2014).

Recent reformulations improve the MV functional's robustness under periodic boundary conditions, removing ill-defined orbital centers and gauge discontinuities by employing a density-convolution (DC) functional and its discretization (TDC), yielding improved numerical stability, smooth gauge gradients, and enhanced convergence (Li et al., 2023).

6. Physical Significance and Applications

The MV spread $\Omega$ quantifies the real-space extent—variance—of Wannier functions. A smaller $\Omega$ indicates highly localized orbitals, essential for:

• Electronic Structure: Accurate tight-binding and Hubbard models.
• Quantum Transport: The magnitude of $\Omega$ directly enters rigorous upper bounds on the low-energy optical spectral weight, superfluid stiffness, and critical temperature $T_c$ in flat-band superconductors (Verma et al., 2021).
• Modern Theory of Polarization: Exponentially localized Wannier functions underpin the computation of electrical polarization, orbital magnetization, and response functions.
• Topological Diagnostics: The divergence of $\Omega$ serves as a topological invariant; its finiteness or divergence directly diagnoses the Chern class of the Bloch bundle (Monaco et al., 2016, Monaco et al., 2016).

The gauge-invariant part $\Omega_\mathrm{I}$ is bounded below by the Brillouin-zone-averaged quantum metric, ensuring minimal real-space delocalization dictated solely by band geometry (Verma et al., 2021).

7. Mathematical Existence, Regularity, and Algorithmic Considerations

Existence of MLWFs minimizing the MV functional is guaranteed for $d < 4$ under mild analytic and spectral conditions on the Hamiltonian. Minimizers are real-analytic in $k$ (jointly in all spatial directions) and yield exponentially decaying Wannier functions in real space (Panati et al., 2011). In nontrivial topological phases, Sobolev regularity breaks down (no global $H^1$ frame), leading to a divergence of $\Omega$ (Monaco et al., 2016, Monaco et al., 2016).

Convergence and robustness of MV-based algorithms depend critically on the presence (or absence) of topological obstructions and the smoothness of the gauge. For topologically nontrivial phases, minimization of $\Omega$ cannot reach a finite value; numerical algorithms typically become unstable or fail to converge as the mesh is refined (Monaco et al., 2016, Monaco et al., 2016).

---

References:

• (Monaco et al., 2016): Optimal decay of Wannier functions in Chern and Quantum Hall insulators
• (Monaco et al., 2016): The Localization Dichotomy for gapped periodic quantum systems
• (Panati et al., 2011): Bloch bundles, Marzari–Vanderbilt functional and maximally localized Wannier functions
• (Li et al., 2023): An unambiguous and robust formulation for Wannier localization
• (Verma et al., 2021): Optical Spectral Weight, Phase Stiffness and Tc Bounds for Trivial and Topological Flat Band Superconductors
• (Mustafa et al., 2015): Automated construction of maximally localized Wannier functions: Optimized projection functions method
• (Wang et al., 2014): Selectively Localized Wannier Functions
• (Modugno et al., 2011): Maximally localized Wannier functions for ultracold atoms in one-dimensional double-well periodic potentials