Tight binding
The tight-binding model, also known as the tight-binding approximation, is a semi-empirical quantum mechanical method used to describe the electronic band structure of crystalline solids by approximating the wavefunctions as linear combinations of localized atomic orbitals centered on lattice sites, with electrons primarily bound to individual atoms but able to "hop" to neighboring sites via weak orbital overlaps.[1] This approach contrasts with nearly free electron models by emphasizing strong atomic potentials that confine electrons, resulting in discrete energy bands formed from bonding and antibonding combinations of orbitals.[2] Developed in the early 20th century and formalized through works like the Slater-Koster parameterization in 1954, the model treats the solid as a lattice where the Hamiltonian includes on-site energy terms (representing isolated atomic levels) and hopping integrals (quantifying inter-atomic interactions that decay rapidly with distance).[3] In one dimension, for instance, the energy dispersion relation simplifies to E(k) = \epsilon_s + 2\gamma \cos(ka), where \epsilon_s is the site energy, \gamma is the nearest-neighbor hopping parameter, k is the wavevector, and a is the lattice constant, yielding a bandwidth of $4|\gamma| and periodic bands within the Brillouin zone.[2] The Bloch theorem is satisfied by constructing wavefunctions as \psi_{nk}(\mathbf{r}) = \frac{1}{\sqrt{N}} \sum_{\mathbf{R}} e^{i\mathbf{k}\cdot\mathbf{R}} \phi_n(\mathbf{r} - \mathbf{R}), where \phi_n are atomic orbitals and \mathbf{R} are lattice vectors.[1] This model excels in providing intuitive insights into band formation, gaps, and effective masses, particularly for materials with localized electrons like semiconductors and transition metals, and extends to higher dimensions and multiple orbitals for complex structures such as graphene or transition metal oxides.[4] Parameters like hopping integrals are often fitted empirically to experimental data, making it computationally efficient for large systems compared to ab initio methods, though it relies on approximations such as the two-center integral neglect of distant interactions.[3] Applications span predicting optical properties, conductivity, and phase transitions in solids, as well as molecular systems via extensions like density functional tight binding.[1]Overview
Introduction
The tight-binding model is an approximate method used in solid-state physics to calculate the electronic band structures of crystalline solids. It posits that electron wavefunctions in a crystal can be represented as linear combinations of atomic-like orbitals localized at each lattice site, assuming weak overlap between these orbitals on neighboring sites, which results in electron hopping between atoms. This approach captures the transition from isolated atomic states to delocalized bands in a periodic lattice.[5] The physical motivation for the tight-binding model lies in simplifying the complex many-body Schrödinger equation for electrons in a solid. By treating electrons as predominantly bound to atomic cores within the crystal, with occasional quantum mechanical tunneling to adjacent sites due to the periodic potential, the model reduces the problem to manageable interactions while respecting the underlying periodicity of the lattice. This framework provides an intuitive picture of how chemical bonding emerges in solids, bridging atomic physics and collective electronic behavior.[6] A key advantage of the tight-binding model is its computational efficiency compared to ab initio methods like density functional theory, enabling calculations for large systems with thousands of atoms while retaining quantum mechanical insights. It excels at offering qualitative understanding of band formation, where discrete atomic energy levels broaden into continuous bands through interatomic coupling, and it serves as a foundational tool for exploring material properties such as conductivity and optical response.[7] The model builds upon the Bloch theorem, which ensures wavefunctions in periodic systems take the form of plane waves modulated by lattice periodicity, setting the stage for more detailed formulations.[5] Emerging from early 20th-century advancements in quantum theory of solids, the tight-binding model has evolved into a versatile approximation widely applied across materials science.[5]Historical Background
The tight-binding model originated from early efforts in quantum mechanics to describe electron behavior in periodic crystal lattices, building on foundational ideas in band theory. In 1928, Felix Bloch, in his doctoral dissertation, proposed that electron wavefunctions in a periodic potential could be expressed as plane waves modulated by periodic functions, now known as Bloch waves; this framework laid the groundwork for the tight-binding approximation by suggesting that crystal states could be constructed from atomic-like orbitals perturbed by the lattice. Bloch's work marked a pivotal shift from free-electron models to those accounting for lattice periodicity, enabling the conceptual basis for energy bands in solids. During the 1930s and 1940s, the model evolved through connections to localized orbitals, with key contributions linking Bloch states to more intuitive atomic representations. Gregory H. Wannier introduced Wannier functions in 1937 as maximally localized orbitals derived from Bloch waves, providing a basis for interpreting tight-binding states in terms of overlapping atomic orbitals while preserving translational symmetry. These developments solidified the tight-binding method as a practical tool for approximating electronic structure beyond nearly free-electron limits. A major milestone came in 1954 with the work of John C. Slater and George F. Koster, who systematized the evaluation of tight-binding matrix elements through tables based on angular momentum projections, transforming the model into a computationally tractable interpolation scheme for periodic potentials.[8] Their approach, often called the Slater-Koster method, facilitated applications to diverse crystal structures by parameterizing hopping integrals from experimental or ab initio data. In the 1980s and 1990s, the tight-binding model matured into semi-empirical frameworks, notably through Walter A. Harrison's universal parameters, which scaled hopping integrals inversely with interatomic distance to predict band structures across the periodic table with minimal adjustable inputs. Harrison's scheme, detailed in his 1980 monograph and subsequent papers, emphasized transferable parameters derived from second-moment approximations, enabling broader use in materials modeling without full self-consistency.Theoretical Foundations
Assumptions and Approximations
The tight-binding model relies on the fundamental assumption that electrons in a crystalline solid can be described using localized atomic orbitals, such as s, p, or d orbitals, as basis functions centered at each lattice site. These orbitals are chosen to mimic the wave functions of isolated atoms and are presumed to decay rapidly away from their atomic cores, ensuring minimal overlap between orbitals on distant sites. This localization reflects the strong binding of electrons to individual atoms under the influence of deep atomic potentials, allowing the model to capture the essential physics of electron motion via tunneling between nearby atoms rather than free propagation.[1][9] A key simplification in the model is the orthogonality approximation, which treats these basis orbitals as approximately orthogonal, effectively setting small overlap integrals between orbitals on different sites to zero. This neglects the subtle but non-zero spatial extensions of the wave functions, enabling straightforward diagonalization of the Hamiltonian without complex non-orthogonal basis corrections. In the basic tight-binding framework, interactions are further restricted to pairwise hopping terms, primarily between nearest neighbors, while three-body or higher-order interactions are ignored to maintain computational tractability.[1][9] The model's validity is strongest in insulators and wide-bandgap semiconductors, where electrons remain tightly bound to atomic sites with limited delocalization, leading to well-separated energy bands. It performs less accurately in metals, where conduction electrons exhibit more free-electron-like behavior due to weaker relative binding. This approach contrasts sharply with the nearly free electron model, which assumes a weak periodic potential perturbing plane-wave states and is better suited to metallic systems, whereas tight-binding excels in scenarios dominated by strong, localized atomic potentials. Potential errors stem primarily from the approximated treatment of overlap integrals, which can distort wave function normalization and band widths if orbital extensions are more significant than assumed, particularly in transition metals or systems with diffuse orbitals.[1]Translational Symmetry and Normalization
In periodic crystals, translational symmetry requires that the single-electron wavefunctions satisfy Bloch's theorem, which states that the wavefunction \psi_{\mathbf{k}}(\mathbf{r}) can be expressed as \psi_{\mathbf{k}}(\mathbf{r}) = e^{i \mathbf{k} \cdot \mathbf{r}} u_{\mathbf{k}}(\mathbf{r}), where u_{\mathbf{k}}(\mathbf{r}) is a periodic function with the periodicity of the lattice, \mathbf{k} is a wavevector in the first Brillouin zone, and \mathbf{r} is the position vector.[1] This form ensures that under a lattice translation by a vector \mathbf{R}, the wavefunction acquires only a phase factor: \psi_{\mathbf{k}}(\mathbf{r} + \mathbf{R}) = e^{i \mathbf{k} \cdot \mathbf{R}} \psi_{\mathbf{k}}(\mathbf{r}).[10] In the tight-binding approximation, the Bloch wavefunction is constructed as a linear combination of atomic orbitals centered at lattice sites, known as the tight-binding ansatz: \psi_{\mathbf{k}}(\mathbf{r}) = \sum_{\mathbf{R}} c_{\mathbf{R}} \phi(\mathbf{r} - \mathbf{R}), where \phi(\mathbf{r}) is an atomic orbital (e.g., an s- or p-orbital) localized around the origin, \mathbf{R} sums over all lattice vectors, and the coefficients c_{\mathbf{R}} incorporate the Bloch phase to satisfy translational symmetry: c_{\mathbf{R}} = N^{-1/2} e^{i \mathbf{k} \cdot \mathbf{R}}, with N being the number of unit cells.[1] This Bloch sum form, \psi_{\mathbf{k}}(\mathbf{r}) = N^{-1/2} \sum_{\mathbf{R}} e^{i \mathbf{k} \cdot \mathbf{R}} \phi(\mathbf{r} - \mathbf{R}), directly fulfills Bloch's theorem, as the periodic part u_{\mathbf{k}}(\mathbf{r}) emerges from the modulated sum of translated orbitals.[10] Proper normalization of the Bloch wavefunction requires \int |\psi_{\mathbf{k}}(\mathbf{r})|^2 d\mathbf{r} = 1 over all space. Substituting the ansatz yields the normalization integral: \int \psi_{\mathbf{k}}^*(\mathbf{r}) \psi_{\mathbf{k}}(\mathbf{r}) d\mathbf{r} = N^{-1} \sum_{\mathbf{R}, \mathbf{R}'} e^{-i \mathbf{k} \cdot (\mathbf{R} - \mathbf{R}')} \int \phi^*(\mathbf{r} - \mathbf{R}) \phi(\mathbf{r} - \mathbf{R}') d\mathbf{r}.[1] Defining the overlap between orbitals at sites separated by \mathbf{R}'' = \mathbf{R}' - \mathbf{R} as S(\mathbf{R}'') = \int \phi^*(\mathbf{r}) \phi(\mathbf{r} - \mathbf{R}'') d\mathbf{r}, the integral simplifies to \sum_{\mathbf{R}''} S(\mathbf{R}'') e^{i \mathbf{k} \cdot \mathbf{R}''}. Assuming the overlap integrals S(\mathbf{R}'') are negligible for \mathbf{R}'' \neq 0 (consistent with the orthogonality approximation), the normalization reduces to S(0) = \int |\phi(\mathbf{r})|^2 d\mathbf{r} = 1, assuming the atomic orbital itself is normalized.[10] In practice, atomic orbitals at neighboring sites overlap, leading to a non-trivial overlap matrix S_{ij}(\mathbf{k}) = \langle \phi_i | e^{-i \mathbf{k} \cdot \mathbf{R}} | \phi_j \rangle = \sum_{\mathbf{R}} e^{i \mathbf{k} \cdot \mathbf{R}} S_{ij}(\mathbf{R}), where S_{ij}(\mathbf{R}) = \int \phi_i^*(\mathbf{r}) \phi_j(\mathbf{r} - \mathbf{R}) d\mathbf{r}.[10] The full normalization condition then becomes \sum_{ij} b_i^*(\mathbf{k}) S_{ij}(\mathbf{k}) b_j(\mathbf{k}) = 1, where \mathbf{b}(\mathbf{k}) are coefficients in a multi-orbital basis per site. However, for simplicity in many applications, especially with orthogonalized orbitals or when overlaps are small, the overlap matrix is approximated as the identity S_{ij}(\mathbf{k}) \approx \delta_{ij}, yielding \sum_i |b_i(\mathbf{k})|^2 = 1 and the Bloch sums normalized to unity per \mathbf{k}-point.[11] For finite crystals with N sites, the sum over \mathbf{R} is truncated, introducing corrections to the normalization; the factor N^{-1/2} ensures the wavefunction amplitude scales correctly, approaching the infinite-crystal limit as N \to \infty. This derivation highlights how translational symmetry not only dictates the form of the wavefunction but also constrains its normalization through the periodic structure.[1]Mathematical Formulation
The Tight-Binding Hamiltonian
The tight-binding approximation begins with the full Hamiltonian of the crystal, \hat{H} = \hat{T} + \hat{V}, where \hat{T} is the kinetic energy operator and \hat{V} represents the total potential energy arising from interactions with ionic cores and other electrons.[12] This Hamiltonian governs the many-electron problem in the solid, but solving it exactly is computationally prohibitive for extended systems. To make progress, the method employs a variational approach, seeking approximate solutions to the Schrödinger equation by projecting onto a basis of localized atomic-like orbitals \{\phi_\mu\}, typically centered on lattice sites and resembling atomic orbitals. The trial wavefunction is constructed as a linear combination \psi = \sum_\mu c_\mu \phi_\mu, and the energy is minimized subject to normalization, leading to the effective tight-binding Hamiltonian in this basis.[12] The projected tight-binding Hamiltonian takes the form \hat{H}_\text{TB} = \sum_{\mu\nu} |\phi_\mu\rangle H_{\mu\nu} \langle \phi_\nu|, where the matrix elements are H_{\mu\nu} = \langle \phi_\mu | \hat{H} | \phi_\nu \rangle. These elements capture the essential physics: the diagonal on-site terms \varepsilon_\mu = H_{\mu\mu} = \langle \phi_\mu | \hat{H} | \phi_\mu \rangle correspond to the energy levels of isolated atoms, shifted by the crystalline environment, while the off-diagonal hopping terms t_{\mu\nu} = H_{\mu\nu} for \mu \neq \nu describe the amplitude for an electron to transfer between orbitals \phi_\mu and \phi_\nu due to overlap and potential coupling.[12] In practice, hopping is assumed to decay rapidly with inter-orbital distance, allowing truncation to nearest-neighbor interactions for efficiency without significant loss of accuracy. For periodic crystals, translational symmetry is exploited via Bloch's theorem, transforming the basis into Bloch states \phi_{\mu \mathbf{k}}(\mathbf{r}) = \frac{1}{\sqrt{N}} \sum_{\mathbf{R}} e^{i \mathbf{k} \cdot \mathbf{R}} \phi_\mu(\mathbf{r} - \mathbf{R}), where \mathbf{R} are lattice vectors and N is the number of unit cells. The resulting \mathbf{k}-dependent Hamiltonian in reciprocal space is the Fourier transform H_{\mu\nu}(\mathbf{k}) = \sum_{\mathbf{R}} e^{i \mathbf{k} \cdot \mathbf{R}} H_{\mu\nu}(\mathbf{R}), with a similar form for the overlap matrix S_{\mu\nu}(\mathbf{k}) = \sum_{\mathbf{R}} e^{i \mathbf{k} \cdot \mathbf{R}} S_{\mu\nu}(\mathbf{R}), where S_{\mu\nu}(\mathbf{R}) = \langle \phi_\mu | \phi_\nu(\mathbf{R}) \rangle accounts for non-orthogonality of the localized basis.[12] The eigenenergies E_n(\mathbf{k}) are obtained by solving the secular equation \det \left[ H(\mathbf{k}) - E_n(\mathbf{k}) S(\mathbf{k}) \right] = 0, or equivalently, the generalized eigenvalue problem \sum_\nu [H_{\mu\nu}(\mathbf{k}) - E_n(\mathbf{k}) S_{\mu\nu}(\mathbf{k})] c_{\nu n}(\mathbf{k}) = 0, which yields the band structure E_n(\mathbf{k}) throughout the Brillouin zone. This formulation, derived variationally, provides a tractable means to compute electronic properties while retaining the periodicity of the lattice.[12]Tight-Binding Matrix Elements
In the tight-binding model, the matrix elements of the Hamiltonian are defined in the basis of localized atomic orbitals \phi_\mu(\mathbf{r}) centered at lattice sites. The general form of these elements is given by H_{\mu\nu}(\mathbf{R}) = \langle \phi_\mu(\mathbf{r}) | \hat{H} | \phi_\nu(\mathbf{r} - \mathbf{R}) \rangle, where \mathbf{R} is the lattice vector separating the sites of the two orbitals, and \hat{H} is the single-particle Hamiltonian including kinetic energy and potential terms. This expression captures the overlap integrals between orbitals on different sites, forming the core of the model's semi-empirical parameterization.[1] The diagonal matrix elements, known as on-site energies, correspond to \mathbf{[R](/page/R)} = [0](/page/0) and are denoted \varepsilon_\mu = \langle \phi_\mu | \hat{[H](/page/H+)} | \phi_\mu \rangle. These represent the energy levels of isolated atoms adjusted for the crystalline environment and can vary by orbital type (e.g., s, p, d). In alloys or heterostructures, the on-site energies may be site-dependent to account for compositional variations, such as different atomic species at lattice positions.[1] Off-diagonal elements, or hopping integrals, arise for \mathbf{R} \neq 0 and are typically denoted t_{\mu\nu}(\mathbf{R}) = \langle \phi_\mu | \hat{H} | \phi_\nu(\mathbf{r} - \mathbf{R}) \rangle. These describe electron transfer between sites and decay rapidly with increasing distance |\mathbf{R}|, often exponentially due to the localized nature of the orbitals, justifying truncation in practical calculations.[3] Symmetry properties impose significant constraints on the matrix elements. Hermiticity of the Hamiltonian ensures H_{\mu\nu}(\mathbf{R}) = H_{\nu\mu}(-\mathbf{R})^*, while time-reversal symmetry further relates elements involving complex orbitals. Lattice symmetries, such as those of the point group at each site, reduce the number of independent parameters by relating integrals based on rotations and inversions, as systematized in the Slater-Koster formalism. In reciprocal space, exploiting translational invariance, the matrix elements become k-dependent: for diagonal terms, H_{\mu\mu}(\mathbf{k}) = \varepsilon_\mu, independent of \mathbf{k}; for off-diagonal terms, H_{\mu\nu}(\mathbf{k}) = \sum_{\mathbf{R}} e^{i \mathbf{k} \cdot \mathbf{R}} t_{\mu\nu}(\mathbf{R}), where the sum is over all lattice vectors \mathbf{R} connecting the \nu orbital to the \mu site.[1] This Fourier transform yields the Bloch Hamiltonian whose eigenvalues give the band structure. A common simplification is the nearest-neighbor approximation, where the sum over \mathbf{R} is truncated to the smallest distances (e.g., first or second neighbors), as farther hoppings contribute negligibly due to their decay. This reduces computational complexity while retaining essential band features, particularly in simple lattices like face-centered cubic metals.[1][3]Evaluation and Connections
Evaluation of Matrix Elements
The evaluation of tight-binding matrix elements t_{\alpha \beta}(\mathbf{R}) = \langle \phi_\alpha(\mathbf{r}) | H | \phi_\beta(\mathbf{r} - \mathbf{R}) \rangle typically relies on approximations that balance computational feasibility with accuracy, focusing on two-center interactions between atomic orbitals on neighboring sites.[8] These elements capture the hopping amplitudes and on-site energies essential for constructing the Hamiltonian, often parameterized through analytical expressions or numerical fits.[8] A foundational analytical approach is the Slater-Koster interpolation scheme, which expresses the matrix elements t_{l m, l' m'}(\mathbf{R}) between orbitals of angular momentum l, l' and magnetic quantum numbers m, m' in terms of a few two-center integrals, such as (ss\sigma), (sp\sigma), (pd\sigma), and their \pi counterparts, modulated by angular factors derived from spherical harmonics.[8] This method assumes that interactions beyond nearest neighbors decay rapidly and that the potential is dominated by pairwise atomic contributions, allowing the full directional dependence to be captured via direction cosines (l, m, n) along the bond vector \mathbf{R}.[8] For instance, the hopping between an s-orbital and a p-orbital is given by t_{s p_x} = (sp\sigma) l, where l = \cos\theta is the direction cosine relative to the x-axis.[8] The angular factors in Slater-Koster tables systematically tabulate these dependencies for common orbital pairs, enabling efficient computation without explicit integration over the Hamiltonian.[8] Representative examples for p-d interactions include:| Orbital Pair | \sigma Factor | \pi Factor |
|---|---|---|
| p_x d_{x^2 - y^2} | \sqrt{3} \sin^2\theta \cos^2\phi | -\sqrt{3} \sin 2\theta \cos\phi / 2 |
| p_x d_{xz} | \sqrt{3} \cos\theta \sin\theta \cos\phi | (\cos^2\theta - \sin^2\theta) \cos\phi / 2 |
| p_z d_{z^2} | \cos\theta | -\sin\theta / \sqrt{3} |
Connection to Wannier Functions
Wannier functions provide a localized representation of the electronic states in a crystal, obtained as the inverse Fourier transform of the delocalized Bloch wavefunctions. For a band indexed by n, the Wannier function centered at lattice site \mathbf{R} is given by w_n(\mathbf{R}, \mathbf{r}) = \frac{1}{\sqrt{N}} \sum_{\mathbf{k}} e^{-i \mathbf{k} \cdot \mathbf{R}} \psi_{n \mathbf{k}}(\mathbf{r}), where N is the number of unit cells, the sum is over the Brillouin zone, and \psi_{n \mathbf{k}}(\mathbf{r}) are the Bloch states.[18] This transformation bridges the periodic, extended nature of Bloch functions to localized orbitals, facilitating interpretations akin to atomic-like states in solids.[19] In the tight-binding approximation, particularly for isolated energy bands that are narrow and well-separated from others, the Wannier functions closely resemble the atomic orbitals from which the model is constructed. This approximation holds when the wavefunctions are strongly localized around atomic sites, with minimal overlap between neighboring sites, allowing the Wannier functions to serve as an effective basis for modeling interatomic interactions.[19] The overlap integral between a Wannier function w_n(\mathbf{R}, \mathbf{r}) and the corresponding atomic orbital \phi_a(\mathbf{r} - \mathbf{R}) approaches unity in this limit, quantifying the fidelity of the approximation and enabling the projection of ab initio results onto a minimal atomic basis.[19] To achieve the most localized representation, especially for composite bands spanning multiple atomic orbitals, maximally localized Wannier functions (MLWFs) are constructed by minimizing a localization functional known as the spread \Omega. This functional measures the spatial delocalization and is defined as \Omega = \sum_n \left[ \langle w_n | r^2 | w_n \rangle - |\langle w_n | \mathbf{r} | w_n \rangle|^2 \right], where the sums are over Wannier functions within the unit cell.[20] The spread \Omega decomposes into a gauge-invariant part \Omega_I, which cannot be altered, and gauge-dependent parts \tilde{\Omega} and \Omega_{OD} that account for intra-cell off-diagonal and inter-band off-diagonal contributions, respectively: \Omega = \Omega_I + \tilde{\Omega} + \Omega_{OD}. Minimization proceeds by optimizing unitary transformations U_{mn}(\mathbf{k}) that rotate the Bloch subspaces at each \mathbf{k}-point, effectively disentangling entangled bands to yield localized functions; this is achieved via iterative steepest-descent algorithms on the unitary matrices.[20] The resulting MLWFs preserve the symmetries of the underlying bands while maximizing localization.[21] The tight-binding Hamiltonian in the Wannier basis takes the form of a real-space hopping matrix, with elements H_{mn}(\mathbf{R} - \mathbf{R}') = \langle w_m(\mathbf{R}) | \hat{H} | w_n(\mathbf{R}') \rangle, which directly corresponds to the site-to-site transfer integrals in the atomic orbital basis when the Wannier functions are orthogonal and approximate those orbitals.[19] For entangled bands, the transformation U_{mn}(\mathbf{k}) disentangles the subspace, projecting onto a smooth, isolated set of bands to construct the effective Hamiltonian. This equivalence underscores how MLWFs enable unbiased, quantitative tight-binding models derived from first-principles calculations.[20] These localized Wannier representations are particularly valuable for applications in electron transport, where hopping parameters inform conductivity, and in defect physics, where localized states around impurities can be analyzed.[21] The overlap between MLWFs and atomic orbitals further allows for efficient interpolation of band structures and forces across the Brillouin zone, enhancing computational efficiency in large-scale simulations.[19]Formalisms and Extensions
Second Quantization
In the tight-binding model, second quantization provides a framework to describe many-electron systems by expressing the Hamiltonian in terms of fermionic creation and annihilation operators, facilitating the inclusion of electron interactions and correlations beyond the single-particle approximation.[22] This formulation starts from the first-quantized tight-binding Hamiltonian and maps it to an operator form suitable for many-body perturbation theory and numerical methods.[22] The fundamental operators are the creation operator c_{i\sigma}^\dagger and annihilation operator c_{i\sigma}, which add or remove an electron with spin \sigma at lattice site i, respectively. These satisfy the fermionic anticommutation relations \{ c_{i\sigma}, c_{j\sigma'}^\dagger \} = \delta_{ij} \delta_{\sigma\sigma'} and \{ c_{i\sigma}, c_{j\sigma'} \} = \{ c_{i\sigma}^\dagger, c_{j\sigma'}^\dagger \} = 0, ensuring the Pauli exclusion principle.[22] The number operator is defined as n_{i\sigma} = c_{i\sigma}^\dagger c_{i\sigma}, with total occupancy at site i given by n_i = n_{i\uparrow} + n_{i\downarrow}.[22] The second-quantized tight-binding Hamiltonian for interacting electrons is H = \sum_{i,j,\sigma} t_{ij} c_{i\sigma}^\dagger c_{j\sigma} + \frac{1}{2} \sum_{i,j,k,l,\sigma,\sigma'} V_{ijkl} c_{i\sigma}^\dagger c_{j\sigma'}^\dagger c_{l\sigma'} c_{k\sigma}, where the first term represents kinetic energy via hopping amplitudes t_{ij}, and the second term accounts for two-body interactions with matrix elements V_{ijkl}.[22] For the non-interacting case, the interaction term vanishes, yielding H_0 = \sum_{i,j,\sigma} t_{ij} c_{i\sigma}^\dagger c_{j\sigma}. Under translational invariance, this diagonalizes in momentum space via the Fourier transform c_{k\sigma} = \frac{1}{\sqrt{N}} \sum_i e^{-i \mathbf{k} \cdot \mathbf{R}_i} c_{i\sigma}, resulting in H_0 = \sum_{\mathbf{k},\sigma} \varepsilon_{\mathbf{k}} c_{\mathbf{k}\sigma}^\dagger c_{\mathbf{k}\sigma}, where \varepsilon_{\mathbf{k}} is the dispersion relation and N is the number of sites.[22] To incorporate on-site electron repulsion, the Hubbard model adds the term H_U = U \sum_i n_{i\uparrow} n_{i\downarrow}, where U > 0 is the Coulomb repulsion strength, restricting double occupancy and capturing Mott insulation at half-filling for large U.[23] This interaction arises from the leading V_{iiii} term in the general potential, with off-site contributions often neglected for simplicity.[23] The derivation from the first-quantized to second-quantized form involves expanding the many-body wave function in the tight-binding basis of localized orbitals and substituting into the expectation value of the Hamiltonian, yielding the operator expression through Wick's theorem or direct mapping for fermions; in one dimension, the Jordan-Wigner transformation can relate spin models to this fermionic representation, though direct quantization suffices for electron systems.[22] Mean-field approximations, such as Hartree-Fock, treat interactions by decoupling the four-operator terms: for the Hubbard U term, n_{i\uparrow} n_{i\downarrow} \approx \langle n_{i\uparrow} \rangle n_{i\downarrow} + n_{i\uparrow} \langle n_{i\downarrow} \rangle - \langle n_{i\uparrow} \rangle \langle n_{i\downarrow} \rangle, leading to an effective single-particle Hamiltonian with renormalized site energies and hoppings that can produce magnetic order like antiferromagnetism. This approximation captures qualitative features of correlation effects while remaining tractable within the tight-binding framework.Modern Extensions
Modern extensions of the tight-binding model have addressed limitations in handling complex interactions, environmental dependencies, and computational scalability, particularly for materials beyond simple periodic lattices. Self-consistent tight-binding approaches integrated with GW approximations or DFT embedding improve screening effects by iteratively updating the electronic structure to account for many-body interactions and dielectric responses.[24] These methods embed tight-binding Hamiltonians within higher-level frameworks, such as GW for quasiparticle corrections, to enhance accuracy in predicting band gaps and excitations in semiconductors and insulators while maintaining computational efficiency.[25] Incorporation of multi-orbital basis sets and spin-orbit coupling has extended the model to topological materials, enabling the description of symmetry-protected edge states and nontrivial band topologies. The Kane-Mele model, a seminal tight-binding formulation for graphene, adds intrinsic spin-orbit interactions to the nearest-neighbor hopping terms, opening a topological gap and realizing the quantum spin Hall effect in two dimensions.[26] This extension has been pivotal for modeling 2D materials like transition metal dichalcogenides, where spin-orbit effects drive valleytronics and topological phase transitions.[27] To capture angular dependencies and environmental effects, three-body tight-binding models introduce interactions involving bond angles and triplet atomic configurations, improving predictions of energies and forces in covalent and metallic systems. A 2023 model parametrizes these three-body terms for 65 elements, achieving near-DFT accuracy for band structures and phonons in diverse crystals at a fraction of the computational cost.[28] Such environment-dependent hoppings account for structural distortions and coordination variations, essential for accurate simulations of amorphous or defective materials. Semi-empirical parametrizations like GFN1-xTB (2017) and GFN2-xTB (2019) refine the tight-binding framework for broad applicability in molecular dynamics and non-covalent interactions, with parameters fitted to DFT reference data for elements up to radon. GFN1-xTB emphasizes geometries, frequencies, and thermochemistry, offering robust performance for organic and inorganic clusters. GFN2-xTB enhances this with multipole electrostatics and density-dependent dispersion, yielding mean absolute errors below 3 kcal/mol for interaction energies in biomolecular systems and nanostructures.[29] These methods excel in applications to 2D materials and nanomaterials, such as optimizing graphene derivatives or metal-organic frameworks, where they balance speed and fidelity for large-scale simulations.[30] Machine learning parameterizations, particularly post-2020 neural network approaches, learn hopping integrals and on-site energies directly from ab initio datasets, enabling transferable tight-binding models for disordered or defect-laden systems. Neural networks trained on projected density of states predict sparse tight-binding parameters for point defects in semiconductors, reducing parameterization effort while preserving electronic structure fidelity.[31] These advances facilitate rapid prototyping of models for nanostructures, with accuracies approaching DFT for band gaps and defect levels in semiconductors. Recent 2025 developments, such as GPU-accelerated methods like GPUTB, further enable efficient simulations of large-scale systems with ab initio precision.[32] Despite these improvements, tight-binding models remain approximate for strongly correlated systems, where local Coulomb interactions lead to Mott insulation or magnetic ordering not captured by single-particle hoppings. Hybrids with dynamical mean-field theory (DMFT) address this by embedding correlated orbitals within a tight-binding lattice, self-consistently solving impurity problems to include fluctuation effects and spectral functions.[33] DFT+DMFT implementations accurately describe transition metal oxides, revealing correlation-driven band renormalizations beyond standard tight-binding scope.Examples
One-Dimensional s-Band
The tight-binding model provides a simple yet insightful framework for understanding electronic band formation in a one-dimensional (1D) lattice of atoms, each contributing an s-orbital. Consider an infinite chain of identical atoms spaced by lattice constant a, where the on-site energy is \epsilon = 0 and electrons hop only between nearest neighbors with hopping amplitude t < 0. This setup approximates the electrons as localized on atomic sites, with weak overlap leading to band dispersion.[1] To derive the energy dispersion, apply Bloch's theorem, expressing the wavefunction as \psi_k(r) = \frac{1}{\sqrt{N}} \sum_n e^{i k n a} \phi_s(r - n a), where \phi_s is the s-orbital centered at site n, N is the number of sites, and k is the wavevector in the first Brillouin zone -\pi/a < k \leq \pi/a. The Hamiltonian in second-quantized form is H = \sum_n \epsilon c_n^\dagger c_n + t \sum_n (c_n^\dagger c_{n+1} + c_{n+1}^\dagger c_n), but in the single-particle basis, the matrix elements yield the secular equation. For the periodic chain, Fourier transform to k-space gives the 1x1 Hamiltonian matrix H(k) = 2 t \cos(ka), as off-diagonal terms from distant sites are neglected. Solving the eigenvalue problem, the dispersion relation is E(k) = -2 |t| \cos(ka), with the band extending from -2|t| to +2|t| and bandwidth $4|t|.[34][1] This cosine dispersion contrasts with the free-electron parabolic form E(k) = \frac{\hbar^2 k^2}{2m}, which is nearly linear near k=0 but lacks the flattening at zone edges due to lattice periodicity. In the tight-binding case, the band is symmetric around E=0, with group velocity v_g = \frac{dE}{dk} = 2 |t| a \sin(ka) vanishing at the zone center (k=0) and boundaries (k = \pm \pi/a), reflecting reduced electron mobility at band edges.[1] At half-filling (one electron per site, ignoring spin for simplicity), the Fermi level lies at E_F = 0, filling states up to k_F = \pm \pi/(2a). The Fermi "surface" reduces to two points in 1D, separating occupied and unoccupied states. The density of states per site is g(E) = \frac{1}{\pi \sqrt{4 t^2 - E^2}} for |E| < 2|t|, diverging as $1/\sqrt{|E \pm 2|t||} at the band edges due to the flat dispersion, which enhances van Hove singularities compared to the free-electron g(E) \propto 1/\sqrt{E}.[35] Extensions beyond nearest-neighbor hopping introduce next-nearest terms t', modifying E(k) = -2|t| \cos(ka) - 2|t'| \cos(2ka), which can skew the band and alter the Fermi surface. Dimerization, as in the Su-Schrieffer-Heeger (SSH) model with alternating hoppings |t| (1 \pm \delta), opens a gap at half-filling and hosts topologically protected zero-energy edge states in finite chains, illustrating how lattice distortions stabilize solitons in conducting polymers like polyacetylene.[36]Interatomic Matrix Elements
In the tight-binding model, interatomic matrix elements, often referred to as hopping integrals, describe the overlap between atomic orbitals on neighboring sites and are approximated using the two-center integral approach, neglecting contributions from three or more centers. These integrals are classified by the orbital types involved (e.g., s-s, s-p, p-p, s-d) and the symmetry of the bond (σ for head-on overlap or π for side-on overlap). For common sp-bonded systems, the key two-center integrals include ssσ, spσ, ppσ, and ppπ, with additional sdσ, pdσ, pdπ, ddσ, ddπ, and ddδ for transition metals incorporating d-orbitals. Their radial dependence typically follows a power-law decay proportional to 1/d² for nearest-neighbor distances d, as derived from Harrison's universal tight-binding theory, though longer-range interactions often incorporate exponential decay for improved accuracy.[37] Harrison's universal parameters provide a scalable framework for these hopping integrals across the periodic table, expressed as V_{ll'l''} = \eta_{ll'l''} \left( \frac{\hbar^2}{m d^2} \right), where \eta are dimensionless coefficients, \hbar is the reduced Planck's constant, m is the electron mass, and d is the interatomic distance. Standard values for the \eta parameters in sp systems are \eta_{ss\sigma} = -1.32, \eta_{sp\sigma} = 1.42, \eta_{pp\sigma} = 2.22, and \eta_{pp\pi} = -0.63, where the signs reflect bonding and antibonding character in σ and π overlaps.[37] For silicon, using d \approx 2.35 Å and \hbar^2 / m \approx 7.62 eV Ų, this yields V_{ss\sigma} \approx -1.8 eV, illustrating the scale for semiconductors; these parameters are adjusted element-specifically via a scaling factor \gamma_s but maintain universal ratios for qualitative band structure predictions. The Slater-Koster formalism expresses these two-center integrals in terms of direction cosines (l, m, n) along the bond axis, enabling computation of the full Hamiltonian matrix. For sp³-hybridized systems like diamond-structure semiconductors, the effective σ-bond hopping in hybrids is h_\sigma = \frac{1}{4} V_{ss\sigma} + \frac{\sqrt{3}}{2} V_{sp\sigma} + \frac{3}{4} V_{pp\sigma} and the π-bond hopping is h_\pi = \frac{3}{4} V_{pp\pi}, capturing directional bonding in structures such as Si or GaAs. For d-orbitals in metals, integrals like sdσ and ddσ incorporate additional angular momentum, with canonical ratios ddσ:ddπ:ddδ = -6:4:-1 for qualitative descriptions. Material-specific parameters are often fitted to ab initio or experimental band structures for quantitative accuracy. In semiconductors, nearest-neighbor values dominate, while in metals, second-neighbor contributions are significant due to broader d-band overlap. The distance dependence beyond nearest neighbors is modeled as t(R) = t_0 e^{-q(R - a)}, where a is the equilibrium bond length, q \approx 1-2 Å^{-1}, and t_0 is the nearest-neighbor value, ensuring rapid decay and computational efficiency.[37]| Orbital Pair | Bond Type | Typical Magnitude (eV) | Material Example |
|---|---|---|---|
| s-s | σ | -2.3 | Si (Vogl et al.) |
| s-p | σ | +2.1 | Si (Vogl et al.) |
| p-p | σ | +3.9 | Si (Vogl et al.) |
| p-p | π | +1.4 | Si (Vogl et al.) |
| s-s | σ | -1.9 | GaAs (Jancu et al.) |
| s-p | σ | +2.1 | GaAs (Jancu et al.) |
| p-p | σ | +3.6 | GaAs (Jancu et al.) |
| p-p | π | +1.4 | GaAs (Jancu et al.) |
| s-d | σ | -1.4 | Cu (Papaconstantopoulos) |
| d-d | σ | -0.07 | Cu (Papaconstantopoulos) |
| d-d | π | +0.04 | Cu (Papaconstantopoulos) |