Skip to main content

Crate vanhove

Crate vanhove 

Source
Expand description

§vanhove

CI crates.io docs.rs license

Model densities of states with integrable van Hove singularities, their convolutions, and accurate integration of spectral functions.

Many quantities in condensed matter physics are frequency integrals $\int A(\omega) f(\omega) d\omega$ of a spectral function $A(\omega)$ against some function $f(\omega)$. Doing such integrals numerically is awkward whenever $A(\omega)$ has van Hove singularities: they defeat generic adaptive quadrature even though the integrals themselves converge.

This crate handles them by splitting each singular model into a smooth part $R(\omega)$ and analytically known asymptotics $S_p(\omega)$ around every singular point $\Omega_p$,

$$ \int A(\omega) f(\omega) d\omega = \int R(\omega) f(\omega) d\omega + \sum_p \int S_p(\omega) [f(\omega) - f(\Omega_p)] d\omega + \sum_p f(\Omega_p) \int S_p(\omega) d\omega. $$

Only the first two terms are integrated numerically, and both integrands are regular; the last term uses a closed-form value of $\int S_p(\omega) d\omega$. The numerical work is delegated to the bilby quadrature.

§Models

FunctionDensity of states
discrete(levels, weights)Sum of discrete levels, $\sum_p w_p \delta(\omega - \varepsilon_p)$
flat(eps, d, delta)Flat band with Fermi-like edges, sharp for delta = 0
gaussian(eps, sigma)Gaussian
semicircle(eps, r)Semicircle (Wigner), square-root band edges
powerlaw(eps, r, w)One-sided power-law band edge, divergent for r < 0
pseudogap(eps, r, d)Pseudogap, $\lvert\omega-\epsilon\rvert^r$ at the band center
chain(eps, t)Linear chain, square-root divergent edges
bethe(z, eps, t)Bethe lattice of coordination number z, square-root band edges
square(eps, t)Square lattice, logarithmic van Hove peak
simple_cubic(eps, t)Simple cubic lattice, square-root cusps at four van Hove points
triangular(eps, t)Triangular lattice, asymmetric with a logarithmic peak
honeycomb(eps, t)Honeycomb lattice, two logarithmic peaks about a Dirac point
kagome(eps, t)Kagome lattice, flat band $\delta$-peak at a band edge
lieb(eps, t)Lieb lattice, flat band $\delta$-peak at the band center

Every model except discrete() is normalized to unit spectral weight. Spectral functions can be scaled by real numbers, negated, added and subtracted, so mixed discrete/continuous spectra are built by simple arithmetic. They can also be displaced in frequency (shifted()), reflected (mirrored()) and convolved (conv()). broadened() evaluates a Lorentzian-broadened spectral function at a given frequency, and precomputed() replaces a regular part that is expensive to evaluate with a Chebyshev interpolation of it.

A model defined outside the crate implements the ContinuousSF trait, supplying its support, its regular part and the closed form of each singular part, and SpectralFunction::from_continuous() turns it into a spectral function.

§Convolution

conv() computes the convolution $\int A(\nu) B(\omega - \nu) d\nu$ of two spectral functions. The singular structure of the result is derived in closed form: its singular points are the pairwise sums of the frequencies where either operand stops being smooth, and the asymptotics at each of them follows from those of the operands. Only the smooth remainder of the convolution is interpolated, so the result can be integrated and convolved again like any built-in model. The continuous parts of both operands must have bounded supports.

§Usage

[dependencies]
vanhove = "0.2"
use num_complex::Complex64;
use vanhove::models::{chain, discrete, semicircle, square};

// 80% square-lattice band plus 20% of a single discrete level at ω = 3
let dos = 0.8 * square(0.0, 1.0) + 0.2 * discrete(&[3.0], &[1.0]);
assert_eq!(dos.total_weight(), 1.0);

// First spectral moment ∫A(ω) ω dω
let m1 = dos.integrate(|omega| omega, None).unwrap();
assert!(m1.converged);
assert!((m1.value - 0.6).abs() < 1e-10);

// Retarded Green's function of a semicircular band at z = 0.5 + 10⁻³i,
// just above the real axis
let z = Complex64::new(0.5, 1e-3);
let g = semicircle(0.0, 2.0)
    .integrate_complex(|omega| 1.0 / (z - omega), None)
    .unwrap();
assert!(g.value.im < 0.0);

// Two linear chains convolve into the square lattice
let sq = chain(0.0, 1.0).conv(&chain(0.0, 1.0), None);
assert!((sq.continuous_at(1.0) - square(0.0, 1.0).continuous_at(1.0)).abs() < 1e-9);

integrate() takes an optional absolute tolerance (1e-10 by default) and returns a Result, since the underlying adaptive quadrature refuses a request it cannot make sense of, such as a tolerance that is not a number. A request it accepts yields an Integral, which contains the value, an estimate of its absolute error, and whether each quadrature it is assembled from has reached the tolerance.

SpectralFunction is Send and Sync, so a frequency scan can be spread over threads, each evaluating continuous_at(), integrate() or broadened() at frequencies of its own.

§Documentation

API documentation is available at https://docs.rs/vanhove. The mathematics behind the splitting, integrate() and conv() is laid out in the theory module.

Formulas in the docs are rendered with KaTeX, so building them locally requires the extra header:

RUSTDOCFLAGS="--html-in-header katex-header.html" cargo doc --no-deps --open

§License

Licensed under either of

at your option.

Modules§

discrete
Discrete spectral function.
interp
Chebyshev interpolation of the regular part of a spectral function.
models
Model densities of states commonly used in physics.
segment
Segments of the frequency axis.
singularity
Closed-form description of an integrable singularity of a spectral function.
theory
The mathematics underneath.

Structs§

Integral
Integral computed by quadrature, with an estimate of its absolute error.
SpectralFunction
Spectral function as a weighted sum of discrete resonances and continuous contributions.

Traits§

ContinuousSF
Continuous spectral function possibly containing integrable singularities.