This package provides subroutines representing and manipulating Green's functions in the imaginary time or the Matsubara-frequency domain.
Imaginary-time Green's functions encode the thermodynamic properties of quantum many-body systems. They are typically singular and hard to deal with at low temperatures in numerical calculations.
We provide the following components to ease the numerical manipulation of the Green's functions:
-
Algorithms to generate the discrete Lehamnn representation (DLR), a generic and compact representation of Green's functions proposed in Ref. [1]. DLR generally only requires
$\sim \log(1/T)\log(1/\epsilon)$ numbers to represent a Green's function at a temperature T up to a given accuracy ϵ. This package provides two algorithms: one algorithm is based on a conventional QR algorithm, and another is based on a functional QR algorithm. The latter extends DLR to extremely low temperatures. -
Dedicated DLR for Green's functions with the particle-hole symmetry (e.g., phonon propagator) or particle-hole antisymmetry (e.g., superconductor gap function).
-
Fast and accurate Fourier transform between the imaginary-time domain and the Matsubara-frequency domain with a cost
$O(\log(1/T)\log(1/\epsilon))$ and an accuracy$\sim 100\epsilon$ . -
Fast and accurate Green's function interpolation with a cost
$O(\log(1/T)\log(1/\epsilon))$ and accuracy$\sim 100\epsilon$ . -
Fit a Green's function with noise.
This package has been registered. So, type import Pkg; Pkg.add("Lehmann") in the Julia REPL to install.
In the following demo, we will show how to compress a Green's function of ~10000 data points into ~20 DLR coefficients and perform fast interpolation and Fourier transform up to the accuracy ~1e-10.
using Lehmann
β = 100.0 # inverse temperature
Euv = 1.0 # ultraviolt energy cutoff of the Green's function
rtol = 1e-8 # accuracy of the representation
isFermi = false
symmetry = :none # :ph if particle-hole symmetric, :pha is antisymmetric, :none if there is no symmetry
diff(a, b) = maximum(abs.(a - b)) # return the maximum deviation between a and b
dlr = DLRGrid(Euv, β, rtol, isFermi, symmetry) #initialize the DLR parameters and basis
# A set of most representative grid points are generated:
# dlr.ω gives the real-frequency grids
# dlr.τ gives the imaginary-time grids
# dlr.ωn and dlr.n gives the Matsubara-frequency grids. The latter is the integer version.
println("Prepare the Green's function sample ...")
Nτ, Nωn = 10000, 10000 # many τ and n points are needed because Gτ is quite singular near the boundary
τgrid = collect(LinRange(0.0, β, Nτ)) # create a τ grid
Gτ = Sample.SemiCircle(dlr, :τ, τgrid) # Use semicircle spectral density to generate the sample Green's function in τ
ngrid = collect(-Nωn:Nωn) # create a set of Matsubara-frequency points
Gn = Sample.SemiCircle(dlr, :n, ngrid) # Use semicircle spectral density to generate the sample Green's function in ωn
println("Compress Green's function into ~20 coefficients ...")
spectral_from_Gτ = tau2dlr(dlr, Gτ, τgrid)
spectral_from_Gω = matfreq2dlr(dlr, Gn, ngrid)
# You can use the above functions to fit noisy data by providing the named parameter ``error``
println("Prepare the target Green's functions to benchmark with ...")
τ = collect(LinRange(0.0, β, Nτ * 2)) # create a dense τ grid to interpolate
Gτ_target = Sample.SemiCircle(dlr, :τ, τ)
n = collect(-2Nωn:2Nωn) # create a set of Matsubara-frequency points
Gn_target = Sample.SemiCircle(dlr, :n, n)
println("Interpolation benchmark ...")
Gτ_interp = dlr2tau(dlr, spectral_from_Gτ, τ)
println("τ → τ accuracy: ", diff(Gτ_interp, Gτ_target))
Gn_interp = dlr2matfreq(dlr, spectral_from_Gω, n)
println("iω → iω accuracy: ", diff(Gn_interp, Gn_target))
println("Fourier transform benchmark...")
Gτ_to_n = dlr2matfreq(dlr, spectral_from_Gτ, n)
println("τ → iω accuracy: ", diff(Gτ_to_n, Gn_target))
Gn_to_τ = dlr2tau(dlr, spectral_from_Gω, τ)
println("iω → τ accuracy: ", diff(Gn_to_τ, Gτ_target))A set of basis files have been precalculated and stored in the folder basis. They cover most of the use cases. For edge cases, you may generate your own basis file use this script.
In the above script, user can choose the folder to store the generated basis file. To use the new basis file, pass the folder as an argument when creating DLRGrid struct. More information can be found in the documentation
If this library helps you to create software or publications, please let us know and cite
libdlr by Jason Kaye and Hugo U.R. Strand.
Contributions are very welcome, as are feature requests and suggestions. Please open an issue if you encounter any problems.