minor change

Author: fsxbhyy

test/grid.jl

@@ -0,0 +1,185 @@
+function fine_ωGrid_test(Λ::Float, degree, ratio::Float) where {Float}
+    ############## use composite grid #############################################
+    N = Int(floor(log(Λ) / log(ratio) + 1))
+
+    grid = CompositeGrid.LogDensedGrid(
+        :cheb,# The top layer grid is :gauss, optimized for integration. For interpolation use :cheb
+        [0.0, Λ],# The grid is defined on [0.0, β]
+        [0.0,],# and is densed at 0.0 and β, as given by 2nd and 3rd parameter.
+        N,# N of log grid
+        Λ / ratio^(N - 1), # minimum interval length of log grid
+        degree, # N of bottom layer
+        Float
+    )
+
+    return grid
+    #return vcat(-grid[end:-1:1], grid)
+end
+
+function fine_τGrid_test(Λ::Float,degree,ratio::Float) where {Float}
+    ############## use composite grid #############################################
+    # Generating a log densed composite grid with LogDensedGrid()
+    npo = Int(ceil(log(Λ) / log(ratio))) - 2 # subintervals on [0,1/2] in tau space (# subintervals on [0,1] is 2*npt)
+    grid = CompositeGrid.LogDensedGrid(
+        :gauss,# The top layer grid is :gauss, optimized for integration. For interpolation use :cheb
+        [0.0, 1.0],# The grid is defined on [0.0, β]
+        [0.0, 1.0],# and is densed at 0.0 and β, as given by 2nd and 3rd parameter.
+        npo,# N of log grid
+        0.5 / ratio^(npo-1), # minimum interval length of log grid
+        degree, # N of bottom layer
+        Float
+    )
+    #print(grid[1:length(grid)÷2+1])    
+    #print(grid+reverse(grid))
+    # println("Composite expoential grid size: $(length(grid))")
+    println("fine grid size: $(length(grid)) within [$(grid[1]), $(grid[end])]")
+    return grid
+
+end
+
+@inline function A1(L::T) where {T}
+
+    return T(2 * expinti(-L) - 2 * expinti(-2 * L) - exp(-2 * L) * (exp(L) - 1)^2 / L)
+
+end
+
+@inline function A2(a::T, beta::T, L::T) where {T}
+
+    return T(expinti(-a * L) - expinti(-(a + beta) * L))
+    #return T(-shi(a * L) + shi((a + 1) * L))
+end
+
+"""
+``F(x, y) = (1-exp(x+y))/(x+y)``
+"""
+@inline function A3(a::T, b::T, L::T) where {T}
+    lamb = (a + b) * L
+    if abs(lamb) > Tiny
+        return (1 - exp(-lamb)) / (a + b)
+    else
+        return T(L * (1 - lamb / 2.0 + lamb^2 / 6.0 - lamb^3 / 24.0))
+    end
+end
+
+function uni_ngrid( isFermi, Λ::Float) where {Float}
+    ngrid = Float[]
+    n = 0
+    while (2n+1)*π<Λ
+        append!(ngrid, n)
+        n += 1
+    end
+    if isFermi
+        return vcat(-ngrid[end:-1:1] .-1, ngrid)
+    else
+        return  vcat(-ngrid[end:-1:2], ngrid)
+    end
+end
+
+function find_closest_indices(a, b)
+    closest_indices = []
+    for i in 1:length(b)
+        min_diff = Inf
+        closest_idx = 0
+        for j in 1:length(a)
+            diff = abs(a[j] - b[i])
+            if diff < min_diff
+                min_diff = diff
+                closest_idx = j
+            end
+        end
+        push!(closest_indices, closest_idx)
+    end
+    return closest_indices
+end
+
+function Freq2Index(isFermi, ωnList)
+    if isFermi
+        # ωn=(2n+1)π
+        return [Int(round((ωn / π - 1) / 2)) for ωn in ωnList]
+    else
+        # ωn=2nπ
+        return [Int(round(ωn / π / 2)) for ωn in ωnList]
+    end
+end
+function nGrid_test(isFermi, Λ::Float, degree, ratio::Float) where {Float}
+    # generate n grid from a logarithmic fine grid
+    np = Int(round(log(10*10 *10*Λ) / log(ratio)))
+    xc = [(i - 1) / degree for i = 1:degree]
+    panel = [ratio^(i - 1) - 1 for i = 1:(np+1)]
+    nGrid = zeros(Int, np * degree)
+    for i = 1:np
+        a, b = panel[i], panel[i+1]
+        nGrid[(i-1)*degree+1:i*degree] = Freq2Index(isFermi, a .+ (b - a) .* xc)
+    end
+    unique!(nGrid)
+    #return nGrid
+    if isFermi
+        return vcat(-nGrid[end:-1:1] .-1, nGrid)
+    else
+        return  vcat(-nGrid[end:-1:2], nGrid)
+    end
+end
+
+function find_indices_optimized(a, b)
+    indices = []
+    for i in 1:length(b)
+        for j in 1:length(a)
+            if abs(b[i] - a[j])<1e-16 
+                push!(indices, j)
+                break
+            end
+        end
+    end
+    return indices
+end
+
+function ωQR(kernel, rtol, print::Bool = true)
+    # print && println(τ.grid[end], ", ", τ.panel[end])
+    # print && println(ω.grid[end], ", ", ω.panel[end])
+    ################# find the rank ##############################
+    """
+    For a given index k, decompose R=[R11, R12; 0, R22] where R11 is a k×k matrix. 
+    If R11 is well-conditioned, then 
+    σᵢ(R11) ≤ σᵢ(kernel) for 1≤i≤k, and
+    σⱼ(kernel) ≤ σⱼ₋ₖ(R22) for k+1≤j≤N
+    See Page 487 of the book: Golub, G.H. and Van Loan, C.F., 2013. Matrix computations. 4th. Johns Hopkins.
+    Thus, the effective rank is defined as the minimal k that satisfy rtol≤ σ₁(R22)/σ₁(kernel)
+    """
+    Nτ, Nω = size(kernel)
+
+    u, σ, v = svd(kernel)
+    rank, err = 1, 0.0
+    for (si, s) in enumerate(σ)
+        # println(si, " => ", s / σ[1])
+        if s / σ[1] < rtol
+            rank = si - 1
+            err = s[1] / σ[1]
+            break
+        end
+    end
+    print && println("Kernel ϵ-rank = ", rank, ", rtol ≈ ", err)
+
+    Q, R, p = qr(kernel, Val(true)) # julia qr has a strange design, Val(true) will do a pivot QR
+    # size(R) == (Nτ, Nω) if Nω>Nτ
+    # or size(R) == (Nω, Nω) if Nω<Nτ
+
+    for idx = rank:min(Nτ, Nω)
+        if Nω > Nτ
+            R22 = R[idx:Nτ, idx:Nω]
+        else
+            R22 = R[idx:Nω, idx:Nω]
+        end
+        u2, s2, v2 = svd(R22)
+        # println(idx, " => ", s2[1] / σ[1])
+        if s2[1] / σ[1] < rtol
+            rank = idx
+            err = s2[1] / σ[1]
+            break
+        end
+    end
+    print && println("DLR rank      = ", rank, ", rtol ≈ ", err)
+
+    # @assert err ≈ 4.58983288255442e-13
+
+    return p[1:rank]
+end