EC.jl implemented for all 3-body systems

This commit is contained in:
Nuwan Yapa 2025-01-10 15:35:16 -05:00
parent 9084820ddc
commit 29bbceac03
5 changed files with 35 additions and 166 deletions

15
EC.jl
View File

@ -23,18 +23,18 @@ end
If a list is provided for ref_eval, they are used as reference values for picking the closest eigenvalues at each sampling point.
If a single number is provided for ref_eval, it is used as a reference for the first point, and the previous eigenvalue is used as the reference for each successive point.
Set orthonormalize_threshold=0 to skip Gram-Schmidt orthonormalization and use GEVP. Otherwise this value is used as the threshold for dropping redundant vectors."
function train!(EC::affine_EC, c_vals; ref_eval=nothing, orthonormalize_threshold=1e-5, CAEC=false, verbose=true, tol=1e-5)
function train!(EC::affine_EC, c_vals; ref_eval=-10.0, orthonormalize_threshold=1e-5, CAEC=false, verbose=true, tol=1e-5)
training_vecs = Vector{ComplexF64}[]
current_E = -10.0 # randomly chosen reference value
for c in c_vals
verbose && println("Training for c = $c")
global current_E
if ref_eval isa Number
current_E = ref_eval
ref_eval = nothing
elseif !isnothing(ref_eval)
current_E = pop!(ref_eval)
current_E = popfirst!(ref_eval)
end
H = EC.H0 + c .* EC.H1
@ -64,22 +64,21 @@ function train!(EC::affine_EC, c_vals; ref_eval=nothing, orthonormalize_threshol
EC.trained = true
end
"Extrapolate using a train EC model for a given range of c values
"Extrapolate using a trained EC model for a given range of c values
If a list is provided for ref_eval, they are used as reference values for picking the closest eigenvalues at each point.
If a single number is provided for ref_eval, it is used as a reference for the first point, and the previous eigenvalue is used as the reference for each successive point."
function extrapolate!(EC::affine_EC, c_vals; ref_eval=nothing, verbose=true, tol=1e-5)
function extrapolate!(EC::affine_EC, c_vals; ref_eval=EC.training_E[end], verbose=true, tol=1e-5)
@assert EC.trained "EC model must be trained using train() before extrapolation"
current_E = -10.0 # randomly chosen reference value
for c in c_vals
verbose && println("Extact solution for c = $c")
global current_E
if ref_eval isa Number
current_E = ref_eval
ref_eval = nothing
elseif !isnothing(ref_eval)
current_E = pop!(ref_eval)
current_E = popfirst!(ref_eval)
end
H = EC.H0 + c .* EC.H1

View File

@ -1,20 +1,20 @@
using Plots
include("../EC.jl")
training_c = [1.1, 0.9, 0.7, 0.5]
extrapolating_c = 0.0 : 0.2 : 1.2
training_ref = reverse([1.4750633616275919 - 0.0003021770706749637im
training_ref = [1.4750633616275919 - 0.0003021770706749637im
1.9567078295375822 - 0.0007646829108872369im
2.4351117758403076 - 0.001281037843108658im
2.9096543462392357 - 0.002962488527470604im])
2.9096543462392357 - 0.002962488527470604im]
exact_ref = reverse([4.076662025307587-0.012709842443350328im,
extrapolating_ref = [4.076662025307587-0.012709842443350328im,
3.613318119833891-0.007335804709990623im,
3.1453431847006783-0.004030580410326795im,
2.672967129943755-0.00211498327461944im,
2.196542557810288-0.0010719835443437104im,
1.7164583929199813-0.0005455212208182736im,
1.233088227541505-0.0003070320106485624im])
1.233088227541505-0.0003070320106485624im]
include("../p_space_3body_resonance.jl")
H0 = H
@ -32,54 +32,11 @@ end
@time "Vp2" Vp2 = W_left * Vp2_HO * transpose(W_right)
@time "Vp" Vp = Vpb + Vp2
# free memory
basis = Hb_blocks = Hb = basis_ho = V2_HO = W_right = W_left = V2 = nothing
GC.gc()
exact = ComplexF64[]
training = ComplexF64[]
extrapolated = ComplexF64[]
training_vecs = Vector{ComplexF64}[]
for c in training_c
println("Training for c = $c")
global current_E = pop!(training_ref)
local H = H0 + c .* Vp
local evals, evecs = eigs(H, sigma=current_E, maxiter=5000, tol=1e-5, ritzvec=true, check=1)
global current_E = nearest(evals, current_E)
push!(training, current_E)
push!(training_vecs, evecs[:, nearestIndex(evals, current_E)])
end
weights = repeat(kron(ws, ws), jmax + 1)
weights_mat = spdiagm(weights)
println("Original EC dimensionality = $(length(training_vecs))")
@time "Gram-Schmidt" training_vecs = gram_schmidt!(training_vecs, weights; verbose=true) # orthonormalization
EC = affine_EC(H0, Vp, weights)
train!(EC, training_c; ref_eval=training_ref, CAEC=false)
extrapolate!(EC, extrapolating_c; ref_eval=extrapolating_ref)
EC_basis = hcat(training_vecs...)
H0_EC = transpose(EC_basis) * weights_mat * H0 * EC_basis
Vp_EC = transpose(EC_basis) * weights_mat * Vp * EC_basis
for c in extrapolating_c
println("Extrapolating for c = $c")
global current_E = pop!(exact_ref)
local H = H0 + c .* Vp
local evals, _ = eigs(H, sigma=current_E, maxiter=5000, tol=1e-5, ritzvec=false, check=1)
global current_E = nearest(evals, current_E)
push!(exact, current_E)
# extrapolation
H_EC = H0_EC + c .* Vp_EC
evals = eigvals(H_EC)
push!(extrapolated, nearest(evals, current_E))
end
scatter(real.(training),imag.(training), label="training")
scatter!(real.(exact),imag.(exact), label="exact")
scatter!(real.(extrapolated),imag.(extrapolated), label="extrapolated")
savefig("temp/Berggren_R2R.pdf")
exportCSV(EC, "temp/Berggren_R2R.csv")
plot(EC, "temp/Berggren_R2R.pdf")

View File

@ -1,7 +1,7 @@
include("../EC.jl")
training_ref = -0.72763
exact_ref = 4.0766890719636635 - 0.01275892774109674im
extrapolating_ref = 4.0766890719636635 - 0.01275892774109674im
training_c = [2.0, 1.9, 1.8]
extrapolating_c = 0.0 : 0.2 : 1.2
@ -15,7 +15,7 @@ Vp_of_r(r) = -exp(-(r/3)^2)
EC = affine_EC(H0, Vp)
train!(EC, training_c; ref_eval=training_ref, CAEC=true)
extrapolate!(EC, extrapolating_c; ref_eval=exact_ref)
extrapolate!(EC, extrapolating_c; ref_eval=extrapolating_ref)
exportCSV(EC, "temp/HO_B2R.csv")
plot(EC, "temp/HO_B2R.pdf")

View File

@ -1,56 +1,16 @@
using DelimitedFiles, Plots
include("../ho_basis_3body_resonance.jl")
include("../EC.jl")
current_E = 5.9673 - 0.0006im
ref_E = 5.9673 - 0.0006im
training_c = 2.0 : -0.2 : 1.2
extrapolating_c = 1.05 .- [0.0 : 0.1 : 0.4; 0.45 : 0.05 : 0.60]
@time "H0" H0 = T1 + T2
include("../ho_basis_3body_resonance.jl")
@time "H0" H0 = T1 + T2 + T_cross
# free memory
basis = T1 = T2 = V1_cache = V_relative_cache = V1 = V_relative = U = V2 = nothing
GC.gc()
EC = affine_EC(H0, V)
train!(EC, training_c; ref_eval=ref_E, CAEC=false)
extrapolate!(EC, extrapolating_c)
exact = ComplexF64[]
training = ComplexF64[]
extrapolated = ComplexF64[]
training_vecs = Vector{ComplexF64}[]
for c in training_c
println("Training for c = $c")
local H = H0 + c .* V
local evals, evecs = eigs(H, nev=3, ncv=24, which=:LI, maxiter=5000, tol=1e-5, ritzvec=true, check=1)
global current_E = nearest(evals, current_E)
push!(training, current_E)
push!(training_vecs, evecs[:, nearestIndex(evals, current_E)])
end
println("Original EC dimensionality = $(length(training_vecs))")
@time "Gram-Schmidt" training_vecs = gram_schmidt!(training_vecs; verbose=true) # orthonormalization
EC_basis = hcat(training_vecs...)
H0_EC = transpose(EC_basis) * H0 * EC_basis
V_EC = transpose(EC_basis) * V * EC_basis
for c in extrapolating_c
println("Extrapolating for c = $c")
local H = H0 + c .* V
local evals, evecs = eigs(H, nev=3, ncv=24, which=:LI, maxiter=5000, tol=1e-5, ritzvec=true, check=1)
global current_E = nearest(evals, current_E)
push!(exact, current_E)
# extrapolation
H_EC = H0_EC + c .* V_EC
evals = eigvals(H_EC)
push!(extrapolated, nearest(evals, current_E))
end
exportCSV("temp/NCSM.csv", (training, exact, extrapolated), ("training", "exact", "extrapolated"))
scatter(real.(training),imag.(training), label="training")
scatter!(real.(exact),imag.(exact), label="exact")
scatter!(real.(extrapolated),imag.(extrapolated), label="extrapolated")
savefig("temp/NCSM.pdf")
exportCSV(EC, "temp/HO_R2R.csv")
plot(EC, "temp/HO_R2R.pdf")

View File

@ -1,6 +1,5 @@
using Arpack, SparseArrays, LRUCache
using DelimitedFiles, Plots
include("../ho_basis.jl")
include("../EC.jl")
Λ = 0
m = 1.0
@ -31,62 +30,16 @@ println("Basis size = ", basis.dim)
@time "Ha" Ha = T1 + T2 + Va
@time "Eigenvalues" target_evals, _ = eigs(Ha, nev=5, ncv=50, which=:SR, maxiter=5000, tol=1e-5, ritzvec=false, check=1)
display(target_evals)
# free memory
basis = T1 = T2 = V1_cache = V_relative_cache = V1 = V_relative = U = V2 = nothing
GC.gc()
training_c = [-0.5, -0.65, -0.8, -1, -1.2]
extrapolating_c = [0.8, 0.6, 0.4, 0.2, 0.1, 0.0, -0.1, -0.2, -0.3]
current_E = -0.5173809356244544
ref_E = -0.5173809356244544
exact = ComplexF64[]
training = ComplexF64[]
extrapolated = ComplexF64[]
training_vecs = Vector{ComplexF64}[]
EC = affine_EC(Ha, Vb)
train!(EC, training_c; ref_eval=ref_E, CAEC=true) # try CAEC=false !!!
extrapolate!(EC, extrapolating_c)
for c in training_c
print("Training for c = $c: ")
H = Ha + c .* Vb
evals, evecs = eigs(H, nev=3, ncv=24, which=:SR, maxiter=5000, tol=1e-5, ritzvec=true, check=1)
global current_E = nearest(evals, current_E)
println(current_E)
push!(training, current_E)
push!(training_vecs, evecs[:, nearestIndex(evals, current_E)])
end
training_vecs = vcat(training_vecs, conj(training_vecs)) # CA-EC
println("Original EC dimensionality = $(length(training_vecs))")
@time "Gram-Schmidt" training_vecs = gram_schmidt!(training_vecs; verbose=true) # orthonormalization
EC_basis = hcat(training_vecs...)
Ha_EC = transpose(EC_basis) * Ha * EC_basis
Vb_EC = transpose(EC_basis) * Vb * EC_basis
current_E = -0.3005521915662689 - 0.13612069020686351im
for c in extrapolating_c
print("Extrapolating for c = $c: ")
H = Ha + c .* Vb
evals, evecs = eigs(H, nev=3, ncv=24, which=:SR, maxiter=5000, tol=1e-5, ritzvec=true, check=1)
global current_E = nearest(evals, current_E)
println(current_E)
push!(exact, current_E)
# extrapolation
H_EC = Ha_EC + c .* Vb_EC
evals = eigvals(H_EC)
push!(extrapolated, nearest(evals, current_E))
end
# exportCSV("temp/dis_HO_B2R.csv", (training, exact, extrapolated), ("training", "exact", "extrapolated"))
scatter(real.(training),imag.(training), label="training")
scatter!(real.(exact),imag.(exact), label="exact")
scatter!(real.(extrapolated),imag.(extrapolated), label="extrapolated")
savefig("temp/dis_HO_B2R.pdf")
exportCSV(EC, "temp/dis_HO_B2R.csv")
plot(EC, "temp/dis_HO_B2R.pdf")