Example 2: Load from SmoQyDQMC and run SmoQyDEAC

Usage:

$ julia --threads=auto SmoQyDQMC.jl

SmoQyDEAC uses multithreading for parallelizing runs. Multithreading it recommended. --threads=auto will run a thread for each core available, or 2x for hyperthreading cores

In this example we explicitly load output from SmoQyDQMC. After running a simulation with SmoQyDQMC you will need to use the correlation_bins_to_csv method for the desired correlation function prior to using the loading script.

This example utilizes the script SmoQyDQMCloader.jl to parse SmoQyDQMC csv files. This script may be moved to another repository at a future date.

# First we import our SmoQyDQMC csv parser and needed packages for the example
include("SmoQyDQMCloader.jl")
using FileIO
using SmoQyDEAC

Create directory for outputs

output_directory = "SmoQyDQMC_DEAC_Out/";
mkpath(output_directory);

Load data from fermion greens correlation functions This puts real in the format real["ORBITAL\_ID\_1","ORBITAL\_ID\_2","TAU","K\_1","K\_2","K\_3","BIN","PID"]`

# This example was from a 1D Holstein run with β = 20.0,
#   ORBITAL\_ID\_1 ∈ {1},
#  ORBITAL\_ID\_2 ∈ {1},
#  TAU          ∈ {0.0,Δτ,...,β-Δτ,β} for Nτ = 201,
#  Kx           ∈ {1,...,32},
#  Ky           ∈ {1},
#  Kz           ∈ {1},
#  Bin          ∈ {1,...,Nbin} for Nbin = 100,
#  PID          ∈ {1},
#
# Some dimensions are 1 deep. They are kept to ensure generality.
#
# See the [`SmoQyDQMCloader.jl`](https://github.com/SmoQySuite/SmoQyDEAC.jl/blob/main/scripts/SmoQyDQMCloader.jl) file for more information
input_directory = "SmoQyDQMC_sim-1/"

dims,real,image,sgnr,sgni,β = load_from_SmoQyDQMC(simulationfolder=input_directory,
                                                correlation="greens",
                                                space="momentum",
                                                type="time_displaced",bin=true);
# dims is a dictionary which tells you what each dimension corresponds to.
println(dims)

Dict("ORBITAL_ID_2" => 2, "K_3" => 6, "BIN" => 7, "TAU" => 3, "K_1" => 4, "PID" => 8, "ORBITAL_ID_1" => 1, "K_2" => 5)

Eliminate unnecessary dimensions

Gτ = real[1,1,:,:,1,1,:,1];
println(size(Gτ))

(201, 32, 100)

Set parameters for DEAC

τs = collect(LinRange(0.0,β,size(Gτ,1)))
Nkx = size(Gτ,2)
number_of_bins = 2;
runs_per_bin = 10 ;
checkpoint_file = joinpath(output_directory,"DEAC_checkpoint.jld2");
nω = 401;
ωmin = -10.;
ωmax = 10.;
ωs = collect(LinRange(ωmin,ωmax,nω));

Run DEAC over all k points in the x direction. For speed in this example I run 1:1 instead of 1:Nkx

for kx in 1:1 # 1:Nkx
    output_file = joinpath(output_directory, string(kx) * ".jld2");
    # put in [bins,τ] shape
    Gτ_temp = Matrix{Float64}(Gτ[:,kx,:]');
    deac_dict = DEAC_Binned(
        Gτ_temp,
        β,
        τs,
        ωs,
        "time_fermionic",
        number_of_bins,
        runs_per_bin,
        output_file,
        checkpoint_file;
        fitness = [1.5,1.0],
        find_fitness_floor = false,
        number_of_generations = 20000,
        verbose = true
    )
end

Bootstrapping to 1005 samples
SIMD GEMM faster than BLAS, using SmoQyDEAC's gemmSIMD!()
[1.5, 1.0]
Target fitnesses are: [1.5, 1.0]

  Bin   1 | Run    1 | Fitness 1.403758 | Generations 20000
  Bin   1 | Run    2 | Fitness 1.414428 | Generations 20000
  Bin   1 | Run    3 | Fitness 1.421095 | Generations 20000
  Bin   1 | Run    4 | Fitness 1.424724 | Generations 20000
  Bin   1 | Run    5 | Fitness 1.443135 | Generations 20000
  Bin   1 | Run    6 | Fitness 1.430003 | Generations 20000
  Bin   1 | Run    7 | Fitness 1.413530 | Generations 20000
  Bin   1 | Run    8 | Fitness 1.425282 | Generations 20000
  Bin   1 | Run    9 | Fitness 1.401075 | Generations 20000
  Bin   1 | Run   10 | Fitness 1.405111 | Generations 20000
Finished bin 1 of 2
  Bin   2 | Run    1 | Fitness 1.422864 | Generations 20000
  Bin   2 | Run    2 | Fitness 1.417702 | Generations 20000
  Bin   2 | Run    3 | Fitness 1.415956 | Generations 20000
  Bin   2 | Run    4 | Fitness 1.412540 | Generations 20000
  Bin   2 | Run    5 | Fitness 1.416825 | Generations 20000
  Bin   2 | Run    6 | Fitness 1.428601 | Generations 20000
  Bin   2 | Run    7 | Fitness 1.423893 | Generations 20000
  Bin   2 | Run    8 | Fitness 1.418302 | Generations 20000
  Bin   2 | Run    9 | Fitness 1.421228 | Generations 20000
  Bin   2 | Run   10 | Fitness 1.414050 | Generations 20000
Finished bin 2 of 2

Saving data to SmoQyDQMC_DEAC_Out/1.jld2 and deleting checkpoint file

Run Statistics
 Expected 0th moment:   1.00
 DEAC 0th moment:       1.000
 0th moment difference: 0.000%
 Mean generations/run:  26002.550
 Total Run time:        20.799s
 Run time per bin:      10.399s
 Run time per genome:   10.399s

Note, these did not converge to a fitness of 1.0 within 20,000 generations. The number of generations is limited for speed when running this example.

This page was generated using Literate.jl.