thestaralight/semiclassicalsfi.jl

Implementation of classical/semiclassical methods in strong-field ionization of atoms and molecules. [Note: This library is currently under construction, and would be available in the recent months! :D]

semiclassicalsfi.jl's Issues

[2D] Adding support for 2D simulation

• 2D simulation

[MO-ADK] Adding support for MO-ADK

• MO-ADK

[traj_simu/dt] Finding a better default `traj_dt` for non-adaptive method or give option to use adaptive method

The program runs too slow under the current default traj_dt=0.1, thus we need to find out a better default value which reaches a balance between speed and accuracy.
Anyone who did test on this is welcomed to post his/her suggestions here!

[feat/unit] Adding support of specifying units when passing physical quantities

Allowing specifying units seems to be a better choice than restricting to atomic units.
The Unitful package is capable of handling the task.
Maybe this can be done in v1.4.1 or 1.5~

Help Wanted: new version of this package slows down on real mechine

Dear Developer:
Thank you for this amazing programme!

I first set it up in a Virtual Machine without using CUDA about a week ago. and it runs just fine.I tested the 'minimal example', after changing some variable and disabling GPU, the program runs at a speed of 1.6 s/it using 4 threats. latter I'd like to install it on a real machine with CUDA enabled. yet after changing most key words to fit the new package, this program runs at 22 s/it using 4 threats with CUDA enabled, and 1 min/it with CUDA disabled. The speed of newer version of pkg is also tested in my VM and the slow down is also observed (90s/it under -t 4 ).

the latter PC could passed the test(SemiclassicalSFI) in 260min. Both are using pkg 1.4 version, though key words are different but the inputs are the same. I was wondering if this is anything to do with my environments, both are using julia 1.92 installed from snap, and python 3.10, in ubuntu 22.04.

I have enclosed both input file in the end to see if there is anything wrong. please inform me if more information is needed.

In the end, I'd like to thank you again for sharing this well documented package, as an experimentalist this allow us to better understand our experiments.
Best Regard
Kefei

---

enclosed input files

Inputs in real that runs slow______

using SemiclassicalSFI

filename = "SCSFI_HydLike_Ip_0.5662.h5"  # output file name


t = SemiclassicalSFI.Targets.HydrogenLikeAtom(Ip=0.5662, Z=1)


l = SemiclassicalSFI.Lasers.Cos2Laser(peak_int=3e14, wave_len=800, cyc_num=6, ellip=0.0)


SemiclassicalSFI.performSFI(

    init_cond_method = :ADK,
    laser = l,
    target = t,
    

    sample_t_interval = (-300,300),
    sample_t_num = 20000,

    final_p_max = (3,3,3),
    final_p_num = (800,800,1),

    traj_t_final = 400,
    save_3D_spec = false,

    ss_kd_max = 2.,
    ss_kd_num = 500,
    ss_kz_max = 2.,
    ss_kz_num = 150,

    traj_phase_method = :CTMC,

    traj_GPU = true,

    save_path = filename
    )

Inputs in VM that runs faster______

using SemiclassicalSFI

filename = "SCSFI_HydLike_Ip_0.5662.h5"  # output file name


t = SemiclassicalSFI.Targets.HydrogenLikeAtom(Ip=0.5662, Z=1)


l = SemiclassicalSFI.Lasers.Cos2Laser(peak_int=3e14, wave_len=800, cyc_num=6, ellip=0.0)


SemiclassicalSFI.performSFI(

    ionRateMethod = :ADK,
    laser = l,
    target = t,
    

    sample_tSpan = (-300,300),
    sample_tSampleNum = 20000,


    finalMomentum_pMax = (3,3,3),
    finalMomentum_pNum = (800,800,1),

    simu_tFinal = 400,
    save_3D_momentumSpec = false,

    ss_kdMax = 2.,
    ss_kdNum = 500,
    ss_kzMax = 2.,
    ss_kzNum = 150,

    simu_phaseMethod = :CTMC,

    simu_GPU = false,

    save_fileName = filename
    )

[PKG/test] Writing unit tests for the package

[WFAT] Adding support for UHF calculation & Becke integration grids for WFAT

• UHF
• Becke integration grids

Two Issues on Development of Molecular Systems

Recently we will focus on building the basic framework of molecular simulations, like we did in the atomic ones. Here I'll present two issues calling for discussions on the (semi)classical simulations of molecular systems.

1. Concrete types relating to molecules need to be set up, which should inherit Target supertype. But I wonder what concrete types should be created. Is a single type Molecule <: Target enough? What basic properties should be set?
2. WFAT rate method would be our recent focus, which requires calculations on the molecular orbital. However, due to lack of stable and well-maintained quantum chemistry packages in current Julia ecosystem, we cannot rely on a single package. My opinion is that we adopt pyscf (in Python) as the first QC package, and leave an interface for future support for other packages. How does it sound?

[PKG] Wrapping the program in a julia package

• Generating the package
• #10
• Writing readme
• #11
• Merging into the master

[Atomic/misc] Current Assignments on Development on Atomic System

While we have published v1.2, atomic part of the library still needs perfection, and some functions are still incomplete. Here I'll list some TODOs.

• ADK
• SFA
• SFA-AE's rate prefix. [Branch name: RatePrefix]
• Non-dipole simulation support. [Branch name: Non-dipole]
• Saving simulation abstract in the h5 file. [Branch name: SimulationAbstract]

Anyone who's interested may create a branch from dev, and name it as its feature name.

[WFAT] Adding support for libcint integration to accelerate XC potential calculation

Currently the XC potential calculation is the most time-consuming task in WFAT, and is carried out by calling pyscf. However, the PyCall does not support multi-threading, and data transmission between Julia and Python would lower the efficiency. Calling libcint directly would solve this bottleneck in efficiency.

• Directly calling the libcint as the XC potential integration scheme

[PKG/docs] Writing docs via `Documenter.jl`