Running without pyALF - ALF — Algorithms for Lattice Fermions

Running ALF directly using the compiled Fortran executable and parameter files. This gives full control over every aspect of the simulation.

Directory Setup¶

A simulation directory needs three files:

my_sim/
├── parameters     # Fortran namelists defining the model and QMC parameters
├── seeds          # One random seed per MPI rank (one integer per line)
└── ALF.out        # The compiled executable (or a symlink to it)

A ready-made template is provided in Scripts_and_Parameters_files/Start/.

The Parameters File¶

The parameters file uses Fortran namelist syntax. A minimal example for the Hubbard model:

&VAR_ham_name
ham_name = "Hubbard"
/
&VAR_lattice
L1 = 6
L2 = 6
Lattice_type = "Square"
Model = ""
/
&VAR_Model_Generic
Dtau   = 0.1d0
Beta   = 10.d0
N_SUN  = 1
N_FL   = 2
Symm   = .true.
/
&VAR_QMC
NSweep     = 50
NBin       = 20
Nwrap      = 10
Ltau       = 1
LOBS_ST    = 1
LOBS_EN    = 1
CPU_MAX    = 24.d0
Sequential = .true.
/
&VAR_Hubbard
Ham_T  = 1.d0
Ham_U  = 4.d0
Mz     = .false.
/
&VAR_errors
n_skip = 0
N_rebin = 1
/

Key Namelists¶

Namelist	Purpose
`&VAR_ham_name`	Selects the Hamiltonian (e.g., `"Hubbard"`, `"Kondo"`, `"tV"`)
`&VAR_lattice`	Lattice type and dimensions (see [[Predefined Lattices]])
`&VAR_Model_Generic`	Universal parameters: `Dtau`, `Beta`, `N_SUN`, `N_FL`, twist angles, projector settings
`&VAR_QMC`	QMC control: sweeps, bins, stabilization, update schemes (see [[Tuning and Best Practices]])
`&VAR_<Model>`	Model-specific parameters (e.g., `&VAR_Hubbard` for hopping and interaction)
`&VAR_errors`	Analysis parameters: `n_skip`, `N_rebin`, `N_Cov`

The Seeds File¶

One integer per line, one per MPI rank. The template in Scripts_and_Parameters_files/Start/seeds provides over 2000 seeds — enough for typical production runs. If you run with more MPI ranks than seeds, ALF will generate additional seeds from the provided ones.

Running¶

# Serial (single Markov chain)
./ALF.out

# MPI (one Markov chain per rank)
mpirun -np 20 ./ALF.out

ALF creates a RUNNING file at startup and removes it on clean exit. If RUNNING already exists, ALF refuses to start — this prevents accidental duplicate runs. Delete it manually if a previous run crashed.

Output Files¶

With HDF5¶

File	Content
`data.h5`	All observables (scalar, equal-time, time-displaced) in a single HDF5 file
`confout_N.h5`	Saved configuration for rank `N` (for restarts)
`info`	Run metadata: timings, acceptance rates, stabilization precision

Without HDF5¶

File	Content
`*_scal`	Scalar observables (one value per bin)
`*_eq`	Equal-time correlation functions
`*_tau`	Time-displaced correlation functions
`confout_N`	Saved configuration for rank `N` (for restarts)
`info`	Run metadata: timings, acceptance rates, stabilization precision

Restarts¶

ALF supports warm restarts: new bins are appended to existing output files, and the Markov chain continues from the saved configuration.

# Rename output configs to input configs
bash out_to_in.sh

# Resubmit — ALF detects confin_* and continues
./ALF.out

For tempering runs, use out_to_in_temper.sh which loops over all Temp_*/ subdirectories.

Batch Parameter Scans¶

Scripts_and_Parameters_files/Start_chain.c (a shell script despite the .c extension) automates parameter scans. It reads a Sims file where each row defines a parameter set:

Y  6  6  4.0  1.0  0.0  0.0  0.0  10.0  0.1  10  50  20  24.0
Y  8  8  4.0  1.0  0.0  0.0  0.0  10.0  0.1  10  50  20  24.0
stop

The first column (Y/N) controls whether to run that row. Remaining columns are substituted into placeholders in parameters and the job script via sed. Each row gets its own directory with a copy of the Start/ template.

Analysis¶

After a run completes, post-process with the analysis tools:

# HDF5 output
$ALF_DIR/Analysis/ana_hdf5.out

# Plain-text output
$ALF_DIR/Analysis/ana.out *

See [[Analysis Tools]] for details.