Two and three dimensions
The recommended API uses the number of lattice extents to determine the dimension. Apart from the vector length and logical link-field size, the API is the same as in the four-dimensional tutorial.
Initialize JACC before Gaugefields so that these examples remain portable to GPU backends:
import JACC
JACC.@init_backend
using GaugefieldsTwo dimensions
U2 = gauge_configuration(
(32, 48);
colors=3,
start=:hot,
seed=1234,
)
@assert length(U2) == 2
@assert gauge_lattice_size(U2) == (32, 48)
@assert gauge_halo_width(U2) == 1
@assert size(U2[1]) == (3, 3, 32, 48)U2[1] and U2[2] contain the links in the first and second directions. The Polyakov loop is measured in the second, final direction. With MPI and no explicit process grid, the default decomposition is (1, nranks).
U(1) fields
Set colors=1 for a two-dimensional U(1) field. General Wilson-loop actions use the same interface:
U1 = gauge_configuration(
(32, 48);
colors=1,
start=:hot,
seed=1234,
)
loops = make_loops_fromname("plaquette"; Dim=2)
append!(loops, loops')
action = GaugeAction(U1)
push!(action, 1.0, loops)
action_value = evaluate_GaugeAction(action, U1)See Wilson loops and gauge actions for custom U(1) paths and general-action gradient flow.
Three dimensions
U3 = gauge_configuration(
(24, 24, 48);
colors=3,
start=:hot,
seed=1234,
)
@assert length(U3) == 3
@assert gauge_lattice_size(U3) == (24, 24, 48)
@assert gauge_halo_width(U3) == 1
@assert size(U3[1]) == (3, 3, 24, 24, 48)U3[1], U3[2], and U3[3] contain the links in the three directions. The Polyakov loop is measured in the third, final direction. With MPI and no explicit process grid, the default decomposition is (1, 1, nranks).
The LM backend supports the same reproducible hot starts, measurements, heatbath updates, gradient flow, stout smearing, I/O, GPU execution, and MPI domain decomposition in 2D and 3D. The compatibility LegacyBackend has additional historical restrictions; in particular, its 3D configuration must be requested with halo=0.
Special initial configurations
The high-level gauge_configuration constructor deliberately limits start to :cold and :hot. Gaugefields also retains the specialized Oneinstanton and Oneinstanton_SUN_embedded constructors. Their LM forms use isMPILattice=true, a positive halo, and PEs for the process grid:
instanton2 = Oneinstanton(
2,
1,
32,
48;
isMPILattice=true,
PEs=(1, 1),
verbose_level=0,
)
embedded4 = Oneinstanton_SUN_embedded(
3,
8,
8,
8,
16;
NDW=1,
isMPILattice=true,
PEs=(1, 1, 1, 1),
verbose_level=0,
)For MPI, the product of PEs must equal the number of ranks. These names are specialized compatibility constructors rather than new start choices; their full parameter history is kept in Legacy API.