Migrating from Wannier90

If you already run wannier90.x, this package slots into the same place in your workflow. It reads the same input files your DFT interface produces — .win, .amn, .mmn, .eig — so nothing upstream changes. You can drive it two ways: as drop-in binaries writing the same output files, or from Julia with results as native data structures instead of a .wout to parse.

The pipeline mapping

A typical Wannier90 run:

1. SCF + NSCF (DFT)                    → wavefunctions
2. wannier90.x -pp seedname            → seedname.nnkp
3. pw2wannier90 (or your interface)    → seedname.amn/.mmn/.eig (+ .spn/.uHu/.dmn as needed)
4. wannier90.x seedname                → .wout, .chk, _hr.dat, band files
5. postw90.x seedname                  → AHC, DOS, kpath, BoltzWann, … .dat files

Steps 1–3 are unchanged. Steps 2, 4, and 5 map one-to-one:

wannier90.jl -pp seedname     # step 2 (byte-identical .nnkp)
wannier90.jl seedname         # step 4 (same outputs, same formats)
postw90.jl seedname           # step 5 (same .dat files)

The commands come either from a repository clone (julia --project=. bin/wannier90.jl …) or, for a pkg> add-installed package, from using WannierFunctions; install_cli(), which writes these launchers to ~/.julia/bin.

Or as a library, replacing steps 4–5 with data you can compute on:

using WannierFunctions

model = read_model("seedname")               # .win/.amn/.mmn/.eig
res   = run_wannier(model)                   # windows via keywords for entangled bands

H = hamiltonian_operator(model, res)         # the H(R) behind _hr.dat
E = bands(H, kpath)                          # replaces bands_plot

bm = BerryModel("seedname")                  # from the .chk you (or we) wrote
anomalous_hall(bm; fermi_energy=…, kmesh=…)  # replaces postw90's berry_task = ahc

Checkpoints interchange in both directions: wannier90.x restart=plot consumes our .chk, and BerryModel consumes theirs. bin/w90chk2chk.jl converts .chk ↔ .chk.fmt.

Where your .win keywords go

The short version: they work. The parser is strict (unknown keywords error with a did-you-mean; recognised-but-unsupported ones warn once), and the supported set covers the wannierise, disentanglement (energy windows, dis_spheres, PDWF projectability, symmetry- adapted), plotting/output, transport, and all postw90 module keywords. The precise policy, keyword semantics matched to the reference source, and the (short) list of behavioural differences are in Wannier90 compatibility.

What you get back (library route)

Instead of scraping a .wout:

• res.spread.Ω, .ΩI, .ΩOD, .ΩD — the Omega I/D/OD/Total lines, in Ų.
• res.spread.centres (3 × num_wann, Cartesian Å) and .spreads — the per-WF lines.
• res.U, res.omega_trace, res.niter, res.converged — the gauge and convergence trace.
• Disentangled runs: res.omega_I, res.dis.omega_I_trace.

These reproduce the reference .wout numbers to the test-suite tolerances or better — see Validation.

Behavioural notes to expect

• Convergence semantics differ by optimiser. The .win/CLI path uses :w90, where — matching Wannier90 — the convergence check is off unless conv_window > 1 and the loop runs the full num_iter. The Julia-native default :rcg has a real convergence criterion and converged means what it says. Same minima either way (asserted by tests).
• Units. Centres/spreads come back in Å / Ų even for a bohr cell, matching the .wout convention; energies are eV. The bohr constant is the reference default (CODATA2006).
• use_ws_distance defaults to true (as in postw90) — mind it when comparing against old runs that disabled it.
• Γ-only runs the reference's real-orthogonal algorithm and returns exactly real gauges.
• One step further than migration: with DFTK you can skip the file round-trip entirely — see Getting started, Workflow B.