dyberforge (Workbench SDK)

Open-source simulation SDK for H-cat components: cat-state generation from squeezed vacuum, single-mode cat loss curves, and an illustrative resource model. These are component simulations; they make no claim about a full machine.

Install (development)

conda create -n dyber python=3.11 -y && conda activate dyber
pip install numpy scipy qutip matplotlib stim pymatching
pip install -e .            # from forge/sdk/

Use

import dyberforge as df
df.generate_cat(s_db=6.0, k=1, N=40)        # cat fidelity, parity, Wigner-ready state
df.loss_curves(2.0, [10**-0.32])            # single-mode loss curves (component level)
df.estimate(gamma=0.005, target_logical_err=1e-9)  # model values (illustrative), not a device spec

Run the example: python examples/quickstart.py.

Notes

• The physics modules are vendored into dyberforge/, so the package is fully self-contained: pip install dyberforge works standalone with no path injection.
• Public science only.
• estimate() returns model values. Model value: abstract circuit model with assumed loss heralds. Not a hardware operating point.
• generate_cat() and loss_curves() are component simulations of single cat modes. The k=1 fits at 3.5 to 8 dB land at or near the alpha = 1 floor: small cats.
• In the H-cat encoding (J. Lee et al., PRX Quantum 5, 030322 (2024)) both logical states carry even cats. The parity these functions report is a loss flag, not the logical value.
• Optional extras: pip install .[devices] (Meep), .[crosscheck] (Strawberry Fields), .[dev] (pytest). The extras install those tools only; this package contains no Meep or Strawberry Fields code.

Changelog

0.1.1

• Labels and documentation only: resource-model thresholds, distances and cat counts are labeled as model values; the H-cat encoding and parity are described as in the published paper; the paper citation is corrected; estimate() results carry a basis field. No numerical change. The quickstart example also prints the fitted cat amplitude.