Code for the paper "Magic state cultivation: growing T states as cheap as CNOT gates"

Usage: recollecting statistics and regenerating plots

These instructions assume you are on a linux system. They were tested on a Debian distribution (specifically gLinux).

Step 0: install system and python dependencies, then confirm unit tests pass.

sudo apt install parallel
pip install -r requirements.txt

pytest src

Step 1: generate circuits tested by the paper.

./step1_make_circuits

Step 2: collect statistics from the circuits generated in step 1.

# Note: this script will appear to do nothing, because it sees that the data is already
# collected and stored in the `assets/` directory. Delete `assets/stats.csv` or change
# `--save_resume_filepath` arguments to force recollection.
#
# Warning: recollecting these statistics will take weeks. Or months.
# You can reduce the time by editing `step2_collect` to use smaller values of
# `--max_shots` or `--max_errors`.
./step2_collect

Step 3: generate various plots from the data collected in step 2.

./step3_plot

Usage: generating a custom circuit

./tools/make_circuits \
    --circuit_type end2end-inplace-distillation \
    --noise_strength 0.001 \
    --gateset css \
    --out_dir out \
    --basis Y \
    --d1 3 5 \
    --d2 "d1*2+1" \
    --r1 "d1" \
    --r2 5

Add --debug to also generate an html file containing a circuit viewer.

Available circuit types are:

• end2end-inplace-distillation: Full construction, including the injection stage and the cultivation stage and the escape stage, ending in a large grafted matchable code.
• escape-to-big-matchable-code: Just the escape stage, without injection or cultivation.
• inject[teleport]+cultivate: Injection and cultivation without escape, with the initial T state injected by creating a degenerate color code then measuring out the extra qubit in the T basis.
• inject[bell]+cultivate: Injection and cultivation without escape, with the initial T state injected by growing the color code with Bell pairs from d=1 to d=3.
• inject[unitary]+cultivate: Injection and cultivation without escape, with the initial T state injected unitarily.
• idle-matchable-code: Idling circuit using the grafted matchable code that the escape stage ends in.
• surface-code-memory: Idling circuit using the surface code.
• surface-code-cnot: A lattice surgery surface code cnot, for scale.
• escape-to-big-color-code: Color code circuit growing from d1 to d2 by preparing Bell pairs.

  

Repository layout

• assets/: Assets created for the paper, like images and 3d models.
  • assets/gen/: Assets generated by running ./step3_plot.
• out/: Ignored by git but various tools (like ./step1_make_circuits) store their output under here.
• src/: Source code directory
  • src/cultiv/: Python utilities for generating and testing the circuits reported in the paper.
    • src/cultiv/_construction/: Code for generating tested circuits.
    • src/cultiv/_decoding/: Code for decoding shots from tested circuits.
  • src/gen/: Python utilities for creating circuit chunks and compiling them into full circuits.
  • src/latte/: Python utilities for testing magic state distillation factories built out of lattice surgery.
• tools/: Bash scripts for performing tasks like generating circuit files, from the command line.
• testdata/: Data used by unit tests.
  • testdata/factory_scripts/: Specifies various magic state distillation factories as a series of pauli product actions. Unit tests iterate through each file to verify correctness.
  • testdata/surgery_scripts/: Specifies various magic state distillation factories as text diagrams of layers of lattice surgery with feedback annotations. Unit tests iterate through each file to verify correctness.
    • testdata/surgery_scripts/ccz_4x3_7_tels.lat: The lattice surgery factory included in the paper's bonus figures.