---
identifier: iit:result/albantakis-2023-fig6b-modular
title: The modular network of the IIT 4.0 paper condenses into three two-unit complexes
vetted_by: "Matteo Grasso"
formalism: iit-4.0-2026
summary: >-
  The six-unit network of Albantakis et al. (2023), Fig 6B, built from three
  weakly connected modules, in state Abcdef condenses along its fault lines
  into the complexes Ab (φ_s = 0.27, Φ = 2.75), cd, and ef (φ_s = 0.12,
  Φ = 3.18 each). Complexes, distinctions, and Φ match the figure under both
  formulations. Under 2026, φ_s is about half the published value for Ab
  and about a quarter for cd and ef.
substrate: >-
  Six binary units A–F with states in {−1, 1}, each a logistic of its
  weighted inputs (Eq 60) with k = 4, in three modules ab, cd, ef. Within a
  module the two units connect to each other with weight 0.6, and every unit
  has a self-connection of 0.2. Between modules there are weak connections
  of 0.1: A→D, A→F, C→B, C→F, E→B, E→D. PyPhi ships the substrate, with the
  weights supplied by the authors, as
  `pyphi.examples.iit4_2023_fig6b_substrate`.
state: "(1, −1, −1, −1, −1, −1) — Abcdef, unit A ON"
phi: 0.26775861814035634
mip: >-
  For the first complex Ab, the directed bipartition cutting B→A. For cd and
  ef, the directed bipartitions cutting C→D and E→F. The same under both
  formulations.
pyphi_version: 2.0.0rc4
script: albantakis-2023-fig6b-modular.py
see_also:
  - iit:measure/phi-s
  - iit:result/albantakis-2023-fig6a-bottleneck
sources:
  - ref: iit:ref/albantakis-2023b
    at: >-
      Fig 6B and its caption; Results and discussion, "Modular systems, fault
      lines, and irreducibility"
---

Albantakis et al. (2023) use this network to show how fault lines limit the
size of complexes. Partitions along the weak connections between modules
yield much lower φ_s than partitions that cut through a module, so the
six-unit system condenses into its three modules.

| complex | φ_s, Fig 6B | 2023 | 2026 | distinctions | Φ, Fig 6B | Φ, computed |
|---|---|---|---|---|---|---|
| Ab | 0.53 | 0.5360 | 0.2678 | 2 | 2.74 | 2.7489 |
| cd | 0.45 | 0.4569 | 0.1153 | 3 | 3.18 | 3.1824 |
| ef | 0.45 | 0.4569 | 0.1153 | 3 | 3.18 | 3.1824 |

The figure truncates φ_s to two decimals (0.536 appears as 0.53). The
complexes and their Φ-structures are the same under both formulations. The
intrinsic information requirement of Mayner et al. (2026) lowers φ_s of every
module, more for cd and ef than for Ab, without changing which candidates
are complexes.

The distinctions of Ab are A → B | B (φ_d = 0.7326) and b → a | a (0.6418),
as in the figure. The distinctions of cd are c → d | d (0.7550),
d → c | c (0.6369), and cd → d | cd (0.1594), with Σφ_d = 1.5513, five
relations, and Σφ_r = 1.6311; ef has the same values for e, f, and ef. The
figure lists φ_d = 0.7551 for the second distinction of cd and of ef, but
its own Σφ_d = 1.5513 equals the computed sum with 0.6369
(0.7550 + 0.6369 + 0.1594); with 0.7551 the sum would be 1.6696. The
figure's row value is therefore inconsistent with its own total, and the
computed value is consistent with it.

The script uses the default configuration of each formulation. To
reproduce, run the script with uv, which installs PyPhi 2.0.0rc4 for the run:

```
uv run --with "pyphi==2.0.0rc4" python canon/albantakis-2023-fig6b-modular.py
```
