---
identifier: iit:result/albantakis-2023-fig7b-inactive
title: The five-unit network of the IIT 4.0 paper in state ABcde specifies 23 distinctions with different purviews
vetted_by: "Matteo Grasso"
formalism: iit-4.0-2026
summary: >-
  The five-unit network of Albantakis et al. (2023), Fig 7B, in state ABcde
  (unit E inactive) is a single complex with 23 of 31 possible distinctions
  and 13498 relations, Φ = 19.32. One distinction, cde, has a tied cause
  purview. Resolved as the paper's S1 Text prescribes, by maximizing Φ, it
  gives these values; the figure's 13111 relations and Φ = 18.56 come from
  the other reading. φ_s is 0.25 under the current formulation and 1.31
  under the 2023 formulation (figure: 1.31).
substrate: >-
  The Fig 7 network (see iit:result/albantakis-2023-fig7a-active) in a
  different state: unit E is OFF instead of ON. PyPhi ships the substrate,
  with the weights supplied by the authors, as
  `pyphi.examples.iit4_2023_fig7_substrate`.
state: "(1, 1, −1, −1, −1) — ABcde, units A, B ON"
phi: 0.24902638133897767
mip: >-
  A directed partition severing 14 of the 25 connections; the script lists
  them. The same under both formulations.
pyphi_version: 2.0.0rc4
script: albantakis-2023-fig7b-inactive.py
see_also:
  - iit:measure/phi-s
  - iit:result/albantakis-2023-fig7a-active
  - iit:result/albantakis-2023-fig7c-inactivated
sources:
  - ref: iit:ref/albantakis-2023b
    at: >-
      Fig 7B and its caption; Results and discussion, "Consciousness and
      activity: Active, inactive, and inactivated units"; S1 Text,
      "Resolving ties in the IIT algorithm"
---

Albantakis et al. (2023) turn unit E from ON to OFF to show that an inactive
unit still contributes to the Φ-structure: the number of distinctions stays
the same, but their purviews and φ_d change, and so do the relations.

| quantity | Fig 7B | 2023 | 2026 |
|---|---|---|---|
| complex | ABcde | ABcde | ABcde |
| φ_s | 1.31 | 1.3133 | 0.2490 |
| distinctions | 23 of 31 | 23 | 23 |
| relations | 13111 | 13498 | 13498 |
| Φ | 18.55 | 19.3169 | 19.3169 |

The relation counts and Φ differ from the figure because of a tie,
described below. The distinctions match the figure's list in mechanism,
purviews, states, and φ_d to four decimals (for example Bc → c | aB with
φ_d = 0.0234 and Bcde → c | a with 0.0020), except that the cause purview
of cde is b where the figure lists c. Σφ_d = 4.5395 and Σφ_r = 14.7774.

The cause purview of the mechanism cde is tied: the readings b and c, both
in state −1, give the same φ_d = 0.0132 and are both congruent with the
system's specified cause state. The two readings give different
Φ-structures, because the cause purview determines which other
distinctions cde overlaps.

| resolution of the cde tie | cause purview | relations | Σφ_r | Φ |
|---|---|---|---|---|
| maximal Φ (S1 Text) | b | 13498 | 14.7774 | 19.3169 |
| figure | c | 13111 | 14.0170 | 18.5565 |

S1 Text of the paper ("Resolving ties in the IIT algorithm") states that
ties in the cause or effect of a distinction "may be resolved at the level
of the cause–effect structure, by selecting the z*c/e that maximizes the
system's structure integrated information Φ". The reading b does that, and
it is the one PyPhi 2.0 selects, so the canon reports it. The figure's
values come from the reading c; PyPhi attributes them to the enumeration
order of the 2023 code. The script substitutes c and prints that result as
well, which reproduces the figure's relation count and Φ (the figure
truncates Φ to 18.55).

The paper states that in panel B "φ_s changes only slightly" and that Φ is
lower than in panel A. Under the 2023 formulation φ_s rises from 1.11 to
1.31; under the current formulation it falls from 0.48 to 0.25. Φ is lower
than in panel A (22.27) under either resolution of the tie.

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-fig7b-inactive.py
```
