---
identifier: iit:result/albantakis-2023-fig2-ab-structure
title: The complex aB of the IIT 4.0 paper specifies three distinctions and seven relations
vetted_by: "Matteo Grasso"
formalism: iit-4.0-2026
summary: >-
  The complex aB in the three-unit substrate of Albantakis et al. (2023)
  specifies the three distinctions of Fig 2 (a: b | b, φ_d = 0.33;
  B: A | Ab, 0.32; aB: b | Ab, 0.07) and seven relations, including the
  relation {a, aB} of Fig 4 with φ_r = 0.036 and all nine faces. Φ = 1.56.
  The Φ-structure is the same under the 2023 and 2026 formulations; only
  φ_s(aB) differs.
substrate: >-
  The three-unit substrate of Fig 1A (see iit:result/albantakis-2023-fig1-triad):
  binary units A, B, C, each a logistic of its weighted inputs (Eq 60) with
  k = 4. The system is aB; unit C is a background condition, causally
  marginalized (Eqs 3–4). PyPhi ships the substrate as
  `pyphi.examples.iit4_2023_fig1a_substrate`.
state: "(−1, 1, 1) — aBC; the system aB is in state (−1, 1)"
phi: 0.04049680811800574
mip: >-
  The directed bipartition cutting A→B, under both formulations.
pyphi_version: 2.0.0rc4
script: albantakis-2023-fig2-ab-structure.py
see_also:
  - iit:result/albantakis-2023-fig1-triad
  - iit:measure/phi-s
sources:
  - ref: iit:ref/albantakis-2023b
    at: >-
      Fig 2 and its caption; Fig 4 and its caption; "Composition and causal
      distinctions"; "Composition and causal relations"
---

Albantakis et al. (2023) unfold the complex aB of their introductory
substrate into its distinctions (Fig 2) and illustrate a relation between two
of them (Fig 4). The script unfolds the full Φ-structure of aB under both
formulations.

| mechanism | cause \| effect, Fig 2 | computed | φ_d, Fig 2 | φ_d, computed |
|---|---|---|---|---|
| a | b \| b | b \| b | 0.33 | 0.3327 |
| B | A \| Ab | A \| Ab | 0.32 | 0.3236 |
| aB | b \| Ab | b \| Ab | 0.07 | 0.0714 |

Fig 4 shows the relation between distinctions a and aB. Their four purviews
all contain unit b in state −1, so all nine faces exist, as the caption
states. The paper gives φ_r({a, aB}) = 0.035; the computed value is 0.0357.
This equals φ_d(aB)/2, and the paper's value is the same quantity computed
from the rounded φ_d(aB) = 0.07.

| quantity | computed |
|---|---|
| distinctions | 3 |
| Σφ_d | 0.7278 |
| relations | 7 |
| Σφ_r | 0.8349 |
| Φ | 1.5627 |

The seven relations are the three self-relations of a, B, and aB, the three
2-relations {a, B}, {a, aB}, {B, aB}, and the 3-relation {a, B, aB}. The paper
reports neither the relation count nor Φ for aB.

Distinctions and relations do not depend on the system-φ measure, which is
the only difference between the 2023 and 2026 formulations. The system's
specified cause–effect state is Ab in both, so the same three distinctions
are congruent. φ_s(aB) is 0.1719 under 2023 and 0.0405 under 2026; the
iit:result/albantakis-2023-fig1-triad entry explains the difference.

The script uses the default configuration of each formulation, with PyPhi's
concrete relation backend so that each relation can be listed. 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-fig2-ab-structure.py
```
