1 of 63

Generalized Contextuality via �EquiRank NMF

Farid Shahandeh, Theo Yianni, Mina Doosti

[2406.19382] Characterizing Contextuality via Rank Separation with Applications to Cloning 

[2506.09133] Complexity of Contextuality.

ESSLLI 2025, Theory and applications of sheaf theory, Bochum, August 8, 2025

2 of 63

story-tellers

3 of 63

is “history” real?

4 of 63

there are “facts”

5 of 63

I like to think that “physics” is “telling a story about continuously occurring phenomena.” And I don’t think there are “rules of nature.” We create rules to tell consistent stories.

6 of 63

prepare-measure scenario

Spekkens (2005), Chiribella et al. (2010), Barrett, PRA (2007)

7 of 63

prepare-measure scenario

Preparation

Turn on the laser (with characteristics X, Y, Z, …) AND align it such and such AND pass the beam through a polarizer.

Turn on the laser (with characteristics X, Y, Z, …) AND align it such and such AND reflect the beam off a polarizing beamsplitter.

Toss a coin AND do (1) above for heads and (2) above for tails.

Spekkens (2005), Chiribella et al. (2010), Barrett, PRA (2007)

8 of 63

prepare-measure scenario

Preparation

Measurement

Turn on the laser (with characteristics X, Y, Z, …) AND align it such and such AND pass the beam through a polarizer.

Turn on the avalanche photodetector AND connect it to the oscilloscope (w. characteristics X, Y, Z) AND align it as such.

Turn on the laser (with characteristics X, Y, Z, …) AND align it such and such AND reflect the beam off a polarizing beamsplitter.

Turn on the photon-number-resolving detector AND connect it to the oscilloscope (w. characteristics X’, Y’, Z’) AND align it as such AND coarse-grain the output as such.

Toss a coin AND do (1) above for heads and (2) above for tails.

Turn the array of APDs AND connect them to oscilloscopes (w. characteristics X, Y, Z) AND align them as such AND postprocess the output as such.

Spekkens (2005), Chiribella et al. (2010), Barrett, PRA (2007)

9 of 63

Prepare-measure scenario

10 of 63

a sentence!

Operational structure

Chiribella et al. (2010)

11 of 63

a sentence!

 

Probabilistic structure

Chiribella et al. (2010)

12 of 63

matrix of conditional outcome probability evaluations

(COPE)

FS, Yianni, Doost (2406.19382), Harrigan et al. (2008)

13 of 63

Matrix of conditional outcome probability evaluations

(COPE)

FS, Yianni, Doost (2406.19382), Harrigan et al. (2008)

14 of 63

matrix of conditional outcome probability evaluations

(COPE)

preparation

outcome

FS, Yianni, Doost (2406.19382), Harrigan et al. (2008)

15 of 63

matrix of conditional outcome probability evaluations

(COPE)

preparation

outcome

FS, Yianni, Doost (2406.19382), Harrigan et al. (2008)

 

16 of 63

Models

17 of 63

Models:

OPT (Chiribella et al.)

Quotiented OPT (Chiribella et al.)

preGPT (FS, unpublished)

GPT (quotiented preGPT) (Hardy, Barrett, J&H, FS)

Quasiprobabilistic (Ferrie et al.)

Ontological (Spekkens)

Noncontextual ontological (Spekkens)

18 of 63

Models:

OPT (Chiribella et al.)

Quotiented OPT (Chiribella et al.)

preGPT (FS, unpublished)

GPT (quotiented preGPT) (Hardy, Barrett, J&H, FS)

Quasiprobabilistic (Ferrie et al.)

Ontological (Spekkens)

Noncontextual ontological (Spekkens)

projections,

set-valued embeddings,

general linear

transformations

19 of 63

Models:

OPT (Chiribella et al.)

Quotiented OPT (Chiribella et al.)

preGPT (FS, unpublished)

GPT (quotiented preGPT) (Hardy, Barrett, J&H, FS)

Quasiprobabilistic (Ferrie et al.)

Ontological (Spekkens)

Noncontextual ontological (Spekkens)

projections,

set-valued embeddings,

general linear

transformations

20 of 63

Models:

OPT (Chiribella et al.)

Quotiented OPT (Chiribella et al.)

preGPT (FS, unpublished)

GPT (quotiented preGPT) (Hardy, Barrett, J&H, FS)

Quasiprobabilistic (Ferrie et al.)

Ontological (Spekkens)

Noncontextual ontological (Spekkens)

21 of 63

Models: Decompositions of COPE matrix

22 of 63

Working example: The boxworld

23 of 63

Working example: The boxworld

preparation

24 of 63

Working example: The boxworld

preparation

25 of 63

Working example: The boxworld

preparation

outcome

26 of 63

SVD

GPT

27 of 63

SVD

GPT

28 of 63

SVD

GPT

29 of 63

SVD

GPT

 

 

30 of 63

GPT

31 of 63

GPT

32 of 63

GPT

 

every prep. and outcome has a unique representation!

33 of 63

34 of 63

NMF

Ontological model

FS, Yianni, Doost (2406.19382)

35 of 63

NMF

Ontological model

FS, Yianni, Doost (2406.19382)

36 of 63

NMF

Ontological model

FS, Yianni, Doost (2406.19382)

37 of 63

NMF

Ontological model

 

 

FS, Yianni, Doost (2406.19382)

38 of 63

NMF

Ontological model

 

 

FS, Yianni, Doost (2406.19382)

39 of 63

NMF

Ontological model

 

 

FS, Yianni, Doost (2406.19382)

 

40 of 63

Ontological model

41 of 63

Ontological model

 

42 of 63

Ontological model

 

43 of 63

Ontological model

 

44 of 63

Ontological model

 

Generalized contextuality means operationally indiscernible propositions are represented differently in the model.

FS, Yianni, Doost (2406.19382)

45 of 63

 

generalized contextuality

 

FS, Yianni, Doost (2406.19382)

46 of 63

Working example: The boxworld

for boxworld, we can satisfy either nonnegativity

or

equirank condition

boxworld does not admit a noncontextual ontological model

 

Yianni, FS (2506.09133)

47 of 63

 

 

 

 

 

 

 

 

 

Yianni, FS (2506.09133)

48 of 63

 

 

 

 

 

 

 

 

Finding equirank NMF is at least exponential in the rank of C. So, it’s very hard!

Yianni, FS (2506.09133)

49 of 63

Sheaf theory?

Agrios (2202.01379)

50 of 63

Bell scenario

A

B

x

y

 

 

 

 

51 of 63

Bell scenario

Williams, Doosti, FS (coming soon), Constantin (1510.02561)

52 of 63

 

 

Bell scenario

Williams, Doosti, FS (coming soon), Constantin (1510.02561)

53 of 63

incidence matrix

global section

Bell scenario

Williams, Doosti, FS (coming soon), Constantin (1510.02561)

54 of 63

 

 

Bell scenario

Williams, Doosti, FS (coming soon)

55 of 63

 

 

incidence matrix

global sections

more data/preparations

Bell scenario

Williams, Doosti, FS (coming soon)

56 of 63

 

 

inner (ontic) dimension is fixed

response functions are fixed

Bell scenario

Williams, Doosti, FS (coming soon)

57 of 63

incidence matrix

Bell scenario

Williams, Doosti, FS (coming soon), Constantin (1510.02561)

global section doesn’t exist

model is sheaf-contextual

58 of 63

 

 

Bell scenario

Williams, Doosti, FS (coming soon)

 

 

59 of 63

Bell scenario

Williams, Doosti, FS (coming soon)

rank-one response function

rank one epistemic state

60 of 63

Bell scenario

Williams, Doosti, FS (coming soon)

rank-one response function

rank one epistemic state

 

generalized noncontextual

61 of 63

sheaf-contextuality

generalized contextuality

 

Williams, Doosti, FS (coming soon)

62 of 63

take-home messages:

  • interpret probabilities as semantics?

  • all (linear) models can be obtained as a decomposition of a COPE matrix.

  • COPE formalism can clarify connection between generalized and sheaf-theoretic (non)contextuality.

  • rank separation may explain quantum advantage (in conjunction with other assumptions).

  • rank separation may shed new light on fundamental limitations of quantum computing.

63 of 63

Lecturer in Computer Science, RHUL, UK

Closing Date: Friday 29 August 2025