1 of 56

 

Roberto Imbuzeiro Oliveira (IMPA)

with Gabriel Leite Baptista da Silva (Groningen)

and Daniel Valesin (Groningen)

arXiv: 2111.11757

Quantitative Methods Seminar @ Krannert

Nov. 30th, 2021

2 of 56

The contact process

A graph with healthy and infected vertices.

3 of 56

The contact process

Infected vertices become healthy at rate 1 (Poisson process).

4 of 56

The contact process

 

5 of 56

The contact process

No contagion because vertex is already infected.

6 of 56

The contact process

Vertex becomes healthy.

7 of 56

The contact process

Nothing happens because already healthy.

8 of 56

The contact process

Nothing happens because starting point is healthy.

9 of 56

The contact process

Contagion

10 of 56

Definition

  •  

11 of 56

Why study the contact process?

Simple model for contagion and epidemics

Susceptible – Infected – Susceptible (SIS) model with spatial component

More generally: a model for “stuff that spreads” but also tends to disappear

12 of 56

Basic facts

  •  

13 of 56

Epidemic threshold

  •  

14 of 56

Large finite boxes

  •  

15 of 56

Random regular graphs

  •  

16 of 56

Power-law graphs are epidemic-prone

  •  

17 of 56

What happens if the graph changes over time?

Does it make it easier or harder for the epidemic to survive?

18 of 56

Switching

  •  

Switching connects all graphs with given degree sequence

19 of 56

Switching dynamics

  •  

20 of 56

Main result (modulo **)

  •  

21 of 56

Intuition: a branching process?

Each individual

has children

according to a fixed

probability

distribution,

Independently.

Maybe contact is

like this, but with

deaths.

(contagion tree)

22 of 56

A branching process with deaths?

Not quite, because of:

  1. finite space: can’t just keep branching forever;
  2. collisions: trying to infect a vertex that is already infected.

However, switching graph makes collisions less common.

23 of 56

Interpretation

Deriving “practical” insights from toy models of epidemics seems in very bad taste these days.

But maybe one can say this:

a randomly changing graph seems to make stuff spread faster.

24 of 56

Remainder of talk

Background and local models

Previous results for finite and infinite static graphs. “Local models.”

The local model in the dynamical graph setting

Definition, analysis, strict inequality for critical parameter.

From local model to actual dynamics in a finite graph

Only a brief sketch.

25 of 56

Background and local models

26 of 56

Infinite graphs

This is where the literature starts.

Allow for true phase transitions between certain death x survival.

27 of 56

Infinite lattices

  •  

Strong survival: finite sets reinfected infinitely many times.

28 of 56

Infinite trees: intermediate phase

  •  

Weak survival: goes extinct over finite sets, but not over full tree

29 of 56

From infinite to finite graphs

  •  

30 of 56

Large boxes

 

31 of 56

Large finite boxes (restatement)

  •  

32 of 56

Random regular graphs

  •  

33 of 56

Random regular graphs (restatement)

  •  

34 of 56

Recall main result for today (modulo **)

  •  

35 of 56

Main proof steps in our case

Find a local limit and study it

This is the herds process described next.

Prove strict inequality for critical parameter

Edge rewiring is a “branching mechanism”.

Prove something similar for process with truncated trees

Makes the last step easier.

Compare “locally” with dynamics in the finite graph.

36 of 56

The local limit of the process over the dynamical graph

37 of 56

Heuristics

Another tree in the graph.

A large tree in the graph

Switch edges?

Swap subtrees!

38 of 56

39 of 56

The local limit

  •  

40 of 56

Herd splits

  •  

41 of 56

Herd process = contact + branching

A new herd evolves independently from all other herds.

In particular, it may also split into new herds.

Built-in branching mechanism.

Makes it easier for the epidemic to survive.

If contact creates “many” particles, herds survive forever.

42 of 56

Contact process is dominated

  •  

43 of 56

Key result about herds process

  •  

44 of 56

Proof sketch (I)

  •  

45 of 56

Proof sketch (II)

  •  

46 of 56

From the herds process to finite graphs

47 of 56

Proof steps

Find a local limit and study it

In our case this is the herds process, a new process described next.

Prove strict inequality for critical parameter

Edge rewiring serves as a kind of “branching mechanism”.

Prove something similar for process with truncated trees

Makes the last step easier.

Compare “locally” with dynamics in the finite graph.

48 of 56

Truncated herds process

  •  

49 of 56

Herd creation is more complicated

  •  

50 of 56

 

51 of 56

Key result about h-herds process

  •  

Proof idea: if herds survives forever, there’s a positive chance of creating many particles within subtrees of bounded diameter 2h. Branching guarantees h-herds can survive forever.

52 of 56

Sketch of end of proof

Find a local limit and study it

In our case this is the herds process, a new process described next.

Prove strict inequality for critical parameter

Edge rewiring serves as a kind of “branching mechanism”.

Prove something similar for process with truncated trees

Makes the last step easier.

Compare “locally” with dynamics in the finite graph.

Can show that there is enough contagion between “nice sets of bounded diameter” . Multitype branching process & stuff.

53 of 56

Conclusion

and some open problems

54 of 56

Our main result

  •  

Proof goes through novel herd process over infinite graphs.

General idea of finding a local limit for the process,

but nontrivial truncation of the process is required.

55 of 56

Open problems

  •  

56 of 56

Thank you!