GENERAL ARTICLE

Social Distancing in Infection Control∗

A Lattice Model Demonstration

Ananda Mukherjee

The article illustrates the efficacy of proper social distancing

methods during disease outbreaks. We inspect the outbreak
by counting the number of ways n individuals are infected,
given an initially infected patient introduced in a lattice(cn ).
Using self-avoiding walks on lattices where the infection spreads
according to the SIR (Susceptible, Infected, Removed) model,
we were able to show an exponential decay of the ratio of ini-
tial and final values of cn after the distancing methods were
Ananda Mukherjee is an
employed. undergraduate student of
mathematics at the RKM
Vivekananda Centenary
1. Introduction College. He is mostly
interested in multiplicative
number theory, differential
Modelling infection spread over biological networks using com-
geometry and mathematical
putations (both deterministic and stochastic) with simple random physics.
walks, self-avoiding random walks, etc., has been substantially
explored before. Mostly questions were raised along the lines of
estimating whether a substantial part of a given network could be
infected, assuming a random node to be initially infected. Bauer
and Lizier [1] present a more nuanced investigation regarding the
role of a single individual on the spreading dynamics rather than
the global properties of the network.
Herein, we shall set up a rather similar walk-counting method
to assess the efficacy of social distancing and quarantine on in-
fluencing the infection spread. Unlike [1], however, our enquiry Keywords
involves inspecting ramifications of the topological properties of Infectious diseases, nCoV, self-
avoiding Walks.
the network and how they are affected when measures such as
social distancing are imposed.

∗
Vol.25, No.5, DOI: https://doi.org/10.1007/s12045-020-0982-2

RESONANCE | May 2020 669

 GENERAL ARTICLE

2. Motivations and Setup

General epidemic spreading models on network structures iden-

tify a complex network with a graph Γ = (V, E) (here V and E are
the vertex and edge sets respectively). It is natural to consider the
spreading process in terms of infection walks in the correspond-
ing graph. A walk of length N is a map from S = {1, ..., N} into
V.
γ : S −→ V

n −→ γn

Such that γi , γi+1 ∈ E ∀i .A walk is said to be self-avoiding if γ

is injective.
An infinite graph Γ = (V, E) is a lattice if it is connected, tran-
sitive and locally finite. For the sake of simplicity, we shall only
look at walks on lattices, in particular the Zd lattice.
Notice that this would immediately restrict the number of suscep-
tible individuals a given node can infect, and we would have to
consider each individual identical in terms of spreading influence
in the network.
Motivated by the behaviour of the present COVID-19 outbreak,
we limit ourselves to the SIR model of epidemic spreading. When
When considering considering infection walks on lattices, a certain node can only
infection walks on infect its immediate neighbours. Given that an arbitrary node can
lattices, a certain node
infect precisely 2d neighbours (an assumption made to fix the
can only infect its
immediate neighbours. number of people an average individual is likely to infect), our
network resembles the Zd lattice.
Proceeding in terms of the SIR model, we consider as susceptible
those nodes which have not been infected by the walk. The in-
fected nodes infect their neighbours, continuing the walk, and are
either classified as deceased or recovered (with immunity) within
a specific duration required by the disease. These deceased or im-
mune nodes comprise the effectively removed individuals. This
final idiosyncracy in the model tacitly encodes the injectivity of

670 RESONANCE | May 2020

 GENERAL ARTICLE

the walk, which pinpoints self-avoiding walks as our primary ob-

ject of study.

3. Self-avoiding Walks on Lattices

The model is defined as in [3], assigning equal probability to

all paths of length n starting from the origin and without self-
intersections. This family of probability measures is not consis-
tent as n is varied, and thus does not define a stochastic process;
the model is combinatorial in nature.
The number of n-step self-avoiding walks from the origin is de-
noted by cn . According to our previous setup, cn represents-given
sufficient time- the number of possible ways n individuals could
be infected given a particular node is infected initially (without
loss of generality, the origin).
The simplest bounds on the behaviour of cn are obtained as fol-
lows. An upper bound on cn is given by the number of walks
which have no immediate reversals, or in other words which never
visit the same site at times i and i + 2. Avoiding immediate rever-
sals allows 2d choices for the initial step, and 2d − 1 choices for
the n − 1 remaining steps, for a total of 2d(2d − 1)n−1 . For a lower
bound we simply count the number of walks in which each step
is in one of the d positive coordinate directions. Such walks are
necessarily self-avoiding. Thus we have

dn ≤ cn ≤ 2d(2d − 1)n−1 .

For better estimates, we look at one further construction. By The concatenation map
S AWn we denote the collection of Self-avoiding walks of length provides an intuitive way
to construct a
n starting from the origin, with the clear consequence that cn =
self-avoiding walk of
|S AWn | length m + n from
smaller self-avoiding
The concatenation map provides an intuitive way to construct
walks of lengths m and n
a self-avoiding walk of length m + n from smaller self-avoiding respectively.
walks of lengths m and n respectively:

Concat : S AWm × S AWn −→ S AWm+n

(γ(1) , γ(2) ) −→ Concat(γ(1) , γ(2) )

RESONANCE | May 2020 671

 GENERAL ARTICLE

Given by

Concat(γ(1) , γ(2) )(k) = γ(1) (k) k = 1, ..., n

Concat(γ(1) , γ(2) )(k) = γ(1) (n) + γ(2) (k − n) − γ(2) (1)

for k = n + 1, ..., n + m.

Since the map so defined ensures the existence of n and m step

self-avoiding walks for every corresponding n + m step walk that
the former pair concatenates to form, the map is clearly onto. It is
not one-one since Concat(γ(1) , γ(2) ) produces a walk whose first
n steps and final m steps are self avoiding, but may nevertheless
intersect itself globally.
The construction of such a map is revealing about the sequence
cn . Since the concatenation map is not injective, but simply sur-
jective, we have

|S AWm+n | ≤ |S AWm × S AWn | = |S AWm ||S AWn |

i.e. cm+n ≤ cm cn ∀m, n ∈ N.

Therefore, logcn is a subadditive sequence: logcn+m ≤ logcn +

logcm . We make use of the folklore subadditivity lemma[2]:

Lemma. If ai ∈ R is a subadditive sequence, (an )n≥1 converges

to in fk≥1 akk

Proof : It suffices to show that

an ak
lim sup ≤ for every k.
n k
We fix k and let
Ak = max ar .
1≤r≤k

Given a positive integer n we let j denote the largest integer which

is strictly less than nk . Then n = jk + r for some integer r with
1 ≤ r ≤ k. Using subadditivity, we have
n
an ≤ jak + ar ≤ ak + Ak .
k

672 RESONANCE | May 2020

 GENERAL ARTICLE

Dividing through by n and taking the lim supn≥1 then gives the
lemma. 
The connective constant of the lattice is defined as

μ = exp( inf N −1 logcN ),

N≥1

and the lemma implies that log μ = limn→∞ n−1 log cn . Hence we
have the better lower bound:

μn ≤ cn ∀ n ∈ N.

4. Effects of Social Distancing

Practices such as social distancing and quarantine are bound to Practices such as social
reduce the number of susceptible nodes that can be infected by a distancing and
quarantine are bound to
given node. In other words, the degree of each vertex is reduced,
reduce the number of
altering the dimension of the lattice. In an effective quarantine, susceptible nodes that
the number of susceptible individuals that a given person is in can be infected by a
contact with should be diminished considerably. given node. In other
words, the degree of
If the exposure of an individual can be curtailed by even half of each vertex is reduced,
its original value, we have the following effect. altering the dimension of
the lattice. In an
If the dimension drops from d to d/2, the number of ways n nodes effective quarantine, the
are infected before and after (say cn and c
n ) can be compared from number of susceptible
individuals that a given
the bounds obtained previously.
person is in contact with
should be diminished
c
n ≤ d(d − 1)n−1 ≤ dn ≤ μn ≤ cn .
considerably.
Now clearly dμ <1 This shows that

c
n dn  d n
≤ ≤ ,
cn cn μ
which decays exponentially with n.
Evidently, even a social distancing regime which prohibits the
number of interactions of an individual to half its usual value
shows an exponential decay of the ratio of the number of ways
n people can be infected in a network. Infection models gener-
ally examine the likelihood of a disease outbreak becoming an

RESONANCE | May 2020 673

 GENERAL ARTICLE

epidemic. Consequently, the feasibility of longer infection walks

primarily dictates whether an outbreak can infect a sizeable pro-
portion of the network, or in other words, become an epidemic. In
case of viral outbreaks such as the novel coronavirus which show
features of the SIR model of infection spreading and might like-
wise yield to our setup of self-avoiding walks, restrictions such
as social distancing provide good results if executed correctly.
This present study can be expanded considerably by taking into
account an average of how many other people a given individ-
ual is likely to infect. That average, which is certainly disease-
specific, would then indicate the dimension of the lattice to be
considered.
Further enquiries could be carried out along the lines of the role
of the connective constant of the lattice in concern. Particularly,
we have not explored whether a regime of social distancing cus-
toms could be so designed as to allow every infected node to be
connected to other susceptible nodes as described by a specific
lattice other than Zd , and whether the connective constant of the
new lattice affects the ratio of infection walks in favour of the
corresponding observations in Zd .

Suggested Reading
[1] Frank Bauer and Joseph T Lizier, Identifying Influential Spreaders and Effi-
ciently Estimating Infection Numbers in Epidemic Models: A Walk Counting
Approach, EPL (Europhysics Letters), 2012.
[2] N Madras, G Slade, The Self-Avoiding Walk, Birkhäuser Basel. Edition 1, 2013.
[3] Roland Bauerschmidt and Hugo Duminil-Copin and Jesse Goodman and Gor-
don Slade, Lectures on Self-Avoiding Walks, 2012.
[4] B Hayes, American Scientist, Vol.86, 1998.
[5] D J Watts and S H Strogatz, Collective Dynamics of ‘Small-world’ Networks,
Nature, Vol.393, 1998.
[6] R van der Hofstad and W Konig, A Survey of One-dimensional Random Poly-
mers, J. Stat. Phys, Vol.103, pp.915–944, 2001.
[7] R van der Hofstad and G Slade, A Generalised Inductive Approach to the Lace
Expansion, Probab. Theory Related Fields, Vol.122, pp.389–430, 2002.
[8] E J Janse van Rensburg, Monte Carlo Methods for the Self-avoiding Walk, J.
Phys. A: Math. Theor., Vol.42, 2009.
[9] I Jensen, A Parallel Algorithm for the Enumeration of Self-Avoiding Polygons
on the Square Lattice, J. Phys. A: Math. Gen., Vol.36, pp.5731–5745, 2003.

674 RESONANCE | May 2020

 GENERAL ARTICLE

[10] Enumeration of Self-avoiding Walks on the Square Lattice, J. Phys. A: Math.

Gen, Vol.37, pp.5503–5524, 2004.
[11] Improved Lower Bounds on the Connective Constants for Two-dimensional Address for Correspondence
Self-avoiding Walks, J. Phys. A: Math. Gen., Vol.37, pp.11521–11529, 2004. Ananda Mukherjee
[12] I Jensen and A J Guttmann, Self-avoiding Polygons on the Square Lattice, J. affiliation
Phys. A: Math. Gen., Vol.32, pp.4867–4876, 1999. Ramakrishna Mission
[13] H Kesten, On the Number of Self-avoiding Walks, J. Math. Phys., Vol.4, Vivekananda Centenary
pp.960–969, 1963. College
Rahara, Howrah
India, 711202
Email:
[email protected]

RESONANCE | May 2020 675