Congestion in confined ant traffic

Nick Gravish*, Gregory Gold, and Andrew Zangwill

School of Physics, Georgia Institute of Technology, Atlanta, GA 30332, USA

Michael A.D. Goodisman

School of Biology, Georgia Institute of Technology, Atlanta, GA 30332, USA

Daniel I. Goldman
School of Physics and School of Biology, Georgia Institute of Technology, Atlanta, GA 30332, USA
(Dated: July 30, 2014)
Many social animals construct subterranean nests where they collectively live and move. In
these environments the physical constraints imposed by the confined, crowded nest tunnels may
present challenges for mobility. To understand how organisms mitigate traffic jams in subterranean
environments we study traffic in the fire ant Solenopsis invicta. We monitor foraging traffic of
fire ant workers in laboratory tunnels of varied diameter 2, 3, 4, 6 mm. We observe that head-on
interactions in bi-directional traffic occurred frequently between oncoming ants and often led to
sustained traffic jams. Traffic jams persisted for a time that increased linearly with the number of
participating ants. The slope of traffic jam duration versus group size increased as tunnel diameter
decreased and diverged at a minimum diameter, Dc =1.47 mm. These results give a measure of
traffic flow sensitivity as a function of tunnel diameter and predict the minimum tunnel diameter
within which two-way traffic can occur in fire ant nests. Further, we compare the traffic flow
sensitivity observed in experiment with the range of tunnel sizes constructed in nature and observe
that fire ants always construct tunnels outside the range of traffic jam sensitivity. To understand
how important behavioral rules are in controlling traffic flow within tunnels we construct a minimal
null model for bi-directional traffic in confined space. Our minimal model reproduces experimental
observations and indicates that ant-tunnel traffic dynamics are due in part to the physical constraints
of subterranean life.
INTRODUCTION

phological or behavioral traits, such as body-size [19] or

interaction dynamics.

Traffic within crowded spaces is observed across a

range of size scales from intracellular [14] to interstate traffic [57]. Motor proteins transport cargo along
crowded microtubules [13, 8], aggregations of cells collectively migrate through the extra-cellular matrix [9, 10],
and animals navigate through crowded aerial, aquatic,
and terrestrial environments [11].
Many social animals live in centralized, enclosed environments such as ant nests, termite mounds, and prairie
dog nests. Social animals rely on the rapid movement of
resources, information, and individuals within the nest
for survival [12, 13]. However, in confined nest environments, the constraints of sensory deprivation, physical crowding, and environmental perturbations all hinder mobility and challenge effective movement. Recent
experiments have explored how individual ants [14], cockroaches [15], and ferrets [16] traverse confined spaces.
However, little is known about the collective locomotion
of subterranean organisms in their natural environments.
Social insects such as ants [12] and termites [17] construct nests in which they collectively live and move (Fig.
1a). Many features of nest morphology (tunnel size,
shape, branching) are conserved across different nests
within species [18]. Thus, an open biological question is
whether these nest morphologies are adaptations for effective subterranean life, or simply related to other mor-

In red imported fire ant (Solenopsis invicta) colonies

the approximately 3.5 mm long workers create subterranean foraging tunnels of diameter 3 14 mm and
up to 50 m in length (Fig. 1b, c). Foraging tunnels
extend horizontally away from the central nest and provide a path along which workers move back and forth
from the nest to the external environment. Foraging typically incurs the highest mortality among ant workers [12]
and thus construction of foraging tunnels are an effective
strategy to minimize worker mortality [20]. However, foraging tunnels restrict foraging traffic to the confines of
the nest and thus tunnel dimensions may limit resource
transport capabilities. In excavations of S. invicta foraging tunnel networks, Tschinkel [21] discovered that the
diameter of tunnels decreased with an increase in the
distance from the nest center. This suggests that traffic
demands may play a role in fire ant nest construction and
design.
Laboratory and field experiments have shed light on
the mechanisms by which ants manage traffic on ant surface foraging trails (see [19] for a review). For example,
lane formation during bi-directional traffic of the army
ant Eciton burchelli [22], trail widening by leaf cutter
ants (Atta colombica) [19, 23], and priority rules for passing [24] are examples of self-organized process that minimize traffic jams along trails. In bi-directional foraging

a)

vironment and along foraging tunnels. For instance, such

interactions allow fire ants to identify invaders [12, 13],
exchange food or water, or be recruited to a new foraging
site [2830]. While tactile interactions along surface trails
may inhibit traffic jams by breaking up the flow [25],
in subterranean tunnels workers are confined in space
and the tactile interactions will inhibit flow. Thus traffic
and tactile interactions are at odds: workers that stop
to interact present physical obstacles to the flow of traffic which may cause temporary traffic jams. Resources
and information move with the flow of traffic within the
nest, and thus smooth traffic flow is likely important for
successful subterranean life.
In this study we test the ability of fire ant colonies to
maintain bi-directional foraging traffic within tunnels of
natural and reduced diameter. From a biological perspective, we seek to understand how ants collectively move
within subterranean tunnels and how diameter affects
this motion. From a physical perspective, we seek to
understand how simple rules of collective motion within
confinement relate to the overall traffic flow of ants in a
colony. To address these questions we perform biological experiments and computer simulations to explore the
dynamics of fire ant foraging traffic in confined tunnels.

1 mm

b) Foraging tunnel network

Mound

5m

c)

Vertical

Foraging

0.2

METHODS

0
0.3
P

Experiment

5
10
L (mm)

15

FIG. 1. a) Images of fire ants within excavated nest tunnels

in a quasi-2D environment (see [26] for details). b) Overhead view of a natural subterranean foraging tunnel network
constructed by an S. invicta colony. Data reproduced from
Markin [27]. c) Size distributions of ant tunnel diameter in
laboratory vertical tunnels and natural horizontal foraging
tunnels. Vertical tunnel data collected by Gravish et al. [14]
and horizontal tunnel data collected by Tschinkel [21]. Bottom shows bodylength distribution of S. invicta workers reproduced from [14].

traffic of Atta colombica, workers do not form lanes and

instead frequently pause to engage in head-on encounters. Counter-intuitively, the rate of traffic flow was maximal in Atta colombica traffic when head-on encounter
frequency was maximum [25]. To explain this it was suggested that ant interactions may have a positive impact
on traffic by breaking up the flow and inhibiting jams of
groups moving in the same direction.
Tactile interactions are an important mode of information acquisition and resource transfer in the nest en-

Fire ant colonies (Solenopsis invicta) were collected in

Georgia in 2011-2012. Colonies were separated from soil
using the water drip method and were housed in plastic
bins that contained an enclosed nest area made from petri
dishes and an open foraging arena. We provided insects
to the colonies as food and water ad libitum.
We monitored unperturbed traffic between a laboratory nest and an open foraging arena (Fig. 2a). The
foraging arena consisted of a 27 cm 17 cm plastic bin
with Fluon coated walls in which we placed a constant
supply of water and food. A lamp placed above the arena
illuminated and heated the foraging zone. A plastic tube
connected the foraging arena to a glass tunnel oriented
horizontally with varied diameter (D = 2, 3, 4, 6 mm)
and length of 11 cm (Fig. 2b). The glass tunnel was
connected to an enclosed plastic nest which was painted
black and contained a moist plaster-of-paris floor.
Five separate groups of 500-2000 worker fire ants of
body length 3.50.5 were removed from their host colony
and placed in the foraging arena. We excluded queens,
males, and brood from the group. The five worker groups
were drawn from three colonies and were monitored in
independent experiments over the course of three months.
Within several hours of being relocated to the foraging
arena the workers migrated to the nest and maintained a
continuous flow of bi-directional traffic from the nest to

a)

Camera
11 cm
Foraging
arena

Tunnel
Nest

P
P
tunnel as n(t) = C1 x (x, t) where x is the sum over
the entire length of the tunnel, and C is a normalization
constant so n(t) = 1 when one ant is in the tunnel.
We analyzed the spatial clustering of ants that occurred within the tunnel by finding spatially connected
regions where (x, t) is non-zero. For each cluster of size
n (binned in increments of 0.5) we computed the density
correlation function

b)
Qn ( ) =

c)

3 mm

x
FIG. 2. Experimental apparatus to monitor traffic in different
diameter tunnels. a) Experiment showing nest and foraging
arena separated by 11 cm long glass tube of diameter D
[2, 3, 4, 6] mm. b) Size relationship between laboratory tunnel
diameter and typical worker body size. c) Image of two ants
in a 2 mm diameter tunnel (top) and a 6 mm tunnel (bottom).
Plots above each image show at a fixed time.

the foraging arena and back. We monitored the foraging

traffic of worker groups for 24-72 hours within each of
the four tunnel sizes. The order of tunnel presentation to
the worker group was randomized across different group
trials.
We recorded video sequences of traffic, 40 seconds in
duration at a rate of 100 Hz, and resolution of 100 x 1328
pixels (120 pixels was equal to 1 cm). After the collection of each video we performed post-processing in Matlab which consisted of dividing each video frame by a stationary background image and thresholding the resultant
image to generate a binary image time-series, It (x, y),
containing only ants. Following image processing a new
video was captured. The time interval between successive videos was 2 minutes. Our experiment consisted
of over 10,000 videos each with 4,000 frames of traffic.
We analyzed spatio-temporal traffic dynamics as a onedimensional flow of density, (x, t) along the length of
the tunnel,
Px (Fig. 2c and Fig. 3). (x, t) is defined as
(x, t) = y It (x, y) where It (x, y) is the experimental
image at time t. We define the number of ants within a

h(x, t0 )(x, t0 + )i h(x, t0 )i

h(x, t0 )2 i h(x, t0 )i

(1)

(introduced in [31]). Qn ( ) is a function which varies

from 0 to 1 with time interval, , and measures how correlated an ants position is over time. Ants in traffic jams
move slowly and thus Qn ( ) remains large (high correlation) for longer time durations than when compared
to free-flow. The brackets h...i are the spatio-temporal
average of the function over the cluster size and time interval [0, 13s]. For the calculation of Qn ( ) we evaluated every fourth video frame to speed up computation.
The correlation function interval was chosen to be long
enough such that Q( ) = 0 for long time intervals.
To track ant motion within the tunnel we used two
techniques. To determine the speed-density relationship
we manually tracked the distance ants traveled during
1.2 second time intervals which corresponded to a freeflow travel distance of approximately 7 bodylenghts. In
addition we measured the local density of ants within
2 bodylengths of the tracked ant at the beginning of
motion. We also used an automated tracker which located all ant clusters between adjacent frames and correlated similar clusters among the frames by minimizing
the traveled distance and cluster size. From the distance
each cluster moved we determined the speed of the cluster.
To study the interaction behavior of ants we hand
tracked ants during antennae contact and measured duration. We define the interaction time, Tint , as the time
from first antennae contact between two ants to the time
when they have passed and there petioles are aligned.

Simulation

We seek to understand how and if ant traffic in tunnels

depends on the details of ant-ant interactions. To explore
this question we implemented a 2D cellular automata in
which we input the rules of interaction and studied the
resultant ant traffic (Fig. 4). In our model ants occupied
lattice sites and moved bi-directionally along the tunnel
length. Ants entered the tunnel from the left or right
at random and advanced along a direction of motion towards the opposite tunnel end. Ants advanced forward
by one lattice site during each iteration; however only a
single ant occupied a lattice site at a time. Two ants

a)

0 c)

b)

10

d)

10

Time (s)

Time (s)

30

30

5
10 40
Position (cm)

5
10 40
Position (cm)

FIG. 3. Space-time representation of ant-traffic in different diameter tunnels. a) Image sequence of ants moving bi-directionally
in a 6 mm tunnel. Images are separated by 14 ms and tunnel length is 10 cm. b) Plot of (x, t) with individual times separated
vertically. Image sequence is taken from the highlighted section in the middle. c) Image sequence of traffic flow from 2 mm
tunnel. d) Plot of (x, t) in a 2 mm tunnel with individual times separated vertically.

a)

bined probability of either ant jumping past each other,

2p p2 . The statistics of head-on encounters follow a
Poisson process and thus the interaction time between
two ants in simulation was determined from the median
value of the exponential distribution function. The median interaction time

of ants in simulation with time-step
dt is ln(2)/ 2p p2 dt. We fixed the length of the simulated tunnel to match that of the experiment with tunnel length l = 31L and grid-length of one bodylength.
The time step of the simulation (dt = 0.175 s) was
chosen such that the free-speed was 2 cm/s, matching
experiment. We varied the width of the tunnel from
Def f = 3 100 lattice spaces.

Deff

Time (s)

b) 0

12
0

x (cm)

FIG. 4. Simulation of bi-directional tunnel traffic. a)

Schematic of cellular automata simulation. Orange ants move
left and black ants move right. b) Space-time plot of (x, t).

adjacent to each other in the head-on direction were permitted to move only by jumping to an open lateral lattice
site with probability p.
By varying p we could vary Tint in simulation to explore how head-on encounters influence traffic. The probability for a head-on encounter to end is given by the com-

To compute flow-density curves in simulation we measured the distance traveled by each ant over 7 time-steps
(a 1.225 s time interval chosen to match experiment).
Similar to experiment, the local density was calculated
as the number of ants within a 2 lattice spacing distance ahead and behind the focal ant. We computed the
mean value of ant speed in density intervals of 1 ant/cm
and scaled the simulation speed to match the experiment
at a single value, the free flow condition (1 ant/cm). We
scaled the simulation for comparison because our simulation parameters were set to match the average free-flow
speed of ants in experiment but speed-density results are
based on maximums speeds achieved (the upper bound
curves in experiment).

Speed (cm/s)

4
2
0
0

6 mm
2 mm

10

10

10

10

Density (ants/cm)
FIG. 5. Fundamental diagram of bi-directional traffic flow in tunnels. Tunnel diameter increases from left to right. Solid lines
are estimates of the bounding curves for the speed-density relationship. Dashed line in left panel is for comparison between
2 mm and 6 mm tunnels. Black and white circles are results from simulation for simulated tunnel diameters of Def f = 2, 3, 4, 6
from left to right.

RESULTS
Experimental results
Ant interaction time

Ants moved bi-directionally through the foraging tunnel (Fig. 3). Often, two ants meeting head-on brushed
their antennae against one anothers head and body (Fig.
2b and SI Movies 1,2). We measured the interaction time,
Tint , between inbound and outbound ants in the four tunnel diameter treatments and found that Tint differed only
in the smallest tunnel (See SI Fig. 2). In D = 3 6 mm
tunnels Tint = 0.45 0.26 s, in the 2 mm tunnel interaction times were skewed to longer durations with a mean
value of Tint = 1.13 1.30.
From automated tracking of the velocity of ant clusters
within the tunnel, we computed speed distributions for
scenarios in which ants moved in traffic, and in which
they moved freely through the tunnel. We observed
that the free speed distribution was roughly gaussian distributed with hvf ree i = 1.93 0.63 cm/s. The speed
distribution in traffic was skewed to the right with the
majority of speeds near zero (See SI). We observed that
these instances of low speed corresponded to situations
where antennation occurred in the tunnel.

Bulk traffic flow

In all tunnels, ant interactions impacted the flow of

traffic nearby creating local traffic-jams (Fig. 3). To
determine if traffic flow was related to ant density we
measured the speed-density relationship within the four
tunnels from experiment. We observed that within all
tunnel sizes, the upper bound of speed vs. density decreased with increasing density (Fig. 5). Comparison
of the upper bound curves of speed vs. density between

large and small diameter tunnels illustrated that speed

decreased with increasing density more rapidly in larger
tunnels than in smaller (Fig. 5 left, solid line 2 mm, and
dashed line 6 mm).
A fundamental measure of traffic throughput is the
traffic flow, F , which is the product of speed and density
(Fig. 6a) [32]. F is the product of speed and density
and is the number of ants that pass a point in space over
time within the tunnel. At zero density, no ants are in the
tunnel and F = 0. At large density ants become jammed
together and speed approaches zero, in which case again
F = 0. Thus the flow is maximized with a value Fmax
at an intermediate density called the carrying capacity.
Flow curves constructed from the upper bound curves of
speed vs. density increased in size with increasing tunnel
diameter (Fig. 6a). The maximum flow achievable in
tunnels of varied diameter increased with tunnel diameter
(Fig. 6c).
Spatio-temporal traffic fluctuations

To gain insight into the process of formation and breakup of traffic jams, we measured spatial and temporal fluctuations of the tunnel density, (x, t). Temporal statistics
were determined by evaluating (x, t) at different fixed
points in space over time. Spatial statistics were determined by evaluating (x, t) in space at fixed times. The
time interval between ants crossing a point in spacethe
wait timeapproximately followed an exponential distribution (Fig. 7a) with a time constant that was similar
among the four tunnels. An exponential wait time distribution indicates that motion within the tunnel can be
considered as a random process.
We define the occupation time as the time an ant occupies a point in space (Fig. 7c). The distribution of occupation times among all tunnels were similar in that they
0.35cm
1.9cm/s
0.18 s
were peaked at a time of bodylength
vf ree
the occupation time of freely moving antsand had a

6
b)
Experiment

a)

Simulation

30

b)

Simulation

10-3

Experiment

10-1

F (ants/s)

F (ants/s)

a) 15

10-5
0

d) 12
Experiment

4
6
D (mm)

20

40

60

Density (ants/cm)

0
0

15

10-7

Simulation

c)

40 0
20
Wait time (s)
L / vfree
d)

20
40
Wait time (s)

10-1
4

10-3
0
0

Fmax (ants/s)

c) 12

5
10
Density (ants/cm)

Fmax (ants/s)

00

4
Deff

FIG. 6. Flow vs. density in tunnels of varying D. a-b) Flow

curves in experiment (a) and simulation (b). Experiment flow
are of increasing diameter as shown by arrow. Dashed box in
(b) shows the axis range of (a). c-d) Maximum flow vs. tunnel
diameter in experiment (c) and simulation (d).

power law tail with a slope 3 for all four tunnel diameters (Fig. 7c).
Finally, we measured the typical size (linear dimension) of interaction clusters. The cluster size of ant interactions was peaked at a value of a single ant bodylength
and had an exponential tail with slope that varied with
tunnel diameter. The magnitude of the exponential
length constant increased with D indicating that the
mean cluster size increased as tunnel sized decreased.
Traffic jam durations

We observed that the time duration of similar sized

jams (clusters) was sensitive to D (Fig. 5). Inspired
by studies of mobility in non-biological soft-matter systems like granular materials [31, 37, 38], we measured
the correlation function for density fluctuations, Qn ( ),
associated with traffic jams of size n. Qn ( ) measures
how quickly high density traffic-fluctuations returned to
steady flow. We observed that Qn ( ) curves decreased
from 1 to 0 over a characteristic time scale which varied over different tunnel diameters and jam sizes (Fig.
8a, 9a). For fixed tunnel diameter, curves of larger n
were shifted to the right indicating that increased
with cluster size. Comparing Qn ( ) curves of similar
n across D indicates that increased with decreasing
D. We measured by fitting an empirical function

Qn ( ) = e[( ) ] where is a fit parameter of order
unity.
The correlation time of ant aggregations, , increased
approximately linearly with increasing n (Fig. 9a) for
all tunnel sizes. The slope of as a function of n is a

10-5
10-7 -2
10
100
10210-2
100
102
Occupation time (s)
Occupation time (s)
FIG. 7. Probability distribution functions for ant stopping
and moving from experiment and simulation. a) Histogram
of wait time from the four tunnels in experiment. b) Distribution of wait times from varied tunnel diameters in simulation. Different color curves correspond to varied diameter
simulations which increase from upper to lower curves. c)
Distribution of site-occupation time from experiment. Note
log-log axes. Dashed line is curve form t3 . d) Distribution
of site-occupation times from simulation. Dashed line is same
as in c) for comparison.

measure of how sensitive traffic flow is to the formation

of jams from intermittent density fluctuations. We fit
lines to vs. n and observed that the normalized slope
of (n) (normalized such that limD = 1) increased with decreasing tunnel diameter (Fig. 10a). We
A
fit vs. n with a function of the form (DD
+ 1
C)
(chosen to match simulation results which we discuss in
the next section). This fit shows that traffic sensitivity
diverges at a tunnel diameter of 1.47 0.2 mm tunnel
diameter within which bi-directional traffic occurs.

Simulation of ant traffic

Wait time (Fig. 7b) and occupation time (Fig. 7d)

probability distributions from our simulations qualitatively matched the experimental results across a range of
D. Wait time distributions from the simulations had a
mixed shape that was not well described by either an exponential or a power law distribution. However the range
of values observed in experiment and simulation were in
accord. Occupation time probability distributions (Fig.
7d) from the simulations matched the experiment well
and were roughly power-law in shape with a slope 3,
similar to experiment.

DISCUSSION

a)

Experiment

Qn(t)

We performed similar measurements of the density correlation function in our cellular automata simulations
(Fig. 8b). The simulations reproduced features of the
experiment: ( (n) increased with n(Fig. 9b), and
diverged with decreasing tunnel diameter (Fig. 10b). To
quantify the effects of tunnel diameter and interaction
time on traffic flow we fit curves from simulation and
A
experiment to the empirical function = (DD
+1
c)
(solid lines in Fig. 10b).

n
b)

0
1

Simulation

We studied fire ant foraging traffic within tunnels of

diameter ranging from 2 - 6 mm. We observed that ant
colonies were able to maintain bi-directional traffic in all
tunnel diameters (SI Movie 2) and tunnel diameter only
appeared to negatively impact traffic flow in the 2 mm
tunnel. Head-on encounters frequently occurred within
the traffic flow and resulted in traffic jams that blocked
flow locally. Head-on encounters consisted of antennation, a primary mechanism of tactile information acquisition by ants [12]. We observed that Tint was larger in
2 mm tunnels compared to the other diameters. This is
likely due to the reduced lateral space which altered the
duration over which ants crossed paths.
We captured the basic features of bi-directional traffic flow and traffic jams that occur within tubular ant
tunnel traffic through a simple cellular automata simulation. The traffic dynamics in the cellular automata are
consistent with many of our experimental observations
(Fig. 5-10) given only three variable parameters. Traffic flow in the simulation at varied Def f exhibited similar speed-density scaling when the free-speed was set to
match that of the experiment (See methods and Fig. 5).
The upper bound curves of speed vs. density in ant traffic were qualitatively similar to measurements from vehicular, and pedestrian traffic [32, 33], and from surface
foraging ant traffic [34]. The density-flow relationship
in simulation across varied tunnel diameters exhibited a
similar shape for lower densities as in experiment (Fig.
6b), however a plateau in flow was observed for high densities in simulation that was not observed in experiment.
Flow-density curves predicted that Fmax should also increase with Def f (Fig. 6d) and we observed a similar
positive trend in experiment.
One goal of our simulations was to determine how tactile interactions or possibly traffic jam avoidance strategies may factor into the resultant traffic dynamics of the
ant colony. Another goal was to determine how the constraints of movement in confined tunnels limit the traffic flow of the ant colony. The agreement between the
simulation and experiment indicates that traffic flow in
subterranean environments is largely constrained by the

Qn(t)

Physical aspects of confined traffic

n
0

10-1

(s)

101

FIG. 8. Correlation function versus time for the 2 mm tunnel in experiment (a) and a traffic simulation (b). Curves of
different color correspond to different size traffic jams with
increasing n [2 15] shown by the arrow from left to right
for both (a) and (b).

physics of collective interactions in confined spaces. It

is not necessary to include complex behavioral rules for
interactions to reproduce the observed traffic dynamics.
The relevant features to this process are that the occupants of a tunnel exclude volume and halt to interact for
short periods of time. In this way the traffic flow of fire
ants within their nest may be similar to the bi-directional
traffic of termites [36, 40, 41], molecular motors along microtubules [1], or human pedestrians in crowded hallways
[4244].
Traffic in natural nests

Field observations of excavated tunnel diameters in

natural foraging tunnels, foraging tunnel entrances, and
laboratory excavated tunnels show that tunnel diameter
in a nest varies from 3-14 mm in diameter with some
tunnels exceeding this range [14, 21, 27] (Fig. 1c). Excavated horizontal foraging tunnels are 6-14 mm in diameter [21, 27] and vertical nest entrance tunnels are slightly
smaller with 3-4 mm diameter [14, 27]. A recent study
has shown that smaller tunnel diameter nest entrances
enhance climbing performance [14] and additionally, the
smaller nest entrance size may enhance security [12].
The tunnel diameter of horizontal foraging tunnels

a)

Experiment

* (s)

b)

0
Simulation

* (s)

where a majority of the colony traffic occurslikely plays

several important roles in traffic dynamics. Our experiments show that the carrying capacity of a tunnel and
the maximum flow of workers increases linearly with D
(Fig. 6c) indicating that we expect to find larger diameter tunnels where traffic flow is higher. Our simulations
also exhibit this result (Fig. 6d) and furthermore studies of road traffic in city centers show that vehicles obey
the same linear relationship between flow and road width
[35]. In natural systems there is observational evidence
that tunnel diameter increases in regions of higher traffic
in fire ant foraging tunnels [21], and termites have been
observed to modify artificial tunnels in regions of high
traffic [36] suggesting that social organisms control their
tunnel diameter based on local traffic demands.
The simulation results have implications for tunnel
construction energetics. The throughput of workers
and thus resource and information throughput
increases linearly with D. However, for a circular crosssection tunnel, D2 soil must be excavated per unit length
and thus the relative energetic cost to traffic flow gains
follows a surface area to volume scaling law. The scaling of resource throughput per energetic cost of the tunnel suggests that larger tunnels are not always better.
These results present two hypothesis for the shape of
natural foraging tunnels: 1) ants should construct noncircular cross sections which have large perimeter (walking path) and small area (excavation volume), 2) ants
should construct many smaller tunnels which would provide the same net throughput as one larger tunnel but
at a lower energetic cost. There is evidence that fire ants
have adopted both strategies to manage effective resource
transport while minimizing energetic costs of construction. From excavations of foraging tunnels Tschinkel [21]
determined that the cross-sections of fire ant foraging
tunnels are elliptical in shape with aspect ratio 2 (the
major axis is horizontal to the ground), thus providing
a larger perimeter for the same cross-sectional area. Additionally, fire ant foraging tunnel networks begin as a
single tunnel that branches many times along its length
into smaller tunnels (Fig. 1b) [27]. The reduction of
tunnel diameter at branches [21] may correspond to an
excavation strategy to maximize perimeter while minimizing energetic costs of excavation.
We can address the question of whether fire ant traffic is susceptible to traffic jams by examining the
curve in relation to the size of tunnels in natural nests
(Fig. 10a). From measurements of the cross-sectional
area of fire ant foraging tunnels in [21], we estimate the
effective diameter to be in the range, D = 7.8 1.9 mm
(tunnels are elliptical in cross-section with eccentricity
of 2 [27]). Vertical nest entrance tunnels in natural
nests were reported in the range of 3-4 mm in diameter and a laboratory x-ray study found that vertical
tunnels were D = 3.7 0.8 [14]. is a diverging
function with decreasing diameter and predicts that bi-

10

FIG. 9. Traffic jam correlation time vs. jam size for different
diameter tunnels. a) Traffic-jam correlation time as a function
of n for increasing tunnel diameter (arrow) in experiment. b)
Traffic-jam correlation time vs. n in simulation for ranges of
Def f [2, 100].

directional traffic should be impossible in tunnels smaller

than Dc = 1.47 mm. This minimum tunnel diameter is
consistent with a previous locomotion study of climbing
in tunnels in which velocity dropped to zero near this
tunnel size[14]. However, over the range of natural tunnel diameters is fairly flat (Fig. 10a). This suggests
that in natural nest tunnels fire ant traffic is not sensitive to fluctuations in density. These observations are
similar to those of above ground foraging traffic in the
ant Leptogenys processionalis in which an absence of a
jammed-phase was observed [39]. Thus we have found
that in addition to behavioral adaptations which some
ant species posses to mitigate traffic jams along trails
(See [19] for review), ants may also modify their environment to allow for smooth traffic flow and resource
acquisition.
By varying the interaction time between ants we are
able to modify the curve in our simulation (Fig.
10b). Increasing Tint results in a shift of to the right
which means that traffic flow becomes increasingly sensitive to traffic jam formation in larger tunnels. This increase in jam sensitivity is due to ants forming blockages
in the tunnels for longer periods of time. We find that
the increase in jam sensitivity is linearly proportional to
Tint . This suggests that to maintain stable traffic flow

a) 3

Experiment

2
1
0
0

Entrance
tunnels

Foraging
tunnels

10
D (mm)

b) 3

15

Simulation

Tint

2
1
0
0

10

20
Deff (mm)

30

FIG. 10. Sensitivity of the slope of as a function of tunnel diameter. a) Results from experiment. Red curve is fit
A
Red points with
function (DD
+ 1 described in text.
c)
bars show diameter of foraging and entrance tunnels (Foraging tunnels from [21], entrance tunnels in lab from [14] and
field from [27]). b) Results in simulation. Different points
and curves correspond to simulations with different interaction times. Arrow indicates increasing Tint .

if ant interaction time increases, tunnel diameter should

increase as well.

in organismal dynamics. Conserved features of nest morphology such as tunnel size or shape [18] are examples of
an organisms extended phenotype. Whether features of
this extended phenotypethe nestare adaptations for
subterranean life are unknown. Our results have implications for the effective design of jam-free subterranean
environments. Observation of the sizes of natural and
laboratory excavated tunnels shows that fire ant groups
construct tunnels outside the size range in which traffic flow becomes sensitive to traffic jams. Thus mobility
within natural foraging tunnels is likely not subject to
large scale traffic jams.
Our traffic model consists of simple interactions in confined space and the resultant dynamics may be applicable
to a broad range of organisms that live and interact in
subterranean environments. The approach of generating predictions from a simple single parameter computational model allows for the sensitive variation of an individual behavior (interaction time) and explore its resultant effect on the complex, collective process of foraging
traffic. This approach is in contrast to other modeling
efforts which include many different behaviors and phenomena but sacrifice the ability to explore simple cause
and effect. Finally, the insight we have gained into collective locomotion in crowded environments may be used
for the design of synthetic systems to collectively navigate disaster zones, extra-terrestrial environments, or the
natural world.
The authors would like to acknowledge Gregg Rodriguez for help with preliminary experiments. Funding
for N.G., D.I.G., and M.A.D.G. provided by NSF PoLS
#0957659 and #PHY-1205878.

[1]
CONCLUSION

We studied the bi-directional traffic of fire ant worker

groups to determine how modulation of tunnel diameter
and ant interaction behavior may influence traffic flow.
We found that the resultant traffic dynamics are in a
large part governed by the physical mechanics of collective interactions in confined space. Through simulation
we find that by modulating the characteristic interaction
time between worker ants, we may vary the sensitivity of
the colony as a whole to the formation and longevity of
traffic jams.
Effective traffic flow and transport of resources within
an ant nest are important for the survival and success
of a colony. The relationship between the behaviors and
nest morphology of social organisms is an open question

[2]
[3]

[4]
[5]
[6]
[7]
[8]
[9]

[10]

School of Engineering and Applied Sciences, Harvard

University, Cambridge, MA 02138, USA
C. Leduc, K. Padberg-Gehle, V. Varga, D. Helbing,
S. Diez, and J. Howard, Proceedings of the National
Academy of Sciences 109, 6100 (2012).
Y. Chai, S. Klumpp, M. J. M
uller, and R. Lipowsky,
Physical review E 80, 041928 (2009).
L. Conway, D. Wood, E. T
uzel, and J. L. Ross, Proceedings of the National Academy of Sciences 109, 20814
(2012).
D. Chowdhury, A. Schadschneider, and K. Nishinari,
Physics of Life reviews 2, 318 (2005).
D. Helbing, Reviews of modern physics 73, 1067 (2001).
T. Nagatani, Reports on progress in physics 65, 1331
(2002).
B. S. Kerner and H. Rehborn, Physical Review E 53,
R1297 (1996).
R. Lipowsky, S. Klumpp, and T. M. Nieuwenhuizen,
Physical review letters 87, 108101 (2001).
T. E. Angelini, E. Hannezo, X. Trepat, J. J. Fredberg,
and D. A. Weitz, Physical review letters 104, 168104
(2010).
K. Painter, Journal of mathematical biology 58, 511

10
(2009).
[11] T. Vicsek and A. Zafeiris, Physics Reports (2012).
[12] B. H
olldobler and E. Wilson, The ants (Harvard University Press, 1990).
[13] D. Cassill, Behavioral Ecology and Sociobiology 54, 441
(2003).
[14] N. Gravish, D. Monaenkova, M. A. Goodisman, and
D. I. Goldman, Proceedings of the National Academy of
Sciences 110, 9746 (2013).
[15] F. R. Jayaram, J., Integrative and Comparative Biology
53(1) (2013).
[16] A. M. Horner and A. R. Biknevicius, Zoology 113, 189
(2010).
[17] J. S. Turner, The extended organism: the physiology of
animal-built structures (Harvard University Press, 2009).
[18] W. R. Tschinkel, Palaeogeography, Palaeoclimatology,
Palaeoecology 192, 321 (2003).
[19] M. Burd, Physica A: Statistical Mechanics and its Applications 372, 124 (2006).
[20] W. R. Tschinkel, The fire ants (Harvard University Press,
2006).
[21] W. R. Tschinkel, Journal of Insect Science 11 (2011).
[22] I. D. Couzin and N. R. Franks, Proceedings of the Royal
Society of London. Series B: Biological Sciences 270, 139
(2003).
[23] J. D. Shepherd, Behavioral Ecology and Sociobiology 11,
77 (1982).
[24] A. Dussutour, S. Beshers, J.-L. Deneubourg,
and
V. Fourcassie, Journal of Experimental Biology 212, 499
(2009).
[25] M. Burd, D. Archer, N. Aranwela, and D. J. Stradling,
The American Naturalist 159, 283 (2002).
[26] N. Gravish, M. Garcia, N. Mazouchova, L. Levy, P. B.
Umbanhowar, M. A. Goodisman, and D. I. Goldman,
Journal of The Royal Society Interface 9, 3312 (2012).
[27] G. P. Markin, J. ONeal, and J. Dillier, Journal of the

Kansas Entomological Society , 83 (1975).

[28] M. Bollazzi and F. Roces, PloS one 6, e17667 (2011).
[29] M. J. Greene and D. M. Gordon, Behavioral Ecology 18,
451 (2007).
[30] H. Torres-Contreras and R. A. Vasquez, Ethology 113,
76 (2007).
[31] K. N. Nordstrom, J. P. Gollub, and D. J. Durian, Physical Review E 84, 021403 (2011).
[32] A. Seyfried, B. Steffen, W. Klingsch, and M. Boltes,
Journal of Statistical Mechanics: Theory and Experiment 2005, P10002 (2005).
[33] W. Daamen and S. P. Hoogendoorn, Transportation Research Record: Journal of the Transportation Research
Board 1828, 20 (2003).
[34] M. Burd and N. Aranwela, Insectes Sociaux 50, 3 (2003).
[35] R. Smeed, Highway Research Record (1967).
[36] S.-H. Lee, P. Bardunias, and N.-Y. Su, Behavioural processes 77, 135 (2008).
[37] S. C. Glotzer, V. N. Novikov, and T. B. Schrder, The
Journal of Chemical Physics 112, 509 (2000).
[38] H. Katsuragi, A. Abate, and D. Durian, Soft Matter 6,
3023 (2010).
[39] A. John, A. Schadschneider, D. Chowdhury,
and
K. Nishinari, Physical review letters 102, 108001 (2009).
[40] S. Sim and S.-H. Lee, Insect Science (2012).
[41] S. J. Ku, N.-Y. Su, and S.-H. Lee, Florida Entomologist
96, 810 (2013).
[42] W. H. Lam, J. Lee, K. Chan, and P. Goh, Transportation
Research Part A: Policy and Practice 37, 789 (2003).
[43] A. Schadschneider, W. Klingsch, H. Kl
upfel, T. Kretz,
C. Rogsch, and A. Seyfried, in Extreme Environmental
Events (Springer, 2011) pp. 517550.
[44] A. Seyfried, O. Passon, B. Steffen, M. Boltes, T. Rupprecht, and W. Klingsch, Transportation Science 43,
395 (2009).