Mathematical model for path finding
P. polycephalum is a large, single-celled amoeboid organism that
can form a dynamic tubular network linking the discovered food
sources during foraging. And the physiological mechanism behind
the tube formation and selection contributes to the Physarum’s ability
of path finding, which is: tubes thicken in a given direction
when the flux through it persists in that direction for a certain
time. With this mechanism, a mathematical model for path
finding has been constructed to simulate the adaptive dynamics
of tubular network. A brief introduction for the model is given
below.
Assume the shape of Physarum is represented by a graph, in
which a plasmodial tube refers to an edge of the graph and a junction
between tubes refers to a node. Two special nodes acting as
food sources are labelled as N1 and N2 in the graph, while other
nodes are labelled as N3, N4, N5, . . .. And the edge between node
i and j is labelled as Mij. The variable Qij denotes the flux through
With the mechanism between the flux and tube thickness
(described as the conductivity ofthe tube),the adaptation equation
for conductivity is constructed as follows:
d
dt
Dij = f(|Qij|) − ˛Dij (3)
It is obvious that positive feedback exists in the model, as f(Q) is
a increasing function with f(0) = 0.
3. Rapid Physarum Algorithm
With experiments of path finding process of the mathematical
model described in Section 2, it is observed that although the
shortest paths can be automatically selected as others disappear,
much time and calculation is required to derive the final solution.
However, in some SP applications, it is required that shortest
paths need to be quickly identified either because an immediate
response isneededor because the shortestpathsneedto be recalculated
repeatedly. For this reason, a number of heuristic approaches
have been advocated for decreasing the computation time. In this
section, with the integration of one heuristic rule extracted from
experiments, a Rapid Physarum Algorithm (RPA) is proposed to
improve the efficiency of the original mathematical model.
3.1. Extraction of the heuristic rule
With observation of path finding process of the model in fully
connected graphs, some phenomena have been found. After several
iterations, the flux through each tube shows regularity in variation,
which means that the flux through some tubes keeps increasing as
the tubes thicken, while other keeps decreasing as the tubes thinner
untiltheflux converge to 0 and the tubes disappear. However, when
the flux through the tube decreases to a very small value. Normally,
when the flux is lower than a small value , a lot of iterations are
still needed until the tube vanishes. ’s value varies from network
to network, which is determined by the properties of the network,
including its structure,the edge length, etc. According to our observation,
when is in the range [0, 0.1] in most of the networks, we
can change the flux associated with corresponding edges to 0. As
we can see in Fig. 1, almost 2/3 of the iterations are used for paths
to adjust its flux to decrease from a small value to zero.
If some heuristic rules can be integrated to reduce the iteration
process by cutting those tubes which will potentially vanish, the
efficiency of the path finding process can be improved with less
computation complexity. Therefore, after a detailed study into our
simulation of the model and results, an empirical heuristic rule is
adopted from those phenomena and integrated into the model.
Rule: If the variation of the flux through the tube maintains a stable
decreasing trend, then the tube will be cut.
This heuristic rule means that if a “stable decreasing trend” is
determined through variation process of the flux which may imply
the flux will continue decrease until it reaches 0, the tube will not
be part of shortest path and is deleted directly with its flux assigned
0. The “stable decreasing trend” criterion is met if Qvij, the times for
which the flux has continuously decreased/increased, satisfies the
following condition:
S(Qvij, Qij) = Qvij + C1 + R1(Qij) + R2(n) (4)
where positive value of Qvij means the flux has continuously
increased for Qvij time while negative value means continuously
decreasing times. And R1(Qij) denotes the influence of current flux
Qij on S. If Qij is very small, the tube is more likely to vanish and
small |Qvij| is needed to identify “stable decreasing trend”. In contrast,
if Qij is very large, more iterations are needed in case of wrong
decision on cutting tubes. What is more, considering that the scale
of the SP problem can also affect the decision criterion, R2(n) is