258
■
Bio-Inspired Computing and Networking
min
w
(i,j)∈E
u
2w
ij
subject to
−1/2
L
−1/2
I −vv
T
w ≥ 0 (10.22)
where w
ij
is the set of undirected edge weights, v =[(¯x
1
)
1/2
···(¯x
M
)
1/2
]
T
,and
= diag(¯x). L is the M ×M weighted Laplacian of G
u
;its(i, j) entries are defined
as L
ij
=−w
ij
if (i, j) ∈ E
u
, L
ii
=
(i, j)∈E
u
w
ij
,andL
ij
= 0 otherwise. Here, the
optimization variable is w and the transition rates k
ij
are derived from w according
to (10.3). This program is similar to program (11) by Sun et al.
39
for finding the
fastest mixing reversible Markov process on a graph. The optimization problem can
be further extended to more general strongly connected graphs.
40
To find the optimal K for a given initial configuration x
0
, K
∗
(x
0
), Metropolis
optimization
41
is used with the ...