Note:   Regarding T. Goff and D. S. Phatak, "Unified transport layer support for data striping and host mobility," IEEE Journal on Selected Areas in Communications, vol. 22, no. 4, pp. 737-746, May 2004.
The analysis in section C., Comparison of Network and Transport Layer Striping, assumes small path loss probabilities. In particular, (3) is an approximation to the exact loss probability of the overall aggregated path:
Lns
=
1 - n

i = 1 
(1 - li)
=
n

i = 1 
li - n

i = 1 
n

j = i + 1 
li ·lj + n

i = 1 
n

j = i + 1 
n

k = j + 1 
li ·lj ·lk - +  .
For small path loss probabilities, the product of two or more li terms quickly becomes insignificant and can be approximated by the first summation from the expression above as simply
Lns n

i = 1 
li  .
An alternate approach to finding a lower bound on the aggregate loss probability is to consider the expected number of packets successfully sent before a loss occurs on each path. The expected number of successfully sent packets is 1 / li for packet loss probability li. This limits the overall expected number of packets sent before a loss is experienced to at most i = 1n 1 / li. The aggregate average loss probability can then be expressed as
Lns = 1

n

i = 1 
1 / li
Lns  .
The end result remains the same in this case since
  


n

i = 1 
1

li
 
< n

i = 1 
  


1

li
 
for 0 < li < 1.