07.01.2013 Views

Lecture Notes in Computer Science 3472

Lecture Notes in Computer Science 3472

Lecture Notes in Computer Science 3472

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

v1<br />

u1<br />

w1<br />

1<br />

2<br />

sP<br />

λ<br />

a<br />

9 Test<strong>in</strong>g Theory for Probabilistic Systems 273<br />

1<br />

2<br />

v2<br />

b b<br />

u2<br />

c d<br />

w2<br />

u ′ 1<br />

w ′ 1<br />

1<br />

2<br />

sQ<br />

v<br />

µ<br />

a<br />

b<br />

1<br />

2<br />

u ′ 2<br />

c d<br />

Fig. 9.17. Q ⊑LS P and P �⊑LS Q.<br />

Example. Consider P, Q ∈ PP shown <strong>in</strong> Figure 9.17 (see also Example 9.8,<br />

page 253). We have Q ⊑LS P and P �⊑LS Q. If we apply the test process T of<br />

Figure 9.16, we can derive for o =1a :(1b :1c :1ω, 1b :0d)<br />

Pr Obs 1<br />

1<br />

1<br />

P,T (o) = 2 · [(1 · 1) · (1 · 1)] + 2 · [(1 · 0) · (1 · 0)] = 2 ,<br />

Pr Obs<br />

1 1 1 1<br />

Q,T (o) =[1 2 · 1+ 2 · 0] · [ 2 · 1+ 2 · 0] = 4 .<br />

So we obta<strong>in</strong> P �⊑LS Q, but recall we have P ∼JY Q, Q ⊑SE P and P �⊑SE Q<br />

(see Example 9.7.2).<br />

⊓⊔<br />

Larsen and Skou proved that the follow<strong>in</strong>g relationship between their notion of<br />

test<strong>in</strong>g and the probabilistic bisimulation holds:<br />

Theorem 9.36. [LS91] Let P, Q ∈ PP be τ-free and reactive. Then<br />

P ∼ bs QiffP∼LS Q.<br />

⊓⊔<br />

Note that the previous theorem also holds for probabilistic processes that are not<br />

f<strong>in</strong>itely branch<strong>in</strong>g but fulfill the m<strong>in</strong>imal probability assumption, i.e. there exists<br />

ɛ>0 such that whenever s a −→ µ either µ(s ′ )=0orµ(s ′ ) >ɛfor all s ′ .The<br />

difference between ∼LS and the test<strong>in</strong>g preorders def<strong>in</strong>ed <strong>in</strong> the previous sections<br />

is that ∼LS considers success or failure ”after each step” whereas test<strong>in</strong>g relations<br />

for probabilistic processes like ⊑SE or ⊑JY take only the success probability after<br />

a (maximal) trace <strong>in</strong>to account. So it is clear that ∼LS dist<strong>in</strong>guishes processes<br />

that can not be dist<strong>in</strong>guished by ⊑SE or ⊑JY . To the best of our knowledge this<br />

relationship has not been further considered yet.<br />

w ′ 2

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!