On
a= yeXpip
s
—
= c4(R)(1+ 83)(1+ v3\S"iB) - GIB) |,
KA9)
respectively, where
\BKy"],
e* yan _ (eOKrv _ et
(SIB) = (BK + Ayty? [(1 _
+
(stp) =(BK + ay?{aay+ e-*[(1- 26" )2pKiy"
;
2(eK _ 1)(BK _ Ay?
_ (2a) I}
The averages (SIB)and (S?1B) with respect to H were each evaluated
numerically for different B values equally spaced over the rangejof 8,
whereupon it was found that oygt”’ is for each given t,O<t <WOy,a
virtually linear function of (X|B¥7™ over a B - and t-dependent range
pf the
latter, and furthermore that corresponding (XIBy¥7 values are vimtually
uniformly distributed over these linear ranges (Figure D3).
The
coefficients {a,b|t} and corresponding (XIS\¥"'-range boundaries {x,,
systematic Latin-Hypercube sampling procedures. Calculations were d
a NeXT workstation using the programs Mathematica (Wolfram, 199[) and
RiskQ (Bogen, 1992). Analyses of quantile convergence indicated that Q01- to
0.99-fractiles obtained are accurate to within ~1 to 5%.
D-7