28.05.2013 Views

Etude des marchés d'assurance non-vie à l'aide d'équilibres de ...

Etude des marchés d'assurance non-vie à l'aide d'équilibres de ...

Etude des marchés d'assurance non-vie à l'aide d'équilibres de ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

tel-00703797, version 2 - 7 Jun 2012<br />

Chapitre 3. Calcul d’équilibre <strong>de</strong> Nash généralisé<br />

We conclu<strong>de</strong> that φF B is strongly semismooth at (0, 0), as proved in Kanzow and Kleinmichel<br />

(1998).<br />

Now, we can express appropriately the generalized Jacobian of the GNEP. We <strong>de</strong>note by<br />

JΦ(z) elements of the generalized Jacobian ∂Φ(z). Using chain rules and previous <strong>de</strong>finitions,<br />

we have<br />

<br />

Jac<br />

JΦ(z) =<br />

xD(x, λ) diag ∇xigi (x) <br />

i , (3.20)<br />

−Da(z)Jac xg(x) Db(z)<br />

where Jac x <strong>de</strong>notes the Jacobian with respect to x and diag[...] represents a block diagonal<br />

matrix, see Appendix 3.5.1 for a exten<strong>de</strong>d representation of the generalized Jacobian JΦ.<br />

The diagonal matrices Da and Db are given by<br />

Da(z) = diag[a 1 (x, λ 1 ), . . . , a N (x, λ N )] and Db(z) = diag[b 1 (x, λ 1 ), . . . , b N (x, λ N )],<br />

with ai (x, λi ), bi (x, λi ) ∈ Rmi <strong>de</strong>fined as<br />

(a i j(x, λ i j), b i j(x, λ i <br />

j)) =<br />

φ ′ a(−gi j (x), λij ), φ′ b (−gi j (x), λij )<br />

<br />

if (−gi j (x), λi j ) = (0, 0),<br />

(ξij, ζij) if (−gi j (x), λij ) = (0, 0),<br />

where φ ′ a (resp. φ ′ b ) <strong>de</strong>notes the <strong>de</strong>rivative of φ with respect to the first (second) argument a<br />

(b) and (ξij, ζij) ∈ ¯ B(pφ, cφ), the closed ball at pφ of radius cφ.<br />

Let us precise the top-left part.<br />

⎛<br />

Jac xD(x, λ) =<br />

⎜<br />

⎝<br />

Jac x1 L1(x, λ 1 ) . . . Jac xN L1(x, λ 1 )<br />

.<br />

Jac x1 LN(x, λ N ) . . . Jac xN LN(x, λ N )<br />

where Li(x, λi ) = ∇xiθi(x)+ mi j=1 ∇xigi j (x)λij . The top-right part is block diagonal and given<br />

by<br />

⎛<br />

Jac x1<br />

⎜<br />

⎝<br />

g1 (x) T 0<br />

. ..<br />

0 Jac xN gN (x) T<br />

⎞<br />

⎟<br />

⎠ .<br />

Let us specify the parameters pφ and cφ for the two consi<strong>de</strong>red complementarity functions.<br />

For the minimum function, using Equation (3.17), we have pφ = (1/2, 1/2) and cφ = 1/2.<br />

For the Fischer-Burmeister function, using Equation (3.19), we have pφ = (−1, −1) and cφ =<br />

1. We refer to Kanzow and Kleinmichel (1998) and Facchinei and Pang (2003) for other<br />

complementarity functions.<br />

If functions θi and gi are Ck+1 , by the chain rule, the root function Φ <strong>de</strong>fined in Equation<br />

(3.20) is Ck except at points z such that gi j (x) = λij = 0. At these points, when the<br />

complementarity function is φ∧, Φ is semismooth, while for φF B, Φ is strongly semismooth.<br />

Extension and local convergence in the semismooth framework<br />

As the Jacobian of the root function is not available, the direction computation of local<br />

methods presented in Subsection 3.2.1 must be adapted. The solution consists in replacing<br />

the Jacobian by an element of the generalized Jacobian.<br />

Consi<strong>de</strong>ring the Newton method (3.12), the direction solves<br />

154<br />

.<br />

⎞<br />

⎟<br />

⎠ ,<br />

Jkdk = −F (zk), (3.21)

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

Saved successfully!

Ooh no, something went wrong!