the generalized eigenvalue problem det(P-λQ) = 0,
we need to transform one of our matrices such that its limiting matrix becomes a.s.
nonsingular. As will be shown below, choosing Pn = Âm and Qn = Bm + n2Am 1 yields a
suitable pair (Pn , Qn ), such that if rank(C(1)C(1)T) = q-r then the q-r largest solutions of
det(P-λQ) = 0 are a.s. positive and free of nuisance parameters, whereas the r smallest
solutions are zero.
Now choose the functions Fk such that
1 n
(6
(7
—∑Fk( t / n ) ^ o (1),
√n t"1
1
n∖Jn
n
Σ tFk( t /n ) = o (1),
t= 1
and for i ≠ j,
ʃʃFi (x ) Fj (y )min(x ,y ) dxdy = 0, (8)
ʃFj(x)jF(y)dydx o 0, (9)
0
ʃFj (x)Fj(x) dx o 0. (10)
Note that the integrals involved are taken over the unit interval [0,1] if not otherwise
indicated, as will be in the sequel. It is a standard exercise in Wiener measure calculus to
show (see, e.g., Billingsley 1968, Phillips 1987, Bierens 1994, Ch.9) that for each k,
( ʌ
Mnz(Fk)/F
M∆ (Fk ) F
V k √
Cd)ʃF,(x) W(x )dx
C (1)(F,(1) W(1) -ʃf,( x ) W(x )dx),
N2q(0, (C(1)C(1)T)®Σk),
(11
where W is a q-variate standard Wiener process, fk is the derivative of Fk, and
Σk
ʃʃFk(x )Fk(y )min(x ,y ) dxdy
0
ʌ
0
ʃFk(x )2 dx
(12