∞
^ δt [Cy (0,ft (y)) + Dx(ft(y)) J fx(y), which is the condition in part a. Hence, the condition in part c
t=1
implies part a and the invasion is a controlled invasion from y.
Let T be the first t such that at > 0 and suppose that T > 0. Then at = 0 for t = 0,...,T-1, while 0 <
aτ ≤yτ = f(xT-1) = fT(y) and xτ = yτ - aτ = fT(y) - aT. Since aτ > 0 there exists an ɛ > 0 such that
f(yτ-1) - f(yT-1 - ε) < aT. Consider the alternative feasible sequence 0T.1 = ε, aT = aT - [f(yT-1) - f(yT-1-ε)], and
at = at for all other t. Then xτ-1 = yτ-1 - ɛ and xt = xt for t ≠Τ-1. Since the sequence {xt,at} is optimal,
∞
0 ≥ ∑ δt [C(a,,y,) + D(x,) - C(a,,yt) - D(X,)]
t=0
= δτ-1[ C(0,yT-ι) + D(yT-ι) - C(ε,yT-ι) - D(yT-ι-ε) ]
+ δT [C(aT,f(yT-ι)) + D(xT) - C(aT+f(yT-1-ε)-f(yT-1),f(yT-1-ε)) - D(xT) ]
Dividing by ε and letting ε → 0 implies
0 ≥ -Ca(0,fT-1(y)) + Dx(fT-1(y)) + δ[Ca(aT,fT(y)) + Cy(aT,fT(y))]fx(fr-1(y))
Since this contradicts the condition in the proposition it must be that T = 0 and A(y) > 0 for all y. ■
Proof of Lemma 4. For purposes of exposition the proof is divided into a sequence of subsidiary lemmas.
Define the lower, right and left Dini derivatives of V at y by
V(y +ε) -V(y) V(y) -V(y -ε)
D+ V ( y ) = lιmιnfe→0 ——-------— and D- V ( y ) = ——---—----.
εε
Lemma 4.1. -Ca(y0-x0,y0) + Dx(x0) + δ D~V(f(x0)) ≤ 0.
Proof. Since x0- is feasible from y0, the principle of optimality implies C(y0-x0+ ,y0) + D(x0- ) +
δV(f(x0-ε)) - C(y0-x0,y0) - D(x0) - δV(f(x0)) ≥ 0. Dividing by ε and taking liminfεt0 establishes the result.
■
24