\$title Test for NL code bug from Dist 23.6 (NLCODE5,SEQ=532) \$ontext This example turfs up a bug in the NL code generation. The 23.5.2 system (last in the 23.5 series) was OK. The problem exists in 23.6.0 alpha, in 23.6.5, and probably all of 23.6 Contributor: Kent Zhao, June 2011 \$offtext SETS t Time periods / 1*5 / tfirst(t) tlast(t) t2(t); tfirst(t) = ord(t)=1; tlast (t) = ord(t)=card(t); t2 (t) = ord(t) <= card(t)-2; display tfirst,tlast,t2; PARAMETERS alpha /0.33/ theta /0.7/ gamma_n(t) Nbar /0.7921/ Am(t) An(t) beta /0.95/ delta /0.04/ gamma_m /0.04/ K0 /1/ gamma_n(t); gamma_n(t) = -0.000321469; * use integer power instead of real power, ord(t)-1 > t.off Am(t) = 1*power(1+gamma_m ,t.off); An(t) = 1*power(1+gamma_n(t),t.off); Variables h(t) total available time K(t) capital stock ; Equation EQ_h(t) h(t) EQ_K(t) K(t) EQ_KT ; EQ_h(t).. h(t) =E= 1 - (Nbar/An(t))**(1/theta); EQ_K(t2(t)).. K(t+2) =E= (Am(t+1)*(K(t+1)/h(t+1))**alpha)*h(t+1) + (1-delta)*K(t+1) - beta*(1-delta + alpha*Am(t+1)*((K(t+1)/h(t+1))**(alpha-1))) *(Am(t)*((K(t)/h(t))**alpha)*h(t) + (1-delta)*K(t) -K(t+1)); EQ_KT(tlast(t)).. (K(t )-(1-delta)*K(t-1)) /(K(t-1)-(1-delta)*K(t-2)) - (Am(t-1)*((K(t-1)/h(t-1))**alpha)*h(t-1)) /(Am(t-2)*((K(t-2)/h(t-2))**alpha)*h(t-2)) =E= 0; model m /EQ_h, EQ_K, EQ_KT/; K.lo(t) = 0.1; h.lo(t) = 0.01; h.up(t) = 1; K.fx(tfirst) = K0; \$ontext h.l(t) = 0.8; K.l(t) = K0; \$offtext execute_loadpoint 'nlcode5'; option cns = conopt; solve m using CNS; abort\$(m.solvestat <> %solvestat.NormalCompletion%) 'Expected solvestat Normal Completion'; abort\$(m.modelstat <> %modelstat.Solved%) 'Expected modelstat Solved'; abort\$(m.iterusd > 3) 'Expected to start at a solution', m.iterusd; \$ontext save a solution for comparison & restarting execute_unload 'badnl.gdx', h, k; \$offtext