In general there is nothing different with nonlinear models in TomSym. They are modeled just like a linear or quadratic problem. The modeling engine takes care of the rest, including first and second order derivatives and problem sparsity patterns.
toms x1 x2
alpha = 100;
% Objective function
f = alpha*(x2-x1^2)^2 + (1-x1)^2;
% Constraints
c = -x1^2 - x2;
con = {
-1000 <= c <= 0
-10 <= x1 <= 2
-10 <= x2 <= 2
};
% Initial conditions
x0 = {
x1 == -1.2
x2 == 1
};
% Compile and solve problem
options = struct;
options.name = 'Rosenbrocks banana';
solution = ezsolve(f,con,x0,options);