Consider the following Lagrangian: $$L=mR\left[\frac{1}{2}R\left(\dot{\theta}^2+\omega^{2}\sin^{2}\theta\right)+g\cos\theta\right],$$ with an associated Hamiltonian $$H=mR\left[\frac{1}{2}R\left(\dot{\theta}^2-\omega^{2}\sin^{2}\theta\right)-g\cos \theta\right].$$
If I take the Euler-Lagrange Equation, along with the small angle approximation $$\sin\theta=\theta,\,\cos\theta=1,$$ I end up with the second order ODE: $$\ddot{\theta}=\left(\omega^2-\frac{g}{R}\right)\theta.$$ However, the conservation of the Hamiltonian (i.e., treating it as a constant), $$\frac{\partial L}{\partial\dot{\theta}}\dot{\theta}-L=H,$$ yields an entirely different differential equation with different solutions to the one generated by using the standard Euler-Lagrange equation: $$\frac{H}{m}=\frac{1}{2}R\omega^2\theta^2-\frac{1}{2}R\dot{\theta}^2+g.$$ This leads to an entirely different result. Why is this happening?