I will ask my question about Lagrange multipliers by using an example in string theory, as this question was inspired by my string theory course, but it applies to every theory with Lagrange multipliers I guess. In string theory, we are taught that one can obtain the Nambu-Goto action from the string sigma-model action by varying the latter with respect to the induced metric $h_{\alpha\beta}$ from $$S_{\sigma}=-\frac{T}{2}\int d^2\sigma\sqrt{-h}h^{\alpha\beta}\partial_{\alpha}X \cdot\partial_{\beta}X$$ (where I use the notation of "String theory and M-theory" written by Becker, Becker and Schwarz), and requiring the variation of the action $S_{\sigma}$ with respect to the induced metric $h^{\alpha\beta}$ be zero. This yields the classical equations of motion and when they satisfied, we obtain the Nambu-Goto action. So far so good...
However, what happens if for some reason the equations of motion are not satisfied? Can something like that take place for whatever reason? For instance one could argue that if we were to promote the components of the induced metric to quantum operators, then the latter components would not have to obey the classical equation of motion. This is an example coming to mind for a circumstance in which the induced metric does not satisfy the equation of motion and, therefore, one can meaningfully demand that the equations of motion be satisfied.
But, the induced metric components are not dynamical. So, there is no way of having a conjugate momentum to promote to quantum operators with. Am I lead to believe that the equations of motion must be satisfied, or am I lead to believe that the equations of motion can not be satisfied for some other reason (and which reason is that)?
I guess the same question applies every time we introduce a Lagrange multiplier into a theory, regardless whether this is string theory or not, but nevertheless...
Any help will be appreciated.