If $\vec{f}\in C^{\infty}(\mathbb{R}^D, \mathbb{R}^K)$, $K=D-d$, such that the Jacobian, $D\vec{f}(\vec{x})$ of $\vec{f}$ has full-rank (i.e. $=K$) on $M=\vec{f}^{-1}(\vec{0})$, then $M$ is a smooth $d$-dimensional submanifold of $\mathbb{R}^D$. This is a consequence of the implicit function theorem.
Question: What sort of smooth manifolds fail to have this representation and are there any classic references for such examples? In other words: are all smooth manifolds the intersection of smooth hypersurfaces?
My intuition tells me that self-intersections are problematic and also possibly orientation-issues. For example, and somewhat informally, consider a figure-eight closed curve in the plane with the point of self-intersection situated at the origin. There is not a unique tangent vector at this point, and thus the only vector that can be orthogonal to all such tangents is the zero vector-so that the gradient of any possible "implicit" function must vanish here--and so $f$ cannot satisfy the hypotheses above. Is this correct?
(I apologize if this is a little elementary--I am self studying differential geometry and it can be a little daunting to navigate the literature, especially with the variety of presentations between intrinsic, extrinsic, local coordinates, etc.)
Edit/Update This nice summary answer on MathOverflow gives a positive answer--locally--which I had expected knowing about the implicit function theorem. But all the classic examples I can think of, e.g. $f(x,y)=x^2+y^2-1$ for the circle, $f(x,y,z) = \left(\sqrt{x^2+y^2}-R)^2\right)+z^2-r^2$, the torus, are all defined globally. (For any of these, if you want to write the manifold in the form, say, $(x,y,F(x,y))$ you have to do this locally, of course.)
Let's consider the circle again. Locally, we know we can define the function $f_\alpha(x,y) = \sqrt{1-x^2}-y$ that has the equation $f_\alpha(x,y)=0$ defining a portion of the manifold. If we were ignorant and started with this, we could easily solve for $y$, and discover the global implicit equation $x^2+y^2-1=0$. I wonder if there are any conditions, and how deep they are, to ensure this is possible in general, if at all...