Reading Bachman's "A Geometric Approach to Differential Forms", in section 7.8.1 about the Lining Number invariant, I have stumbled upon the following assertion.
Let the knot $K$ be defined as a (continuous and differentiable) map from unit circle $S^1$ to 3d Eucledian space $\mathbb{R}^3$, so:
$K: S^1 \to \mathbb{R}^3$
i.e. $K$ is a closed curve in 3d space parametrized by a single real variable.
Let $U=\mathbb{R}^3-K$, i.e. it is the 3d space without the points associated with the knot (curve).
Then there always exists a map: $A:U\to S^1$.
Unfortunatelly author then says that the proof is outside the scope of the book, and I am not sufficiently knowledgeable to know where to find the proof. Can anyone please suggest a good reference for a keen amateur?
Paragraph from the book:
"""
The linking number of a two-component link is precisely the analogous measure that you get when you treat one of the components as the z-axis and the other as the 1-chain over which we integrate. Given any knot $K$ in $\mathbb{R}^3$, there is a function $A: \mathbb{R}^3 − K \to S^1$. (The existence of such a function is, unfortunately, beyond the scope of this book.) If we think of A as a 0-form, then we can differentiate it to get a 1-form on $\mathbb{R}^3 − K$. This 1-form is precisely what we can integrate over a second knot to measure how many times it “links” with $K$. Interestingly, there is also a point $p \in S^1$ such that $A^{-1}(p)$ is a surface whose boundary is $K$. Just as before, the linking number can also be computed just by appropriately counting the number of intersections with this surface.
"""
The standard reason is homological (Hatcher's book is the usual reference). Alexander duality relates the homology of $K$ to the cohomology of $\mathbb{R}^3-K$. In particular, $H^1(\mathbb{R}^3-K;\mathbb{Z})$ is isomorphic to $H_1(K;\mathbb{Z})$ is isomorphic to $\mathbb{Z}$. First cohomology with $\mathbb{Z}$ coefficients is the same as homotopy classes of maps $\mathbb{R}^3-K \to S^1$. If you take the generator $1\in\mathbb{Z}$ and run through the isomorphisms, you get a map $A:\mathbb{R}^3-K\to S^1$, which we can assume is smooth. Supposing $S^1$ is parameterized by $\theta$, then $d\theta$ is its $1$-form. The map $A^*$ is the pullback for $1$-forms, giving a closed $1$-form $A^*d\theta$ on $\mathbb{R}^3-K$. (I'm a little uncomfortable calling $dA$ a $1$-form, but it seems $dA=A^*d\theta$.)
Given an oriented curve $C\subset\mathbb{R}^3-K$, the linking number of $C$ with $K$ is $$\int_C A^*d\theta.$$
Unfortunately, I don't know of an elementary way to see such an $A$ must exist! (Perhaps you could integrate the form in the Gauss linking integral that N. Owad mentions in the comments. The idea would be you fix one knot $K$ permanently, then vary the second knot. Fix a point $x_0\in \mathbb{R}^3-K$ and define $A(x)$ to be the integral along any arc from $x_0$ to $x$ in $\mathbb{R}^3-K$. I haven't checked, but the $2$-form ought to be closed, so $A(x)$ is well-defined modulo $1$, hence $A$ can be thought of as a map to a circle.)
A possibly less-mysterious way to get an $A$ is through a Seifert surface. There is a triangulation of $\mathbb{R}^3-K$ where the Seifert surface is a subcomplex, and then you can define a function to $S^1$ by choosing where all the edges in the triangulation go---paths from one side of the Seifert surface to the other should go around $S^1$ exactly once---and then there is a way to fill in the triangles and tetrahedra. After this, you have to find a smooth approximation, which is easy in theory.