TITLE: Simple curves on non-orientable surfaces.
QUESTION [18 upvotes]: Given an element in the (first) homology group of a surface, I would like to know if it can be represented as a simple closed curve.  For orientable surfaces, this is well-known, but I wasn't able to find a reference for non-orientable surfaces.  
For orientable surfaces, the sphere with $g$ handles has homology $\mathbb{Z}^{2g}$, and an element $(a_1, \dots, a_{2g}) \in \mathbb{Z}^{2g}$ can be represented by a simple closed curve if and only if $gcd(a_1, \dots, a_{2g})=1$.  This is classical and actually not too hard to prove.  
For non-orientable surfaces, the sphere with $k$ crosscaps has homology $\mathbb{Z}_2 \times \mathbb{Z}^{k-1}$. 
Let $(a, b_1, \dots, b_{k-1}) \in \mathbb{Z}_2 \times \mathbb{Z}^{k-1}$.  If $k$ is odd and $gcd(b_1, \dots, b_{k-1})=1$, then we can represent this element as a simple curve, regardless of the value of $a$ by the orientable case.  But there are other homology classes not of this form which can be represented by simple curves.  For example, I think $(0,2,0) \in \mathbb{Z}_2 \times \mathbb{Z}^2$ can be represented by a simple curve on the sphere with 3 crosscaps.  Incidentally, while I am at it, I'd also like to know why it is common to use $\mathbb{Z}_2$-homology when working with non-orientable surfaces, as opposed to $\mathbb{Z}$-homology (which is what I am using).

REPLY [6 votes]: (I found this MO thread while searching for a reference similar to the one requested by the OP, so I'm adding the following reference in case it is helpful to other future visitors)
Representing codimension one homology classes on closed nonorientable manifolds by submanifolds: William H. Meeks, III. Illinois Journal of Mathematics, Volume 23, Number 2, June 1979
In dimension two the theorem in this paper implies that an integer homology class on a connected closed nonorientable surface can be represented by an embedded circle if and only if the class is primitive or twice a primitive class. 
(A class in $H_1(M,Z)$ is called primitive if the induced class in $H_1(M, Z)/\text{Torsion}$ is the zero class or is not a nontrivial multiple of any other class.)
In a further paper (Representing homology classes by embedded circles on a compact surface) Meeks appears to be claiming that the same result also holds for nonorientable surfaces with boundary, but the referenced paper is listed as a preprint and does not appear to have been actually been published (it doesn't, for instance, appear on the list of publications in his CV)