Is it possible to embed de Rham cohomology of a twodimensional closed surface of genus $g\geq 2$ into the differential graded algebra of differential forms (with de Rham differential and wedge product) on the surface as a differential graded subalgebra (in a way that is compatible with canonical projection from closed forms to cohomology, associating to a closed form its cohomology class)?

1$\begingroup$ Does this not follow from the formality of compact Kähler manifolds, proved in Deligne, Pierre; Griffiths, Phillip A.; Morgan, John W.; Sullivan, Dennis (1975), "Real homotopy theory of Kähler manifolds", Inventiones Mathematicae 29 (3): 245–274, doi:10.1007/BF01389853, MR0382702 ? $\endgroup$ – José FigueroaO'Farrill Oct 3 '11 at 23:50

$\begingroup$ As far as I understand, formality just means that there exists some A_\infty morphism from cohomology to de Rham algebra. Generally it would have polylinear components. My question was whether one can find an A_\infty morphism which has only linear component. $\endgroup$ – pmnev Oct 4 '11 at 0:08

$\begingroup$ Jose  this just shows that Sullivan's minimal model maps quasiisomorphically to both algebras. $\endgroup$ – algori Oct 4 '11 at 0:20

$\begingroup$ pmnev  I think you are right. Moreover, I think one can get a genuine morphism of algebras at the expense of enlarging the target: namely, if one replaces the de Rham algebra with the cobar of its bar. $\endgroup$ – algori Oct 4 '11 at 0:24

$\begingroup$ @pmnev,algori: Thanks for your comments. I understand now what the question actually asks. $\endgroup$ – José FigueroaO'Farrill Oct 4 '11 at 0:47
The answer is no. Suppose that $\alpha_1,\ldots,\alpha_g,\beta_1,\ldots,\beta_g$ were closed $1$forms on $M$ such that their cohomology classes were a basis of $H^1(M)$ and they satisfied $\alpha_i\wedge\alpha_j = \beta_i\wedge\beta_j = 0$ while $\alpha_i\wedge\beta_j = \delta_{ij}\ \gamma$ where $\gamma$ is a single $2$form whose cohomology class is nonzero. Then $\gamma$ cannot vanish identically.
Let $U\subset M$ be the open set on which $\gamma$ is nonzero. Then none of the $\alpha_i$ or $\beta_i$ can vanish on $U$. Since $\alpha_1\wedge\alpha_2 = \beta_1\wedge\alpha_2=0$ while $\alpha_2\not=0$ on $U$, it follows that $\alpha_1$ and $\beta_1$ are multiples of $\alpha_2$ on $U$. But this implies that $\alpha_1$ and $\beta_1$ must be linearly dependent on $U$ as well, which implies that $\alpha_1\wedge\beta_1 = 0$.