MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $Y\subset X$ be a codimension $k$ proper inclusion of submanifolds. If we choose a coorientation of $Y$ inside of $X$ (that is, an orientation of the normal bundle), then we get a class $[Y]\in H^k(X)$. If $X$ and $Y$ are oriented, then $[Y]$ may be defined as the fundamental class of $Y$ in the Borel-Moore homology of $X$, which is isomorphic to the cohomology of $X$. What is the simplest definition of $[Y]$ in the general case (where $X$ and $Y$ are not necessarily oriented)?

Note that a simple generalization of this question would be to ask how to define the pushforward in cohomology along a proper oriented map. (Then $[Y]$ would simply be the pushforward of $1\in H^0(Y)$.) I would be happy to know the answer to this more general question, but I asked the simpler version to be as concrete as possible.

share|cite|improve this question
up vote 7 down vote accepted

The easiest definition is via the Pontrjagin--Thom construction (I think). Let $N$ be a tubular neighbourhood of $Y$, isomorphic to the normal bundle. Let $X'$ be the space obtained from $X$ by collapsing the complement of $N$ to a point. Then $X'$ is isomorphic to the Thom space of the normal bundle, and if the normal bundle is oriented then there is the Thom isomorphism: $$ H^\bullet(Y,\mathbf Z) \cong H^{\bullet+\dim X - \dim Y}(X',\mathbf Z).$$ Composing with the natural $H^\bullet(X',\mathbf Z) \to H^\bullet(X, \mathbf Z)$ gives the map you want.

share|cite|improve this answer

Your construction for the oriented case extends to the unoriented case just by using twisted coefficients.

Every manifold $Y$ has a fundamental class $[Y]\in H_\ast^{\mathrm{lf}}(Y;\mathfrak o_Y)$ (where $\mathfrak o_Y$ is the orientation sheaf of $Y$). The coorientation of $Y\subseteq X$ gives an identification of orientation sheaves $\mathfrak o_Y\xrightarrow\sim\mathfrak o_X$ over $Y$, and thus there is a pushforward map: $$H_\ast^{\mathrm{lf}}(Y;\mathfrak o_Y)\to H_\ast^{\mathrm{lf}}(Y;\mathfrak o_X)\to H_\ast^{\mathrm{lf}}(X;\mathfrak o_X)$$ Now use the Poincare duality isomorphism $H^\ast(X;\mathbb Z)\xrightarrow{\cap[X]}H_\ast^{\mathrm{lf}}(X;\mathfrak o_X)$.

share|cite|improve this answer
Thanks, that's a nice perspective! – Nicholas Proudfoot Dec 11 '13 at 16:39

This is an answer to your more general question about how to define $$f_\ast:H^\ast(X;\mathbb{Z})\to H^{\ast+\dim(Y)-\dim(X)}(Y;\mathbb{Z})$$ when $f: X\to Y$ is a proper oriented map.

For $\ell$ sufficiently large, there is an embedding $g: X\hookrightarrow \mathbb{R}^\ell$ which is unique up to isotopy. The map $f':X\to Y\times \mathbb{R}^\ell$ given by $f'(x)=(f(x),g(x))$ is then an embedding, and has a tubular neighbourhood $N$ as in Dan's answer. Note that the normal bundle $\nu_{f'}$ of $f'$ has dimension $\dim(Y)-\dim(X)+\ell$. There is a Pontrjagin-Thom collapse map $F:(Y\times\mathbb{R}^\ell)_+\to T(\nu_f')$, where $+$ denotes a one-point compactification. After making a standard identification, this gives a pointed map $F: \Sigma^\ell Y_+\to T(\nu_{f'})$, where $\Sigma^\ell$ denotes the $\ell$-fold reduced suspension.

Now we can define the pushforward as the composition $$ H^\ast(X)\to \widetilde{H}^{\ast+\dim(Y)-\dim(X)+\ell}(T(\nu_{f'}))\stackrel{F^\ast}{\to}\widetilde{H}^{\ast+\dim(Y)-\dim(X)+\ell}(\Sigma^\ell Y_+) \to H^{\ast+\dim(Y)-\dim(X)}(Y). $$ Here the first map is a Thom isomorphism and the last is a suspension isomorphism, and all coefficients are integers.

share|cite|improve this answer

Here is an extrinsic definition of the class of $Y$. The kernel of the natural map $$ H^*(X) \to H^*(X \setminus Y) $$ is a graded ideal in $H^*(X)$. The lowest degree that this ideal is non-zero in is precisely $codim(Y)$ and in this degree the image is a free $\mathbb{Z}$-module on one generator, which is $\pm[Y]$. This works if, e.g., $X \to Y$ is a proper embedding of smooth manifolds. I learned this in a paper of Feher and Rimanyi (See Definition 2.6 and the examples in 2.7).

To get the correct choice of a sign, at least for varieties, I would compute the degree (as in pushforward to a point) of the associated classes. Only one of these will be positive.

share|cite|improve this answer
Some typos: the map $H^*(X) \to H^*(X \setminus Y)$ is not surjective, and the ideal you want to look at is the kernel of that map. Nice, but it gives you the class of $Y$ up to sign; is there a way to get the right sign? – abx Dec 11 '13 at 5:53
thanks for catching this. in the $K$-theory of varieties $K^0(X) \to K^0(X-Y)$ is surjective (and the kernel is the image of $K^0(Y)$), whence my confusion. – andrewBee Dec 11 '13 at 6:46
That's really cool; thanks! – Nicholas Proudfoot Dec 11 '13 at 16:41

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.