# What's the normal cone to a union?

I have a reducible variety, $X = \bigcup_{i \in I} X_i \subset Y$, and I want to understand the normal cone $C_X Y$.

I write $X_J$ for $\cap_{j \in J} X_j$.

Can $C_X Y$ be expressed in terms of the $C_{X_J} X_K$?

(I take the convention $X_\emptyset = Y$ so that the $C_{X_i} Y$ are among the above.)

In fact in the example that has prompted this question, all the $X_J$ are smooth and connected, (though I do not know that they intersect transversely), and I am actually only interested in numerical information; i.e.,

Can the Segre class $S(X,Y)$ be expressed in terms of the Segre classes $S(X_J, X_K)$ for $J \subset K$ and perhaps also the intersection multiplicities of $X_M$ and $X_N$ along their intersection?

-

No. Take $X$ to be the union of two smooth, genus $0$ curves, say $X_1$ and $X_2$, that intersect in two nodes, say $p$ and $q$. Then $\text{Pic}(X)$ is an extension of $\mathbb{Z}^2$ by $\mathbb{G}_m$. So let $Y$ be a geometric $\mathbb{A}^1$-bundle over $X$ whose restriction to each of $X_1$ and $X_2$ is trivial, yet that is nontrivial on $X$. Identify $X$ with the zero section of this $\mathbb{A}^1$-bundle. The normal cone is a normal bundle, essentially just the data of $Y$ once again. Yet you cannot recover $Y$ from its restriction to the two irreducible components.
Edit. Vivek clarified that he would also like to keep track of the restriction maps between the cones on different strata. That would certainly make a difference in the example above. However, if one allows non-transversal intersections, as Vivek explicitly allows, then there are still counterexamples. For instance, consider a union of three concurrent lines in $\mathbb{P}^2$. The irreducible components are each $\mathbb{P}^1$, and the intersections between components are all one (reduced) point. Now there is simply no way to encode all invertible sheaves by the restriction on strata and restriction homomorphisms between them. (Of course if one allows nilpotent thickenings of the strata, one could encode these invertible sheaves.)
I also allowed in the question to use things like $C_{X_1 \cap X_2}(Y)$ and $C_{X_1 \cap X_2} (X_1)$ and asked it in a rather vague way which might be interpreted as allowing you to use some induced maps between these cones. Does the counterexample exclude that as well? – Vivek Shende Jul 9 '13 at 10:53