The Stacks project

90.20 Applications

We collect some results on deformation categories we will use later.

Lemma 90.20.1. Let $f : \mathcal{H} \to \mathcal{F}$ and $g : \mathcal{G} \to \mathcal{F}$ be $1$-morphisms of deformation categories. Then

  1. $\mathcal{W} = \mathcal{H} \times _\mathcal {F} \mathcal{G}$ is a deformation category, and

  2. we have a $6$-term exact sequence of vector spaces

    \[ 0 \to \text{Inf}(\mathcal{W}) \to \text{Inf}(\mathcal{H}) \oplus \text{Inf}(\mathcal{G}) \to \text{Inf}(\mathcal{F}) \to T\mathcal{W} \to T\mathcal{H} \oplus T\mathcal{G} \to T\mathcal{F} \]

Proof. Part (1) follows from Lemma 90.16.12 and the fact that $\mathcal{W}(k)$ is the fibre product of two setoids with a unique isomorphism class over a setoid with a unique isomorphism class.

Part (2). Let $w_0 \in \mathop{\mathrm{Ob}}\nolimits (\mathcal{W}(k))$ and let $x_0, y_0, z_0$ be the image of $w_0$ in $\mathcal{F}, \mathcal{H}, \mathcal{G}$. Then $\text{Inf}(\mathcal{W}) = \text{Inf}_{w_0}(\mathcal{W})$ and similarly for $\mathcal{H}$, $\mathcal{G}$, and $\mathcal{F}$, see Remark 90.19.4. We apply Lemmas 90.12.4 and 90.19.10 to get all the linear maps except for the “boundary map” $\delta : \text{Inf}_{x_0}(\mathcal{F}) \to T\mathcal{W}$. We will insert suitable signs later.

Construction of $\delta $. Choose a pushforward $w_0 \to w'_0$ along $k \to k[\epsilon ]$. Denote $x'_0, y'_0, z'_0$ the images of $w'_0$ in $\mathcal{F}, \mathcal{H}, \mathcal{G}$. In particular we obtain isomorphisms $b' : f(y'_0) \to x'_0$ and $c' : x'_0 \to g(z'_0)$. Denote $b : f(y_0) \to x_0$ and $c : x_0 \to g(z_0)$ the pushforwards along $k[\epsilon ] \to k$. Observe that this means $w'_0 = (k[\epsilon ], y'_0, z'_0, c' \circ b')$ and $w_0 = (k, y_0, z_0, c \circ b)$ in terms of the explicit form of the fibre product of categories, see Remarks 90.5.2 (13). Given $\alpha : x'_0 \to x'_0$ we set $\delta (\alpha ) = (k[\epsilon ], y'_0, z'_0, c' \circ \alpha \circ b')$ which is indeed an object of $\mathcal{W}$ over $k[\epsilon ]$ and comes with a morphism $(k[\epsilon ], y'_0, z'_0, c' \circ \alpha \circ b') \to w_0$ over $k[\epsilon ] \to k$ as $\alpha $ pushes forward to the identity over $k$. More generally, for any $k$-vector space $V$ we can define a map

\[ \text{Lift}(\text{id}_{x_0}, k[V]) \longrightarrow \text{Lift}(w_0, k[V]) \]

using exactly the same formulae. This construction is functorial in the vector space $V$ (details omitted). Hence $\delta $ is $k$-linear by an application of Lemma 90.11.5.

Having constructed these maps it is straightforward to show the sequence is exact. Injectivity of the first map comes from the fact that $f \times g : \mathcal{W} \to \mathcal{H} \times \mathcal{G}$ is faithful. If $(\beta , \gamma ) \in \text{Inf}_{y_0}(\mathcal{H}) \oplus \text{Inf}_{z_0}(\mathcal{G})$ map to the same element of $\text{Inf}_{x_0}(\mathcal{F})$ then $(\beta , \gamma )$ defines an automorphism of $w'_0 = (k[\epsilon ], y'_0, z'_0, c' \circ b')$ whence exactness at the second spot. If $\alpha $ as above gives the trivial deformation $(k[\epsilon ], y'_0, z'_0, c' \circ \alpha \circ b')$ of $w_0$, then the isomorphism $w'_0 = (k[\epsilon ], y'_0, z'_0, c' \circ b') \to (k[\epsilon ], y'_0, z'_0, c' \circ \alpha \circ b')$ produces a pair $(\beta , \gamma )$ which is a preimage of $\alpha $. If $w = (k[\epsilon ], y, z, \phi )$ is a deformation of $w_0$ such that $y'_0 \cong y$ and $z \cong z'_0$ then the map

\[ f(y'_0) \to f(y) \xrightarrow {\phi } g(z) \to g(z'_0) \]

is an $\alpha $ which maps to $w$ under $\delta $. Finally, if $y$ and $z$ are deformations of $y_0$ and $z_0$ and there exists an isomorphism $\phi : f(y) \to g(z)$ of deformations of $f(y_0) = x_0 = g(z_0)$ then we get a preimage $w = (k[\epsilon ], y, z, \phi )$ of $(x, y)$ in $T\mathcal{W}$. This finishes the proof. $\square$

Lemma 90.20.2. Let $\mathcal{H}_1 \to \mathcal{G}$, $\mathcal{H}_2 \to \mathcal{G}$, and $\mathcal{G} \to \mathcal{F}$ be maps of categories cofibred in groupoids over $\mathcal{C}_\Lambda $. Assume

  1. $\mathcal{F}$ and $\mathcal{G}$ are deformation categories,

  2. $T\mathcal{G} \to T\mathcal{F}$ is injective, and

  3. $\text{Inf}(\mathcal{G}) \to \text{Inf}(\mathcal{F})$ is surjective.

Then $\mathcal{H}_1 \times _\mathcal {G} \mathcal{H}_2 \to \mathcal{H}_1 \times _\mathcal {F} \mathcal{H}_2$ is smooth.

Proof. Denote $p_ i : \mathcal{H}_ i \to \mathcal{G}$ and $q : \mathcal{G} \to \mathcal{F}$ be the given maps. Let $A' \to A$ be a small extension in $\mathcal{C}_\Lambda $. An object of $\mathcal{H}_1 \times _\mathcal {F} \mathcal{H}_2$ over $A'$ is a triple $(x'_1, x'_2, a')$ where $x'_ i$ is an object of $\mathcal{H}_ i$ over $A'$ and $a' : q(p_1(x'_1)) \to q(p_2(x'_2))$ is a morphism of the fibre category of $\mathcal{F}$ over $A'$. By pushforward along $A' \to A$ we get $(x_1, x_2, a)$. Lifting this to an object of $\mathcal{H}_1 \times _\mathcal {G} \mathcal{H}_2$ over $A$ means finding a morphism $b : p_1(x_1) \to p_2(x_2)$ over $A$ with $q(b) = a$. Thus we have to show that we can lift $b$ to a morphism $b' : p_1(x'_1) \to p_2(x'_2)$ whose image under $q$ is $a'$.

Observe that we can think of

\[ p_1(x'_1) \to p_1(x_1) \xrightarrow {b} p_2(x_2) \quad \text{and}\quad p_2(x'_2) \to p_2(x_2) \]

as two objects of $\textit{Lift}(p_2(x_2), A' \to A)$. The functor $q$ sends these objects to the two objects

\[ q(p_1(x'_1)) \to q(p_1(x_1)) \xrightarrow {b} q(p_2(x_2)) \quad \text{and}\quad q(p_2(x'_2)) \to q(p_2(x_2)) \]

of $\textit{Lift}(q(p_2(x_2)), A' \to A)$ which are isomorphic using the map $a' : q(p_1(x'_1)) \to q(p_2(x'_2))$. On the other hand, the functor

\[ q : \textit{Lift}(p_2(x_2), A' \to A) \to \textit{Lift}(q(p_2(x_2)), A' \to A) \]

defines a injection on isomorphism classes by Lemma 90.17.5 and our assumption on tangent spaces. Thus we see that there is a morphism $b' : p_1(x_1') \to p_2(x'_2)$ whose pushforward to $A$ is $b$. However, we may need to adjust our choice of $b'$ to achieve $q(b') = a'$. For this it suffices to see that $q : \text{Inf}(p_2(x'_2)/p_2(x_2)) \to \text{Inf}(q(p_2(x'_2))/q(p_2(x_2)))$ is surjective. This follows from our assumption on infinitesimal automorphisms and Lemma 90.19.11. $\square$

Lemma 90.20.3. Let $f : \mathcal{F} \to \mathcal{G}$ be a map of deformation categories. Let $x_0 \in \mathop{\mathrm{Ob}}\nolimits (\mathcal{F}(k))$ with image $y_0 \in \mathop{\mathrm{Ob}}\nolimits (\mathcal{G}(k))$. If

  1. the map $T\mathcal{F} \to T\mathcal{G}$ is surjective, and

  2. for every small extension $A' \to A$ in $\mathcal{C}_\Lambda $ and $x \in \mathcal{F}(A)$ with image $y \in \mathcal{G}(A)$ if there is a lift of $y$ to $A'$, then there is a lift of $x$ to $A'$,

then $\mathcal{F} \to \mathcal{G}$ is smooth (and vice versa).

Proof. Let $A' \to A$ be a small extension. Let $x \in \mathcal{F}(A)$. Let $y' \to f(x)$ be a morphism in $\mathcal{G}$ over $A' \to A$. Consider the functor $\text{Lift}(A', x) \to \text{Lift}(A', f(x))$ induced by $f$. We have to show that there exists an object $x' \to x$ of $\text{Lift}(A', x)$ mapping to $y' \to f(x)$, see Lemma 90.8.2. By condition (2) we know that $\text{Lift}(A', x)$ is not the empty category. By condition (2) and Lemma 90.17.5 we conlude that the map on isomorphism classes is surjective as desired. $\square$

Lemma 90.20.4. Let $\mathcal{F} \to \mathcal{G} \to \mathcal{H}$ be maps of categories cofibred in groupoids over $\mathcal{C}_\Lambda $. If

  1. $\mathcal{F}$, $\mathcal{G}$ are deformation categories

  2. the map $T\mathcal{F} \to T\mathcal{G}$ is surjective, and

  3. $\mathcal{F} \to \mathcal{H}$ is smooth.

Then $\mathcal{F} \to \mathcal{G}$ is smooth.

Proof. Let $A' \to A$ be a small extension in $\mathcal{C}_\Lambda $ and let $x \in \mathcal{F}(A)$ with image $y \in \mathcal{G}(A)$. Assume there is a lift $y' \in \mathcal{G}(A')$. According to Lemma 90.20.3 all we have to do is check that $x$ has a lift too. Take the image $z' \in \mathcal{H}(A')$ of $y'$. Since $\mathcal{F} \to \mathcal{H}$ is smooth, there is an $x' \in \mathcal{F}(A')$ mapping to both $x \in \mathcal{F}(A)$ and $z' \in \mathcal{H}(A')$, see Definition 90.8.1. This finishes the proof. $\square$


Comments (0)


Post a comment

Your email address will not be published. Required fields are marked.

In your comment you can use Markdown and LaTeX style mathematics (enclose it like $\pi$). A preview option is available if you wish to see how it works out (just click on the eye in the toolbar).

Unfortunately JavaScript is disabled in your browser, so the comment preview function will not work.

All contributions are licensed under the GNU Free Documentation License.




In order to prevent bots from posting comments, we would like you to prove that you are human. You can do this by filling in the name of the current tag in the following input field. As a reminder, this is tag 0DYM. Beware of the difference between the letter 'O' and the digit '0'.