Does the Associated Graded Functor have an Adjoint

$F$ preserves neither limits nor colimits and so does not have an adjoint on either side.

First, $F$ does not preserve finite coproducts. For instance, the coproduct of two copies of $(\mathbb{Z},2\mathbb{Z})$ in $C$ is again just $(\mathbb{Z},2\mathbb{Z})$. But $F(\mathbb{Z},2\mathbb{Z})$ is the graded ring $\mathbb{F}_2[t]$ (with $|t|=1$), and the coproduct of two copies of this graded ring is $\mathbb{F}_2[t,t']$ (with $|t|=|t'|=1$) which is not isomorphic to it.

Second, $F$ does not preserve equalizers. For instance, consider the two morphisms $f,g:(\mathbb{Z}[t],0)\to(\mathbb{Z}[x,y],\mathbb{Z}[x,y])$ which map $t$ to $x$ and $y$, respectively. Their equalizer is $(\mathbb{Z},0)$, which $F$ sends to the graded ring $\mathbb{Z}$ concentrated in degree $0$. But $F(\mathbb{Z}[x,y],\mathbb{Z}[x,y])$ is the zero graded ring, and so $F(f)=F(g)$ and the equalizer of $F(f)$ and $F(g)$ is $F(\mathbb{Z}[t],0)$ which is $\mathbb{Z}[t]$ concentrated in degree $0$.


This is a riff on the more general question "what is the universal property of associated graded?" I won't discuss the rings-and-ideals case at all, and will just restrict my attention to filtered vector spaces.

Classically a filtered vector space is an increasing sequence $V_0 \subseteq V_1 \subseteq V_2 \subseteq \dots $ and it has an associated graded vector space given in degree $i$ by $V_i/V_{i-1}$. One of the first MO questions (#263!) asked whether this functor is a left or a right adjoint, and the answer is that it is neither.

This turns out, however, to be an artifact of working with an insufficiently flexible notion of filtration; it turns out that requiring all the maps $V_i \to V_{i+1}$ to be inclusions is too restrictive and makes the category behave poorly, and we can fix this by working with a generalized notion of filtration given just by a sequence $V_0 \to V_1 \to V_2 \to \dots$ of not-necessarily-injective maps. With this fix, as described by Nicholas Schmidt, taking associated graded is left adjoint to the functor which sends a graded vector space $W_i$ to the sequence $W_0 \xrightarrow{0} W_1 \xrightarrow{0} W_2 \dots$ where all maps are zero, and in particular it now preserves colimits (which behave differently and better in this new category of filtered vector spaces).


There is a lovely algebro-geometric interpretation of a further generalization of this construction where we allow $\mathbb{Z}$-gradings, as follows. The category of $\mathbb{Z}$-graded vector spaces can be identified with the category of quasicoherent sheaves over the stack $B \mathbb{G}_m$ which classifies line bundles; this is a fancy way of saying that an action of $\mathbb{G}_m$ is the same thing as a $\mathbb{Z}$-grading. The category of "$\mathbb{Z}$-filtered" vector spaces, by which I mean sequences $\dots V_i \to V_{i+1} \dots$ possibly extending infinitely in both directions, can in turn be identified with the category of quasicoherent sheaves over the quotient stack $\mathbb{A}^1/\mathbb{G}_m$. There is a natural inclusion

$$i : B \mathbb{G}_m \to \mathbb{A}^1/\mathbb{G}_m$$

given by thinking of $B \mathbb{G}_m$ as $\bullet / \mathbb{G}_m$ and including $\bullet$ into $\mathbb{A}^1$ as the origin, and the associated graded functor turns out to be the pullback $i^{\ast}$ of quasicoherent sheaves along $i$, left adjoint to pushforward $i_{\ast}$.