To which extent is a thin category determined by its graph?

69 Views Asked by At

I'm reading Abstract and Concrete Categories and am currently attempting to solve the very first exercise, which asks to show that a thin category is determined by its graph up to isomorphism.

Here, a category $\mathbf{A}$ is a quadruple $(\mathcal{O},\mathrm{hom},\mathrm{id},\circ)$, such that

  • $\mathcal{O}$ is a class of objects, alternatively denoted $\mathrm{Ob}(\mathbf{A})$,
  • $\mathrm{hom}\colon\mathcal{O}^2\rightarrow\mathcal{U}$ is a mapping, where each element of its image is called a hom-set (here, $\mathcal{U}$ is the universe, i.e. the class of all sets),
  • $\mathrm{id}\colon\mathcal{O}\rightarrow\mathrm{Mor}(\mathbf{A})\colon=\bigcup_{X\in\mathrm{hom}(\mathcal{O}^2)}X$ is a mapping that sends an $\mathbf{A}$-object $A$ to an identity morphism $\mathrm{id}_A\in\mathrm{hom}(A,A)$ and
  • $\circ\colon\mathrm{Mor}(\mathbf{A})\times\mathrm{Mor}(\mathbf{A})\leadsto\mathrm{Mor}(\mathbf{A})$ is a partial mapping that is defined on $(g,f)$ iff $f\in\mathrm{hom}(A,B)$ and $g\in\hom(B,C)$ for $A,B,C\in\mathrm{Ob}(\mathbf{A})$; in that case, $g\circ f\colon=\circ(g,f)\in\mathrm{hom}(A,C)$.

These are subject to the conditions that

  • composition is associative, i.e. $(h\circ g)\circ f=h\circ(g\circ f)$ whenever defined,
  • for any morphism $f\in\mathrm{hom}(A,B)$, we have $\mathrm{id}_B\circ f=f=f\circ\mathrm{id}_A$ and
  • hom-sets are disjoint.

A morphism $f\in\mathrm{Mor}(\mathbf{A})$ is by definition a member of a hom-set $\mathrm{hom}(A,B)$ with $A,B\in\mathrm{Ob}(\mathbf{A})$ and, by disjointedness of hom-sets, it is the member of exactly one hom-set. Thus, setting $\mathrm{dom}(f)=A$ and $\mathrm{cod}(f)=B$ produces two well-defined mappings $\mathrm{dom},\mathrm{cod}\colon\mathrm{Mor}(\mathbf{A})\rightarrow\mathrm{Ob}(\mathbf{A})$, giving what is called domain and codomain of a morphism respectively.

Furthermore, a large graph is defined as a quadruple $(V,E,d,c)$, where $V$ and $E$ are classes (called the class of vertices and edges respectively), and $d\colon E\rightarrow C$ and $c\colon E\rightarrow C$ are mappings giving what is called domain or codomain of each edge respectively. The graph $G(\mathbf{A})$ of a category $\mathbf{A}$ is the graph with $V=\mathrm{Ob}(\mathbf{A})$, $E=\mathrm{Mor}(\mathbf{A})$, $d=\mathrm{dom}$ and $c=\mathrm{cod}$.

A thin category is a category where any hom-set contains at most one element. Let $\mathbf{A}=(\mathcal{O},\mathrm{hom},\mathrm{id},\circ)$ and $\mathbf{A}^\prime=(\mathcal{O}^\prime,\mathrm{hom}^\prime,\mathrm{id}^\prime,\circ^\prime)$ be two thin categories such that $G(\mathbf{A})=G(\mathbf{A}^\prime)$. I.e. $$(\mathrm{Ob}(\mathbf{A}),\mathrm{Mor}(\mathbf{A}),\mathrm{dom},\mathrm{cod})=G(\mathbf{A})=G(\mathbf{A}^\prime)=(\mathrm{Ob}(\mathbf{A}^\prime),\mathrm{Mor}^\prime(\mathbf{A}),\mathrm{dom}^\prime,\mathrm {cod}^\prime),$$ where the $^\prime$ denotes being relative to the category $\mathbf{A}^\prime$. From this, it follows immediately by definition of a tuple that $\mathcal{O}=\mathrm{Ob}(\mathbf{A})=\mathrm{Ob}(\mathbf{A}^\prime)=\mathcal{O}^\prime$, $\mathrm{Mor}(\mathbf{A})=\mathrm{Mor}(\mathbf{A}^\prime)$, $\mathrm{dom}=\mathrm{dom}^\prime$ and $\mathrm{cod}=\mathrm{cod}^\prime$. Let $A,B\in\mathcal{O}=\mathcal{O}^\prime$. We then have by definition that $$\mathrm{hom}(A,B)=\mathrm{dom}^{-1}(\left\{A\right\})\cap\mathrm{cod}^{-1}(\left\{B\right\})=(\mathrm{dom}^\prime)^{-1}(\left\{A\right\})\cap(\mathrm{cod}^\prime)^{-1}(\left\{B\right\})=\mathrm{hom}^\prime(A,B).$$ Thus, $\mathrm{hom}=\mathrm{hom}^\prime$. Now let $A,B,C\in\mathcal{O}$, $f\in\mathrm{hom}(A,B)$ and $g\in\mathrm{hom}(B,C)$ be arbitrary. Then, $g\circ f\in\mathrm{hom}(A,C)$ exists, i.e. the hom-set $\mathrm{hom}(A,C)$ has at least on element, but, since $\mathbf{A}$ is thin, it contains at most one element, hence it has exactly one element. Since $\mathrm{Ob}(\mathbf{A})=\mathrm{Ob}(\mathbf{A}^\prime)$ and $\mathrm{Mor}(\mathbf{A})=\mathrm{Mor}(\mathbf{A}^\prime)$, we may regard $f,g$ as morphisms in $\mathbf{A}^\prime$ and thus $g\circ^\prime f$ exists. By nature of composition, we have $g\circ^\prime f\in\mathrm{hom}(A,C)$, but since this set has exactly one element, it follows that $g\circ f=g\circ^\prime f$ and, since $f,g$ were arbitrary, it follows that $\circ=\circ^\prime$.

Finally, let $A,B\in\mathrm{Ob}(\mathbf{A})$ and $f\in\mathrm{hom}(A,B)$ be arbitrary. By the previous result, it holds that $\mathrm{id}_B\circ^\prime f=\mathrm{id}_B\circ f=f$ and $f\circ^\prime\mathrm{id}_A=f\circ\mathrm{id}_A=f$. Hence, the identity morphisms in $\mathbf{A}$ are also identity morphisms in $\mathbf{A}^\prime$ and, by uniqueness of identities, it follows that $\mathrm{id}=\mathrm{id}^\prime$.

Finally, by the definition of a tuple, it follows that $\mathbf{A}=\mathbf{A}^\prime$.

Now, this doesn't only imply that a graph determines a thin category up to isomorphism (as was required to be shown), but it shows that a thin category is uniquely determined by its graph. This claim is clearly stronger than the initial one and since the proposed proof is pretty trivial, I suspect there must be an error as otherwise the way the exercise was stated seems off. My question now is where any error or potential misunderstanding lies.