5
$\begingroup$

Let $\textbf{Vect}_\infty$ be the category of infinite dimensional vector spaces (over some fixed ground field), and consider the endofunctor $F$ of $\textbf{Vect}_\infty$ whose effect on objects is given by $F(V):=V^{**}/V$, and whose effect on morphisms is the obvious one. (Here $V^{**}/V$ is the cokernel of the canonical monomorphism $V\to V^{**}$.)

Does $F$ reflect monomorphisms?

Does it reflect epimorphisms?

Does it reflect isomorphisms?

(Recall the $F$ reflect monomorphisms if the condition that $F(f)$ is a monomorphism implies that $f$ is also a monomorphism. The reflections of epimorphisms and isomorphisms are defined similarly.)

$\endgroup$
  • $\begingroup$ If $V$ is an infinite dimensional vector space, if $K$ is the ground field, and if $f$ is the endomorphism $(\lambda,v)\mapsto(0,v)$ of $K\times V$, then $F(f)$ is (isomorphic to) the identity of $F(V)$. This shows that the answer to the three questions is No. $\endgroup$ – Pierre-Yves Gaillard Oct 30 '18 at 9:31
5
$\begingroup$

The endofunctor $F:\mathbf{Vect}_\infty\rightsquigarrow\mathbf{Vect}_\infty$ does not reflect monomorphisms. Pick $V$ to be the vector space of sequences $(x_1,x_2,x_3,\ldots)$ with finitely many non-zero terms. Choose $f:V\to V$ to be the left shift operator $$(x_1,x_2,x_3,\ldots)\mapsto (x_2,x_3,x_4\ldots).$$
Clearly, $f$ is not monic. We claim that $F(f)$ is monic.

The kernel of $F(f)$ consists of $\alpha +V$ with $\alpha \in V^{**}$ satisfying $f^{**}\alpha \in V$. Because $f$ is an epimorphism, $f^{**}\alpha=f(v)$ for some $v\in V$. Now, for an arbitrary $\varphi\in V^*$, we have $$\alpha(\varphi\circ f)=f^{**}\alpha(\varphi)=f(v)(\varphi)=\varphi\big(f(v)\big)=(\varphi\circ f)(v)=v(\varphi\circ f).$$ Therefore, $$(\alpha-v)(f^*\varphi)=(\alpha-v)(\varphi\circ f)=0$$ for all $\varphi\in V^*$.

For convenience, let $e_j\in V$ denote the sequence $(0,0,0,\ldots,0,1,0,0,0,\ldots)$, where there is only one $1$ at the $j$th term, and other terms are $0$. So, $V=\bigoplus_{j=1}^\infty Ke_j$ if $K$ is the base field, and $V^*$ can be identified with $\prod_{j=1}^\infty Ke_j$. Hence, we can identify $V$ as a subspace of $V^*$ as well.

Observe that $V^*=Ke_1\oplus \operatorname{im}f^*$. That is, if $\lambda=(\alpha-v)(e_1)$, then $$\alpha=v+\big(\lambda-v_1\big)e_1\in V,$$ where $v=(v_1,v_2,v_3,\ldots)$. That is, $F(f)$ is monic, but $f$ is not.


Note that $F(f)$ in the example above is actually an isomorphism. To show that $F(f)$ is epic, we observe that for $\beta\in V^{**}$, there exists $\alpha \in V^{**}$ such that $\beta=f^{**}\alpha$. We define $$\alpha(\varphi_1,\varphi_2,\varphi_3,\ldots)=\beta(\varphi_2,\varphi_3,\varphi_4,\ldots)$$ for all $\varphi=(\varphi_1,\varphi_2,\varphi_3,\ldots)\in V^*$. That is, $$f^{**}\alpha(\varphi_1,\varphi_2,\varphi_3,\ldots)=\alpha(0,\varphi_1,\varphi_2,\varphi_3,\ldots)=\beta(\varphi_1,\varphi_2,\varphi_3,\ldots)$$ for all $\varphi\in V^{*}$, so $f^{**}\alpha=\beta$. This proves that $F(f)$ is epic. Since it is monic, $F(f)$ is an isomorphism, but as we learned, $f$ is not an isomorphism. Therefore, $F$ does not reflect isomorphisms.


The endofunctor $F$ does not reflect epimorphisms. Pick $V$ to be the same as above, but now let $f$ be the right shift operator $$(x_1,x_2,x_3,\ldots)\mapsto (0,x_1,x_2,x_3,\ldots).$$ Clearly, $f$ is not epic. We claim that $F(f)$ is epic.

Let $\beta \in V^{**}$. We want to find $\alpha \in V^{**}$ such that $\beta-f^{**}\alpha\in V$. With the identification $V^*=\prod_{j=1}^\infty Ke_j$ as before, we take $$\alpha(\varphi_1,\varphi_2,\varphi_3,\ldots)=\beta(0,\varphi_1,\varphi_2,\varphi_3,\ldots)$$ for all $\varphi=(\varphi_1,\varphi_2,\varphi_3,\ldots)\in V^*$ with $\varphi_j=\varphi(e_j)$. Therefore, for all $\varphi\in V^*$, $$\big(\beta-f^{**}\alpha\big)(\varphi)=\beta(\varphi_1,0,0,0,\ldots)=\varphi_1\beta(e_1)=\beta(e_1)\ e_1(\varphi).$$ That is, $$\beta-f^{**}\alpha =\beta(e_1)\ e_1\in V.$$ Consequently, $\beta+V=F(f)(\alpha +V)$, and so $F(f)$ is epic without $f$ being epic.


However, this is true. If $f:V\to W$ is such that $F(f)$ is monic, then $W\cap \big(\operatorname{im}f^{**}\big)\subseteq \operatorname{im} f$. Conversely, if $f:V\to W$ is such that $W\cap \big(\operatorname{im}f^{**}\big)\subseteq \operatorname{im} f$ with the extra condition that $f$ is monic, then $F(f)$ is monic.

To show this, note that the kernel of $F(f)$ consists of $\alpha+V$ with $\alpha \in V^{**}$ satisfying $f^{**}\alpha \in W$. Because $F(f)$ is monic, $\alpha=v$ for some $v\in V$. As $f^{**}\alpha=w$ for some $w\in W\cap\big(\operatorname{im}f^{**}\big)$, $$\varphi\big(f(v)\big)=(\varphi\circ f)(v)=v(\varphi\circ f)=\alpha(\varphi \circ f)=f^{**}\alpha(\varphi)=w(\varphi)=\varphi(w)$$ for all $\varphi \in W^*$. That is, $$\varphi\big(w-f(v)\big)=0$$ for every $\varphi\in W^*$. This shows that $w=f(v)$, and so $w\in\operatorname{im}f$.

The converse can be proven as follows. Let $\alpha\in V^{**}$ be such that $\alpha+V\in\ker F(f)$. Then, $f^{**}\alpha \in W\cap \big(\operatorname{im} f^{**}\big)\subseteq \operatorname{im}f$, so $f^{**}\alpha=f(v)$ for some $v\in V$. Thus, $$(\alpha-v)(\varphi\circ f)=f^{**}\alpha(\varphi)-f(v)(\varphi)=0$$ for all $\varphi\in W^*$, so the injectivity of $f$ implies surjectivity of $f^*:W^*\to V^*$, which means $\alpha=v$.

$\endgroup$
  • 1
    $\begingroup$ Thank you very much for this magnificent answer!!! However I don't understand the third point. It seems to me $F(f)$ is injective iff $f^{**}(\alpha)\in W$ implies $\alpha\in V$. I don't see why your $w$ is an arbitrary vector of $W$. $\endgroup$ – Pierre-Yves Gaillard Oct 29 '18 at 14:47
  • 1
    $\begingroup$ Ach! You are very right. This part will be modified. $\endgroup$ – user593746 Oct 29 '18 at 15:03
  • $\begingroup$ Now I understand! Nice! - Just to make sure I understand correctly the status of my 3 questions after your post, do you agree that the reflection of isomorphisms (or conservatism) remains open? $\endgroup$ – Pierre-Yves Gaillard Oct 29 '18 at 15:36
  • $\begingroup$ All of them have been answered, if I didn't make stupid mistakes. The same example that fails reflection of monomorphisms works as a counterexample for the reflection of isomorphisms too. $\endgroup$ – user593746 Oct 29 '18 at 15:37
  • 1
    $\begingroup$ I will take a look at them later, but I cannot promise to be able to help. $\endgroup$ – user593746 Oct 29 '18 at 16:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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