On Page 13 of Bourbaki's Chapter I on Lie algebras, it is proven that if $\mathfrak g_1, \mathfrak g_2$ are Lie algebras and $\mathrm U$ denotes the functor from the category of Lie algebras over a field $F$ to the category of associative unital $F$-algebras taking the universal envelope algebra, then $$ \mathrm U(\mathfrak g_1 \oplus \mathfrak g_2) \simeq \mathrm U(\mathfrak g_1) \otimes_F \mathrm U(\mathfrak g_2). $$ Given an associative $F$-algebra $A$, denote by $[A]$ the corresponding Lie algebra over $F$ defined by the classical Lie bracket, namely $[x,y] = xy-yx$. This is functorial and $(\mathrm U,[-])$ form an adjoint pair. So I thought that the above was a consequence of the fact that $\mathrm U$ was left-adjoint and commuted with small colimits. However, neither of the above are coproducts, namely $\mathfrak g_1 \oplus \mathfrak g_2$ is not the coproduct in the category of Lie algebras and $\mathrm U(\mathfrak g_1) \otimes_F \mathrm U(\mathfrak g_2)$ is not the coproduct in the category of associative $F$-algebras.

So can anyone figure out categorically why this isomorphism is to be expected, or is it just a coincidence? I usually don't like coincidences. The most confusing part is that this isomorphism is constructed "naturally" in some way, namely we take the $F$-linear maps $\mathfrak g_i \to \mathrm U(\mathfrak g_i) \to \mathrm U(\mathfrak g_1) \otimes_F \mathrm U(\mathfrak g_2)$, use the fact that their images commute (which is precisely why $\mathfrak g_1 \oplus \mathfrak g_2$ is not the coproduct in the category of Lie algebras, i.e. the images do not always commute in general) and construct the map of Lie algebras $\mathfrak g_1 \oplus \mathfrak g_2 \to [\mathrm U(\mathfrak g_1) \otimes_F \mathrm U(\mathfrak g_2)]$ which then corresponds to a map $\mathrm U(\mathfrak g_1 \oplus \mathfrak g_2) \to \mathrm U(\mathfrak g_1) \otimes_F \mathrm U(\mathfrak g_2)$ by the $(\mathrm U,[-])$ adjunction.

The inverse map is also constructed naturally, namely by applying the functor $\mathrm U$ on the inclusions $\mathfrak g_i \to \mathfrak g_1 \oplus \mathfrak g_2$ and tensoring the resulting maps over $F$. It turns out that the two are inverse to each other, but I can't naturally figure out why, I have to dive into formulas with elements to see it.


It is not a coincidence. The tensor product of algebras and the direct sum of Lie algebras are both "commutative coproducts": they are the universal objects admitting a map from two objects whose images commute (this condition makes sense for both algebras and Lie algebras, compatibly with the forgetful functor from algebras to Lie algebras).

  • $\begingroup$ I understand this informally, but how does one remove the " " in "commutative coproducts"? Can I see them as some kind of colimits to use the fact that left adjoints commute with colimits? I can't think of a diagram which means "the images commute". $\endgroup$ – Patrick Da Silva Dec 7 '16 at 23:50
  • $\begingroup$ @Patrick: I don't know if commutative coproducts can literally be seen as colimits, but you don't need this in order to see the isomorphism in a natural way without working with elements. The point is that both the LHS and the RHS are both universal with respect to admitting a map from $\mathfrak{g}_1$ and $\mathfrak{g}_2$ whose images commute; that is, commutative coproducts are a convenient way of seeing why the LHS and the RHS represent the same functor. $\endgroup$ – Qiaochu Yuan Dec 8 '16 at 1:11
  • $\begingroup$ Ah ok, so instead of using adjoints I'd have to go with the good old Yoneda lemma. Thanks! $\endgroup$ – Patrick Da Silva Dec 8 '16 at 1:14

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.