-1

For a unital algebra $A$, and a two-sided ideal $I$ of $A \otimes A$, does there exists a universal quotient algebra $B$ of $A$, with map $q: A \twoheadrightarrow B$, such that $q \otimes q: A \otimes A \twoheadrightarrow B \otimes B$ factors through $(A \otimes A)/I$?

i.e. if there exists another quotient algebra $C$ of $A$ such the map $f: A \otimes A \twoheadrightarrow C \otimes C$ factors through $(A \otimes A)/I$, then the map $A \to C$ factors through $q$?

Hello
  • 173

1 Answers1

2

The answer is "not always". Below, I will give two explicit counterexamples and two explicit examples, and afterwards (this will be the main part) give a general criterion for the existence of this universal quotient. I assume that we work with algebras over a field $K$. This ensures that the tensor product over $K$ preserves monomorphisms.

Examples

a) When $I = 0$, the quotient exists and is just $A$.

b) When $A = K[X]$ and $I = \langle X \otimes X \rangle $ the universal quotient does exist: It is the universal $K$-algebra generated by an element $a$ such that $a \otimes a = 0$. This is equivalent to $a = 0$. So the answer is $K$ with $a=0$.

Counterexamples

a) Let $A = K[X,Y]$ and $I = \langle X \otimes Y \rangle \subseteq A \otimes A$. Here the universal quotient would be a universal $K$-algebra $B$ generated by two commuting elements $x,y \in B$ with $x \otimes y = 0$ in $B \otimes B$. We must have $x \neq 0$, witnessed by the (non-universal) example $1 \otimes 0 = 0$ in $K \otimes K$. Similarly, we must have $y \neq 0$. But linear algebra tells us that $x,y \neq 0$ implies $x \otimes y \neq 0$, a contradiction.

b) Let $A = K[X]$ and $I = \langle X \otimes 1 - 1 \otimes X \rangle$. The universal quotient would be a universal $K$-algebra $B$ generated by an element $x$ satisfying $x \otimes 1 = 1 \otimes x$. By considering $1 \otimes 1 = 1 \otimes 1$ in $K \otimes K$, we see $x \neq 0$ and $1 \neq 0$ in $B$. Therefore, $x \otimes 1 = 1 \otimes x$ means $x = \lambda 1$ for some $\lambda \in K^*$, but $\lambda$ cannot be chosen universally (unless $K = \mathbb{F}_2$ of course, in which case $K$ with $1 \in K$ is the universal quotient).

General criterion

Assume that a universal quotient $q : A \twoheadrightarrow B$ exists for an ideal $I \subseteq A \otimes A$. The first step is to show that the universal property, initially stated only for surjective homomorphisms $A \to C$, actually holds for all homomorphisms of algebras.

Let $f : A \to C$ be any algebra homomorphism such that $f \otimes f : A \otimes A \to C \otimes C$ vanishes on $I$, we need to show that $f$ factors uniquely through $q$. Uniqueness is clear since $q$ is assumed to be surjective. Existence is clear by assumption when $f$ is surjective, but if $f$ is not surjective, consider the image decomposition of $f$ as $f' : A \twoheadrightarrow C'$ followed by $i : C' \hookrightarrow C$. Since $i \otimes i$ is injective, then $f' \otimes f'$ vanishes on $I$. Hence, $f'$ factors through $q$. And this shows that also $f$ factors through $q$.

As always, this universal property can also be stated in terms of a representable functor. Here, we consider the functor $F_I : \mathbf{Alg}_k \to \mathbf{Set}$, a subfunctor of $\hom(A,-)$ actually, that maps an algebra $C$ to the set of homomorphisms $f : A \to C$ such that $f \otimes f$ vanishes on $I$. By what we saw above, $F_I$ is representable if $q$ exists. The converse is also true, since it can be checked that every representing object $A \to B$ of $F_I$ must be surjective.

Every representable functor preserves all limits, and by Freyd's representability criterion, the converse is also true in our case, since the (regular) quotients of $A$ provide a solution set for $F_I$. So we should analyse the condition that $F_I$ preserves all limits, or equivalently, equalizers and products (see here for the equivalence). But by a similar argument as above it is easy to check that $F_I$ preserves equalizers anyway.

Hence, a universal quotient exists for $I$ if and only if $F_I$ preserves products. This means that for a family of homomorphisms $(f_i : A \to C_i)_{i \in I}$, if each $f_i \otimes f_i : A \otimes A \to C_i \otimes C_i$ vanishes on $I$, then $f \otimes f : A \otimes A \to \prod_{i \in I} C_i \otimes \prod_{i \in I} C_i$ vanishes on $I$, where $f : A \to \prod_{i \in I} C_i$ is the homomorphism induced by $f_i$.

The natural map $\prod_{i \in I} C_i \otimes \prod_{i \in I} C_i \to \prod_{i,j \in I} C_i \otimes C_j$ is injective (since we are working over a field). Therefore, $f \otimes f$ vanishes on $I$ if and only if each $f_i \otimes f_j : A \otimes A \to C_i \otimes C_j$ vanishes on $I$.

So the condition reduces to: If two homomorphisms $f : A \to C$, $g : A \to C'$ are given such that $f \otimes f$ and $g \otimes g$ vanish on $I$, then also $f \otimes g$ has to vanish on $I$ (and by symmetry then also $g \otimes f$). Now, the kernel of $f \otimes g$ is equal to $\ker(f) \otimes A + A \otimes \ker(g)$, see here. Therefore:

General Criterion. The universal quotient exists for $I \subseteq A \otimes A$ if and only if for all ideals $J,J' \subseteq A$ with $I \subseteq J \otimes A + A \otimes J$ and $I \subseteq J' \otimes A + A \otimes J'$, we also have $I \subseteq J \otimes A + A \otimes J'$.

I am not sure if this condition can be simplified further.

  • Hi Martin, Thank you very much for your detailed answer! And thanks also for the very clear examples! Really helps – Hello Sep 23 '24 at 15:15
  • You are welcome! I believe that the quotient only exists in "uninteresting" cases, namely when the quotient collapses somehow. At least this is witnessed by the examples I gave. I wonder if the criterion I gave can be used to prove this. – Martin Brandenburg Sep 23 '24 at 16:44
  • Yes, it seems like most of the time this quotient will not exist. Moreover, it seems that the criterion you gave works where the k-algebras are just replaced with vector spaces over k – Hello Sep 24 '24 at 10:50
  • You mean given a subspace $U \subseteq V \otimes V$, we want a universal quotient vector space $q : V \to W$ such that $q \otimes q$ vanishes on $U$? I don't think that exists either always. – Martin Brandenburg Sep 24 '24 at 11:01
  • Yes, that is what I mean. I guess my question is whether your proof in the above actually requires the fact that we are dealing with algebras and algebra maps? – Hello Sep 24 '24 at 11:35