\classheader{2012-11-21} \begin{theorem} Let $A$ be a finite-dimensional algebra over $k$. Then \begin{enumerate} \item $J(A)$ is the maximal nilpotent ideal of $A$, i.e., $J(A)$ is a nilpotent ideal and it contains any other nilpotent ideal. \item $A/J(A)$ is a semisimple algebra. \item $A$ has only finitely many maximal two-sided ideals. \item $J(A) = \bigcap$ maximal two-sided ideals. \end{enumerate} \end{theorem} \begin{proof}[Proof of 1] Let $J:=J(A)$. We have a decreasing chain of two-sided ideals $A \supset J \supset J^2 \supset J^3 \supset \dotsb$, which must stabilize since $A$ is finite-dimensional (hence Artinian). Therefore there exists $N \gg 0$ such that $J^{N+1} = J^N$. Since $J^N$ is a finitely generated $A$-module, Nakayama's lemma implies that $J^N=0$, so $J$ is nilpotent. \end{proof} \begin{proof}[Proof of 2] $J(A/J(A))=0$ so by Problem 6 of Homework 8, we know $A/J(A)$ is semisimple. \end{proof} \begin{proof}[Proof of 3] Wedderburn's theorem implies that $A/J(A) = A_1 \oplus \dotsb \oplus A_n$ where $A_i = \M_{r_i}(D_i)$ is a simple algebra. Any two-sided ideal $I \subset \bigoplus A_i$ has the form $I = I_1 \oplus \dotsb \oplus I_n$ where $I_n$ is a two-sided ideal in $A_i$. The maximal ideals in $A/J(A)$ are $A_1 \oplus \dotsb \oplus A_{i-1} \oplus 0 \oplus A_{i+1} \oplus \dotsb \oplus A_n$. We claim that if $\mathfrak a$ is a maximal two-sided ideal in $A$, then $J(A) \subset \mathfrak a$. If this were not the case, then $J(A) + \mathfrak a = A$. Hence $1 = j+a$ for some $j\in J(A), a \in \mathfrak a$. But then $a= 1-j$ is invertible, a contradiction. In general, we have a bijection \[\{\text{maximal two-sided ideals of }A\text{ containing }J(A)\}\longleftrightarrow \{\text{maximal two-sided ideals of }A/J(A)\}.\] Thus, the number of maximal two-sided ideals is finite. \end{proof} \begin{proof}[Proof of 4] This is clear from our description of maximal ideals in $A/J(A)$ from part 3. \end{proof} \begin{remark} Let $A$ be a ring, $J \subset A$ a two-sided ideal, and $M$ a left (resp.~right) $A$-module. Then $M/JM = (A/J) \otimes_A M$ (resp. $M/MJ=M \otimes_A (A/J)$) is a left (resp.~right) $(A/J)$-module. \end{remark} \begin{theorem} Let $A$ be a finite-dimensional algebra over an algberaically closed field $k$. Let $J=J(A)$, and define $\bar{A} := A/J$. Then $A \cong (T_{\bar{A}} (J/J^2))/I$ where $I$ is a two-sided ideal in the tensor algebra satisfying \[ T^{\ge 2}(J/J^2)\supseteq I \supseteq T^{\ge N}(J/J^2) \] for sufficiently large $N$. \end{theorem} \begin{corollary} Let $A$ be finite-dimensional over an algebraically closed field $k$ such that \[\bar A = A/J(A) = \underbrace{k \oplus \dotsb \oplus k}_{n\text{ times}}.\] Then there exists a finite quiver $Q$ with vertex set $\{1,\dotsc,n\}$ such that $A = kQ/I$ with $(kQ)_{\ge 2} \supseteq I \supseteq (kQ)_{\ge N}$ for some sufficiently large $N$. \end{corollary} \begin{proof} By the remark, $E:=J/J^2$ is an $A/J$-bimodule. Since $A/J$ is a direct sum of fields, we have a decomposition $E = \bigoplus E_{ij}$. Define the quiver $Q$ so that $E$ corresponds to the paths. \end{proof} \begin{proof}[Proof of Theorem 2] The proof proceeds in several steps. {\it Step 1:} We claim that there exists a section $\varepsilon : \bar A \to A$, i.e. that there exists a subalgebra $A' \subset A$ such that $A' \hookrightarrow A \to \bar A$ is an isomorphism. \vspace{-0.1in} \begin{quotation} \begin{lemma}[Lifting of idempotents] If $J \subset A$ is a nilpotent two-sided ideal, let $\bar e_1,\dotsc,\bar e_n$ be a collection of orthogonal idempotents, i.e., $\bar e_i \cdot\bar e_j = \delta_{ij} \bar e_i$. Then there exist orthogonal idempotents $e_i \in A$ such that $e_i \bmod J = \bar e_i$ for all $i=1,\dotsc,n$. \end{lemma} \end{quotation} The above is Corollary 7.5 of Etingof's \emph{Introduction to representation theory} and we omit the proof. Since $\bar A$ is semisimple over an algebraically closed field, it is a direct sum of matrix algebras $\bigoplus_\ell \M_{n_\ell}(k)$. Let $\bar e_{\ell,ij}$ denote the matrix with a $1$ in the $(i,j)$ position as an element of $\M_{n_\ell}(k)$. Then $\{\bar e_{\ell,ii}\}$ forms a collection of orthogonal idempotents in $\bar A$, so by the lemma we can lift them to orthogonal idempotents in $e_{\ell,ii} \in A$. Since $\sum_{\ell,i} e_{\ell,ii} \in 1+J$ is idempotent and invertible, it is $1$. Let $e_\ell = \sum_i e_{\ell,ii}$. Then $e_\ell A e_\ell$ are orthogonal subalgebras of $A$ for distinct $\ell$. Hence it suffices to lift each $\M_{n_\ell}(k)$ to $e_\ell A e_\ell$. We therefore drop the $\ell$ subscript and assume $\sum e_{ii} = 1$. Suppose that for $I$ a two-sided square-zero ideal of $A$, we can lift a matrix subalgebra of $A/I$ to $A$. Then applying this to the sequence $A = A/J^N \to A/J^{N-1} \to \dotsb \to A/J = \bar A$ with $I = J^i/J^{i+1} \subset A/J^{i+1}$, we are done. To prove the square-zero case, i.e., $\bar A = A/I$ and $I^2=0$: first take an arbitrary lift $e_{i,i+1} \in A$ of $\bar e_{i,i+1} \in \bar A$. By multiplying on the left by $1-e_{jj}$ for $j \ne i$ and on the right by $1-e_{jj}$ for $j \ne i+1$, we get $e_{jj} e_{i,i+1} = \delta_{ij}$ and $e_{i,i+1} e_{jj} = \delta_{i+1,j}$. Do the same for $e_{i+1,i}$. Then $e_{i,i+1} e_{i+1,i} - e_{ii} \in I$ implies \[ (e_{i,i+1} e_{i+1,i} - e_{ii})^2 = e_{i,i+1} (r-e_{i+1,i}) + e_{ii} = 0 \] where $r = e_{i+1,i} e_{i,i+1} e_{i+1,i} - e_{i+1,i} \in I$. Thus $e_{i,i+1}(e_{i+1,i} - r) = e_{ii}$. The analogous computation shows that $(e_{i+1,i}-r)e_{i,i+1} = e_{i+1,i+1}$. Replacing $e_{i+1,i}$ with $e_{i+1,i}-r$, we can assume that \[ e_{i+1,i} e_{i,i+1} = e_{i+1,i+1} , \quad e_{i,i+1} e_{i+1,i} = e_{ii}. \] Now set $e_{ij} = e_{i,i+1} e_{i+1,i+2} \dotsb e_{j-1,j}$ and $e_{ji} = e_{j,j-1} \dotsb e_{i+1,i}$ for $i