Past exam of the algebra, analysis and differential equations II course of the University of Oxford / 2010 / Algebra / 1

Write $n=\dim V$. If $Rv=\lambda v$ with $v\ne 0$, then $0=R^{a}v={\lambda }^{a}v$, hence $\lambda =0$. The characteristic polynomial splits over $\mathbb{C}$ and has degree $n$, soThere is indeed an eigenvector for the eigenvalue zero: a nonzero vector in the last nonzero image of $R$ lies in its kernel of a linear map. For the bound, choose $v$ with $R^{a-1}v\ne 0$, using minimality. The vectors $v,Rv,\dots ,R^{a-1}v$ are linearly independent. Otherwise choose the least $j$ with nonzero coefficient in a relation and apply $R^{a-1-j}$; only its nonzero multiple of $R^{a-1}v$ survives, a contradiction. Therefore $\boxed{a\le n}$.

Suppose $S^b=0$. Use the increasing chain of vector subspacesChoose a basis successively, extending a basis of each kernel of a linear map to one of the next kernel of a linear map. If a basis vector first enters at level $j$, its image under $S$ belongs to $\ker S^{j-1}$, spanned entirely by earlier basis vectors. Thus every matrix column has nonzero entries only above its diagonal: the matrix is strictly upper triangular.

Conversely, for such a matrix and its ordered basis $e_1,\dots ,e_n$, one has $S⟨e_1,\dots ,e_j⟩\subseteq ⟨e_1,\dots ,e_{j-1}⟩$. Iterating $n$ times kills every vector. Hence $\boxed{S^n=0}$, establishing both directions without assuming a Jordan form.

The polynomials $t^c$ and $(t-1{)}^d$ are coprime. Bézout's identity gives polynomials $u,v$ with $u(t)t^c+v(t)(t-1{)}^d=1$. SetFor every $x$, the decompositionhas first summand in $V_0$ and second in $V_1$, since $T^c(T-I{)}^d=0$ and polynomial expressions in $T$ commute. If $x$ belongs to both kernels, the same Bézout identity gives $x=0$. Thus $V=V_0\oplus V_1$. Both vector subspaces are invariant under $T$.

On $V_0$, the restriction of $T$ is nilpotent; on $V_1$, the restriction of $T-I$ is nilpotent. Apply (b) separately, then concatenate the two bases. This yieldswhere $A,N$ are strictly upper triangular. Empty blocks are allowed if one subspace is zero. In particular $C=I+N$ has precisely the required diagonal.