Pome

12.31 Symmetric tensors on split elliptic ruled surfaces7CF6C47CAA92

7CF6C47CAA92

To compute the section algebra of a canonical extension, we first compute that of the tangent bundle of $M$. A holomorphic section of $\operatorname {Sym}^ rT_ M$ is the same as a holomorphic function on the total space of the cotangent bundle that is a homogeneous polynomial of degree $r$ on each fiber. The $\mathbb C^*$-action on the rulings lifts to the cotangent bundle and decomposes these polynomials into weights. Each weight space is the space of sections of an explicit vector bundle on $E$, built from powers of $A$ and symmetric powers of a nonsplit bundle $U$ of rank two. Riemann–Roch then gives exact dimensions: the tautological line bundle on $\mathbb P(T_ M)$ is big, while a square-zero quotient shows that the section algebra is not finitely generated.

Keep $E$, the line bundle $A$ of degree $d>0$, the surface $M$ and the normalized factors $b_\lambda $ of lemma 12.30.2, and put

\[ q_\lambda =-b_\lambda '/b_\lambda =2\pi \mathrm{i}dm, \qquad \lambda =m\tau +n. \]

Write $u\, \mathrm{d}z+y\, \mathrm{d}t$ for a covector in the finite ruling chart. Evaluating covectors on the Euler field $t\partial /\partial t$ of the rulings gives the global fiberwise linear function $h=ty$ on the cotangent bundle. The dimensions of the weight spaces measure how many symmetric tensors there are, and their products detect the failure of finite generation.

Let $U$ be the rank-two vector bundle on $E$ whose sections, pulled back to $\mathbb C$, are the linear polynomials $F(z,h,u)$ in formal variables $h,u$ with entire coefficients satisfying

\[ F(z+\lambda ,h,u)=F(z,h,u-q_\lambda h). \]

The invariant polynomial $h$ and the coefficient of $u$ give an exact sequence $0\to \mathcal O_ Eh\to U\to \mathcal O_ E\to 0$.

Lemma 12.31.1.2604E6F7AC76 For every integer $r\geq 0$, the space $H^0(M,\operatorname {Sym}^ rT_ M)$ is the direct sum of finitely many weight spaces, with weights $-r\leq \ell \leq r$. The weight-zero space is $H^0(E,\operatorname {Sym}^ rU)$. For $1\leq k\leq r$, the space of weight $-k$ is

\[ H^0(E,A^ k\otimes \operatorname {Sym}^{r-k}U), \]

and the space of weight $k$ is $H^0(E,A^{-k}\otimes \operatorname {Sym}^{r-k}U)$. Multiplication adds tensor degrees and weights.

Proof. On the cover $\mathbb C\times \mathbb P^1$, the cotangent coordinates of the two ruling charts are related by

\[ v=1/t,\qquad u_\infty =u,\qquad y_\infty =-t^2y, \]

and the lattice acts by

\[ (z,t,u,y)\longmapsto (z+\lambda ,b_\lambda t,u+q_\lambda ty,y/b_\lambda ); \]

both formulas follow by preserving the one-form $u\, \mathrm{d}z+y\, \mathrm{d}t$. In the other ruling chart the action is

\[ (z,v,u_\infty ,y_\infty )\longmapsto (z+\lambda ,b_\lambda ^{-1}v, u_\infty -q_\lambda v y_\infty ,b_\lambda y_\infty ), \]

which is holomorphic at $v=0$ and agrees with the formula in the finite chart. Both formulas preserve $h=ty=-v y_\infty $.

A function that is homogeneous of degree $r$ on the fibers is a sum of terms $g_ b(z,t)y^ bu^{r-b}$, $0\leq b\leq r$, with $g_ b$ holomorphic on $\mathbb C^2$. Since $y=-v^2y_\infty $, holomorphy at $v=0$ says that $v^{2b}g_ b(z,1/v)$ is holomorphic there; there is no cancellation between different $b$, because the monomials in $u_\infty ,y_\infty $ are linearly independent. Expanding in $t$ therefore gives

\[ g_ b(z,t)=\sum _{a=0}^{2b}c_{a,b}(z)t^ a, \]

with entire coefficients.

Give $t$ weight $1$, $y$ weight $-1$ and $u$ weight $0$, so that $t^ ay^ bu^{r-b}$ has weight $\ell =a-b$. The chart changes and the lattice action preserve this grading; equivalently, it is the grading by the cotangent lift of the global action $t\mapsto ct$, $c\in \mathbb C^*$. The bounds $0\leq a\leq 2b$ and $b\leq r$ give $-r\leq \ell \leq r$.

For $\ell =-k<0$, a homogeneous function of weight $\ell $ can be written uniquely as $y^ kF(z,h,u)$ with $F$ homogeneous of degree $r-k$. Its invariance is equivalent to

\[ F(z+\lambda ,h,u) =b_\lambda (z)^ kF(z,h,u-q_\lambda h), \]

which is the transformation rule for sections of $A^ k\otimes \operatorname {Sym}^{r-k}U$. Conversely, each such polynomial defines a function that is holomorphic in both ruling charts, because its exponents satisfy $0\leq a\leq 2b$. For $\ell =k>0$, the condition $a\leq 2b$ shows that a function of weight $\ell $ can be written uniquely as $t^ kh^ kF(z,h,u)$, with $F$ of the same remaining degree $r-k$. Its invariance involves the factor $b_\lambda ^{-k}$, which gives $A^{-k}\otimes \operatorname {Sym}^{r-k}U$. Weight zero consists exactly of the homogeneous polynomials $F(z,h,u)$ with the transformation rule of $\operatorname {Sym}^ rU$. Products of polynomial functions add degrees and weights, which proves the last assertion. $\square$

Lemma 12.31.2.711AF6E8C0D8 The extension $U$ does not split. For every $s\geq 0$,

\[ H^0(E,\operatorname {Sym}^ sU)=\mathbb C h^ s. \]

For $k\geq 1$ and $s\geq 0$,

\[ \begin{aligned} H^0(E,A^{-k}\otimes \operatorname {Sym}^ sU)& =0,\\ H^1(E,A^ k\otimes \operatorname {Sym}^ sU)& =0,\\ h^0(E,A^ k\otimes \operatorname {Sym}^ sU)& =kd(s+1), \end{aligned} \]

where $h^0=\dim H^0$. For $s\geq 1$, taking the coefficient of $u^ s$ defines a surjection

\[ H^0(E,A\otimes \operatorname {Sym}^ sU)\longrightarrow H^0(E,A), \]

whose kernel is $hH^0(E,A\otimes \operatorname {Sym}^{s-1}U)$.

Proof. A splitting of $U$ would give an invariant section $u+a(z)h$ lifting $1$, with $a$ entire and

\[ a(z+1)=a(z),\qquad a(z+\tau )=a(z)-2\pi \mathrm{i}d. \]

The derivative $a'$ would then be lattice periodic, hence constant on the compact torus. The first identity forces this constant to vanish, contradicting the second. Thus the extension does not split.

More generally, write an invariant polynomial of degree $s$ as

\[ F(z,h,u)=\sum _{j=0}^ s a_ j(z)h^{s-j}u^ j. \]

If its largest nonzero $u$-degree is $j>0$, then $a_ j$ is lattice periodic, hence a nonzero constant. Comparing the next coefficients gives

\[ a_{j-1}(z+1)=a_{j-1}(z),\qquad a_{j-1}(z+\tau )=a_{j-1}(z)-j(2\pi \mathrm{i}d)a_ j, \]

and the same derivative argument gives a contradiction. Hence $F=a_0h^ s$ with $a_0$ constant; this also covers $s=0$.

The powers of the invariant line $\mathcal O_ Eh$ give a filtration of $\operatorname {Sym}^ sU$ whose $s+1$ successive quotients are trivial line bundles. A line bundle of negative degree on a compact curve has no nonzero section, since the zero divisor of a nonzero section has nonnegative degree. Hence the filtration twisted by $A^{-k}$ has no global sections. Serre duality and $\Omega _ E^1\simeq \mathcal O_ E$ give $H^1(E,A^ k)=0$, and the cohomology sequences of the filtration twisted by $A^ k$ then give $H^1(E,A^ k\otimes \operatorname {Sym}^ sU)=0$. Riemann–Roch on the curve of genus one, for a bundle of rank $s+1$ and degree $kd(s+1)$, gives the dimension formula.

Finally, the leading coefficient gives the exact sequence of bundles

\[ 0\longrightarrow A\otimes \operatorname {Sym}^{s-1}U \xrightarrow {\ h\ }A\otimes \operatorname {Sym}^ sU \longrightarrow A\longrightarrow 0. \]

Indeed, translating $u$ by a multiple of $h$ does not change the leading coefficient, and a polynomial with zero leading coefficient is divisible by $h$. The vanishing of $H^1(E,A\otimes \operatorname {Sym}^{s-1}U)$ just proved gives the asserted surjectivity and kernel on global sections. $\square$

For a line bundle $\mathcal L$ on a projective manifold $Y$ of dimension $N$, recall that its volume is $\operatorname {vol}(\mathcal L)=\limsup _{r\to \infty }N!\, h^0(Y,\mathcal L^ r)/r^ N$, and that $\mathcal L$ is big if and only if $\operatorname {vol}(\mathcal L)>0$. For a vector bundle $F$ we use the quotient convention introduced in the proof of theorem 12.30.4, in which the projection $\pi :\mathbb P(F)\to Y$ satisfies $\pi _*\mathcal O_{\mathbb P(F)}(r)=\operatorname {Sym}^ rF$.

Theorem 12.31.3.654B70C2AFD7 For every integer $r\geq 0$,

\[ h^0(M,\operatorname {Sym}^ rT_ M) =1+\frac{d\, r(r+1)(r+2)}6. \]

The tautological line bundle $\mathcal O_{\mathbb P(T_ M)}(1)$ is big and has volume $d$.

Proof. By lemma 12.31.1, lemma 12.31.2, the positive weights contribute nothing, the weight zero contributes one dimension, and the weight $-k$ contributes $kd(r-k+1)$ dimensions. Thus

\[ h^0(M,\operatorname {Sym}^ rT_ M) =1+d\sum _{k=1}^ r k(r-k+1) =1+\frac{d\, r(r+1)(r+2)}6. \]

The elliptic curve and its holomorphic line bundle $A$ are algebraic, so $M$ and $\mathbb P(T_ M)$ are projective. The identity $\pi _*\mathcal O_{\mathbb P(T_ M)}(r)=\operatorname {Sym}^ rT_ M$ identifies the section spaces. Since $\mathbb P(T_ M)$ has dimension three,

\[ \operatorname {vol}\bigl (\mathcal O_{\mathbb P(T_ M)}(1)\bigr ) =\limsup _{r\to \infty }\frac{6h^0(M,\operatorname {Sym}^ rT_ M)}{r^3}=d>0, \]

so the tautological line bundle is big. $\square$

Theorem 12.31.4.3B7C9C492833 The complex algebra

\[ R=\bigoplus _{r\geq 0}H^0(M,\operatorname {Sym}^ rT_ M) \]

is not finitely generated.

Proof. Write $R_{r,k}$ for the component of tensor degree $r$ and weight $-k$. By the preceding results only $k\geq 0$ occurs, multiplication adds both indices, and the subalgebra of weight zero is $\mathbb C[h]$. Let $I$ be the ideal formed by all components with $k>0$, and put

\[ S=R/(hR+I^2). \]

Its component of weight zero is $\mathbb C$, and its ideal $J$ of positive weight index satisfies $J^2=0$, so $S=\mathbb C\oplus J$ as a vector space.

For every $r\geq 1$, products in $I^2$ have weight index at least two. Consequently the component of $S$ of degree $r$ and weight index one is

\[ R_{r,1}/hR_{r-1,1}\simeq H^0(E,A). \]

For $r=1$ this holds because $R_{0,1}=0$, and for $r\geq 2$ it is the last assertion of lemma 12.31.2 with $s=r-1$. Each of these components has dimension $d>0$, and they lie in distinct degrees, so $J$ is infinite dimensional.

If $R$ were finitely generated, its quotient $S$ would be finitely generated too. Write a finite set of algebra generators of $S$ as $c_ i+j_ i$, $1\leq i\leq N$, with $c_ i\in \mathbb C$ and $j_ i\in J$. Because $J^2=0$, every polynomial in these generators lies in the finite-dimensional span of $1,j_1,\ldots ,j_ N$. This contradicts the infinite dimension of $J$. $\square$

Cite this page

Comments

Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.