Let $C$ be a smooth projective curve over $\mathbb{C}$. Let $A$ be a degree $d$ line bundle on $C$, and $M$ be a degree 0 line bundle on $C$ such that $M^2=\mathcal{O}_C$, that is, it is a 2-torsion line bundle. Therefore, we have that $deg(A)=deg(A\otimes M)$. What is $h^0(C,A\otimes M)$ in general? Can we say that it is zero?

More precisely,

We have the subvariety $W^r_d(C)=\{A\in Pic^d(C)|h^0(C,A)\geq r+1\}$ of $Pic^d(C)$. A 2-torsion line $M$ acts on $Pic^d(C)$. Assume $0< d < g$ Then what will the action of $M$ on $Pic^d(C)$ do to $W^r_d(C)\setminus W^{r+1}_d(C)$ ? In fact consider $W^0_d(C)=\{A\in Pic^d(C):h^0(C,A)\geq 1\}$. This has expected dimension $\rho(g,0,d)<g$ when $d<g$. This will get translated to some other variety under translation by $M$ right? So that $h^0(A\otimes M)=0$. How do we prove this?

I have attempted the following: Consider $t_M:Pic^d(C)\longrightarrow Pic^d(C)$ given by $B\mapsto B\otimes M$. This is an isomorphism. Now $A\in Pic^d(C)$ is such that $h^0(A)\geq 1$. That is $A\in W^0_d(C)$.We have that $W^0_d(C)\subset Pic^d(C)$. Now, every component of $W^0_d(C)$ has expected dimension (when $d<g$): $\rho(g,0,d)=g-1(g-d+0)=d< g$.\ Now $Pic^d(C)$ is a $g$-dimensional variety, so under translation by $M\in Pic^0(C)$, a general element of the subvariety $W^0_d(C)$ of $Pic^d(C)$ moves to another outside $W^0_d(C)$. Hence for a general $A\in W^0_d(C)$, $h^0(C,A\otimes M)=0$.

Is this correct? How to improve upon this answer?

  • 2
    $\begingroup$ Certainly not. Take for $C$ a hyperelliptic curve: then $W^1_2$ has only one point, which is not preserved by tensor product with any line bundle of order 2. In fact, I believe the situation you describe never happens when $\dim W^r_d<g$. $\endgroup$ – abx Oct 2 '15 at 6:05
  • $\begingroup$ Thanks @abx, if we consider $W^{g-2}_{2g-2}\setminus W^{g-1}_{2g-2}$ this will be preserved by tensor product by a order 2 line bundle. That's why I had this doubt. How do we prove your claim? Can you give me any leads? $\endgroup$ – gradstudent Oct 2 '15 at 15:41
  • 1
    $\begingroup$ Have you heard of Prym differentials? They are sections of the tensor product of the canonical bundle with a 2 torsion line bundle. This 2 torsion bundle determines a double cover D of the curve C, and the Prym differentials on C correspond to differentials on D that are skew symmetric for the sheet exchange involution of D. Maybe that pattern generalizes. (Oh and there are always only g-1 indept Prym diferentials on a curve of genus g.) $\endgroup$ – roy smith Oct 2 '15 at 15:44
  • $\begingroup$ One statement in the question is wrong $W^0_d$ has expected (and actual) dimension d. $\endgroup$ – meh Nov 4 '15 at 14:08
  • 2
    $\begingroup$ Unless I misunderstand your notation $W_0^d = {A \mbox{ s.t. } h^0(A) \geq 1 } = {C_d)} $ and the expected dimension is d . In addition, plugging r=0 into the formula $g -(g-d+r)(r+1) gives the Brill-Noether number of effective line bundles is d, which is to be 'expected'. $\endgroup$ – meh Nov 4 '15 at 17:17

For a definite class of counterexamples, look at theta characteristics. These are defined as the line bundles whose square is the canonical class. So, by definition, any two theta characteristics differ by a two-torsion line bundle. A theta characteristic $L$ is called even or odd according to whether $H^0(C,L)$ is even or odd. The parity of a theta characteristic is a non-trivial quadratic form modulo 2 on the space of theta characteristics and, in particular, is non-constant. (See Mumford Ann Sci ENS 1971). This shows that $W^0_{g-1}$ is not invariant by translation by two torsion.


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.