Peng Zhou

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== From B-fields to the Quantum Torus: a Toy Model in Wrapped Fukaya Categories ======

===== 0. The question =====

Start with the mirror pair

$$
\operatorname{Coh}( ( \mathbb C^* )^2 )
\quad\leftrightarrow\quad
\operatorname{Fuk}(T^*T^2).
$$

On the B-side, quantization of $( \mathbb C^* )^2$ should replace the commutative Laurent polynomial algebra

$$
\mathbb C[x^{\pm 1},y^{\pm 1}]
$$

by the quantum torus

$$
\mathbb C_q[x^{\pm 1},y^{\pm 1}]
=
\mathbb C\langle x^{\pm 1},y^{\pm 1}\rangle/(xy=qyx).
$$

The question is:

What is the corresponding modification on the A-side?

The answer in this toy model is: twist the wrapped Fukaya category of $T^*T^2$ by a $B$-field whose flux through the base torus is nonzero.

===== 1. The ordinary wrapped Fukaya category of $T^*T^2$ =====

Let

$$
M=T^*T^2.
$$

Let $L=T^*_0T^2$ be a cotangent fiber. It is a basic generator of the wrapped Fukaya category.

One expects

$$
HW^0(L,L)\cong \mathbb C[ \pi_1(T^2) ].
$$

Since

$$
\pi_1(T^2)\cong \mathbb Z^2,
$$

this gives

$$
HW^0(L,L)
\cong
\mathbb C[x^{\pm1},y^{\pm1}].
$$

Here $x$ and $y$ correspond to the two basic loops on $T^2$. Since $\mathbb Z^2$ is abelian, the product is commutative:

$$
xy=yx.
$$

===== 2. Twisting the product by a multiplicative 2-cocycle =====

A quantum torus can be viewed as a twisted group algebra of $\mathbb Z^2$.

Let $e_m$ denote the basis element corresponding to

$$
m=(m_1,m_2)\in \mathbb Z^2.
$$

Instead of the ordinary product

$$
e_m e_n=e_{m+n},
$$

define

$$
e_m\star e_n=\sigma(m,n)e_{m+n},
$$

where

$$
\sigma:\mathbb Z^2\times\mathbb Z^2\to \mathbb C^*
$$

is a multiplicative 2-cocycle. The cocycle condition is

$$
\sigma(m,n)\sigma(m+n,k)
=
\sigma(n,k)\sigma(m,n+k).
$$

This is exactly the condition that $\star$ be associative.

For example, take

$$
\sigma(m,n)=q^{m_1n_2}.
$$

Then, if

$$
x=e_{(1,0)},\qquad y=e_{(0,1)},
$$

we get

$$
x\star y=q\,y\star x.
$$

Thus the quantum torus is just a twisted group algebra:

$$
\mathbb C_q[x^{\pm1},y^{\pm1}]
\cong
\mathbb C_\sigma[ \mathbb Z^2 ].
$$

===== 3. How the $B$-field produces the cocycle =====

Now write

$$
T^*T^2\cong \mathbb R^2_{u,v}\times T^2_{\theta,\phi}.
$$

Here $(\theta,\phi)$ are coordinates on the base torus and $(u,v)$ are cotangent fiber coordinates.

Take a $B$-field pulled back from the base:

$$
B=\beta\, d\theta\wedge d\phi.
$$

In the wrapped Floer product, multiplying $e_m$ and $e_n$ is geometrically represented by a holomorphic triangle, or morally by a filling of two composable base loops. A $B$-field weights this triangle by

$$
\exp\left(2\pi i\int_{\Delta(m,n)}B\right).
$$

Thus the product becomes

$$
e_m\star e_n
=
\exp\left(2\pi i\int_{\Delta(m,n)}B\right)e_{m+n}.
$$

This is the desired multiplicative cocycle.

The commutator phase is the $B$-field holonomy around the parallelogram spanned by $m$ and $n$:

$$
\frac{e_m\star e_n}{e_n\star e_m}
=
\exp\left(2\pi i\beta(m_1n_2-m_2n_1)\right).
$$

In particular,

$$
x\star y
=
\exp(2\pi i\beta)\, y\star x.
$$

So if

$$
q=\exp(2\pi i\beta),
$$

then

$$
x\star y=q\,y\star x.
$$

===== 4. Why associativity survives =====

The $A_\infty$-relations come from the boundary of one-dimensional moduli spaces of holomorphic polygons.

With a $B$-field, every polygon is weighted by

$$
\exp\left(2\pi i\int B\right).
$$

When two polygons are glued, the integrals add. Therefore the weights multiply.

For three elements $e_m,e_n,e_k$, associativity compares the two decompositions of a quadrilateral into triangles. Since $B$ is closed, the total $B$-field holonomy of the quadrilateral is independent of the triangulation. Algebraically, this gives

$$
\sigma(m,n)\sigma(m+n,k)
=
\sigma(n,k)\sigma(m,n+k).
$$

Thus the $B$-field holonomy automatically gives a multiplicative 2-cocycle.

===== 5. Relation with the logarithmic holomorphic symplectic form =====

On $( \mathbb C^* )^2$, write

$$
x=e^{u+i\theta},\qquad y=e^{v+i\phi}.
$$

The logarithmic holomorphic symplectic form is

$$
\Omega=d\log x\wedge d\log y.
$$

Expanding,

$$
\Omega
=
(du+i\,d\theta)\wedge(dv+i\,d\phi).
$$

Therefore

$$
\operatorname{Re}\Omega
=
du\wedge dv-d\theta\wedge d\phi,
$$

and

$$
\operatorname{Im}\Omega
=
du\wedge d\phi+d\theta\wedge dv.
$$

The base part of $\operatorname{Re}\Omega$ is

$$
-d\theta\wedge d\phi.
$$

This is exactly the kind of $B$-field term that produces the quantum torus phase.

So the quantum torus deformation can be thought of as coming from the base-torus component of the real part of the logarithmic holomorphic symplectic form.

===== 6. What about the extra $du\wedge dv$ term? =====

The form

$$
du\wedge dv
$$

is exact on the cotangent fibers:

$$
du\wedge dv=d(u\,dv).
$$

So the fiber-fiber part of the $B$-field is gauge-trivial.

More explicitly, if

$$
B_{\mathrm{fib}}=c\,du\wedge dv,
$$

then with

$$
\Lambda=c\,u\,dv
$$

we have

$$
B_{\mathrm{fib}}=d\Lambda.
$$

A $B$-field gauge transformation sends

$$
B\mapsto B-d\Lambda.
$$

Thus $c\,du\wedge dv$ can be gauged away.

At the brane level, one must also change the connection on the Lagrangian. If $L=T^*_0T^2$, then $B_{\mathrm{fib}}|_L\neq 0$. One can choose a connection one-form

$$
A=-2\pi i\,c\,u\,dv
$$

so that

$$
F_\nabla=dA=-2\pi i\,c\,du\wedge dv.
$$

Then

$$
F_\nabla+2\pi i B_{\mathrm{fib}}|_L=0.
$$

After the gauge transformation, both $B_{\mathrm{fib}}$ and this connection term disappear.

By contrast,

$$
d\theta\wedge d\phi
$$

is not exact on the base torus. Its period is the invariant quantity, and this is what gives the quantum torus relation.


===== 7. Moral =====

The ordinary wrapped Fukaya category gives

$$
HW^0(T^*_0T^2,T^*_0T^2)
\cong
\mathbb C[x^{\pm1},y^{\pm1}].
$$

Turning on a base $B$-field

$$
B=\beta\,d\theta\wedge d\phi
$$

twists the wrapped product by a cocycle and gives

$$
HW^0_B(T^*_0T^2,T^*_0T^2)
\cong
\mathbb C_q[x^{\pm1},y^{\pm1}],
$$

where

$$
q=\exp(2\pi i\beta).
$$

So, in this toy model,

$$
\boxed{
\text{quantizing } \operatorname{Coh}( ( \mathbb C^* )^2 )
\text{ corresponds on the A-side to a }B\text{-field twist of } \operatorname{Fuk}(T^*T^2).
}
$$

Or more poetically:

The two wrapped products $xy$ and $yx$ differ by the $B$-field flux through one fundamental parallelogram of the base torus.

===== 8. Branes in the presence of a $B$-field =====

There is one important brane-level point. In the ordinary Fukaya category, an object is not just a Lagrangian submanifold $L$. One usually equips $L$ with extra data, for example a grading, spin or Pin structure, and a local system. Without a $B$-field, this local system is usually flat. In rank one, we can write it locally as a connection $\nabla=d+A$ on a $U(1)$ line bundle over $L$, with curvature $F_\nabla=dA$. Flatness means $F_\nabla=0$.

In the presence of a $B$-field, a disk contribution contains both an interior term and a boundary term. For a holomorphic disk or polygon $u:(\Sigma,\partial\Sigma)\to (M,L)$, the weight looks like $\exp(2\pi i\int_\Sigma u^*B)\cdot \exp(\int_{\partial\Sigma}A)$.

The reason a non-flat connection may be needed is that $B$-fields have gauge transformations. If $B\mapsto B+d\Lambda$, then $\int_\Sigma B$ changes by the boundary term $\int_\Sigma d\Lambda=\int_{\partial\Sigma}\Lambda$. To keep the total disk weight invariant, one must transform the connection $A$ on the brane at the same time. The gauge-invariant combination is morally $2\pi i\,B+F_\nabla$.

For a Lagrangian brane, the usual compatibility condition is $F_\nabla+2\pi i\,B|_L=0$, or equivalently $F_\nabla=-2\pi i\,B|_L$. Thus the connection on $L$ is flat only when $B|_L=0$ in the relevant sense. If $B|_L$ is nonzero, the brane must carry a possibly non-flat line bundle connection whose curvature cancels the restriction of the $B$-field.

For example, take $L=\mathbb R^2_{u,v}\subset T^*T^2$. Suppose the $B$-field has a fiber-fiber part $B_{\mathrm{fib}}=c\,du\wedge dv$. Then $B_{\mathrm{fib}}|_L=c\,du\wedge dv$. A compatible connection is $\nabla=d+A$, where $A=-2\pi i\,c\,u\,dv$. Indeed, $F_\nabla=dA=-2\pi i\,c\,du\wedge dv$, so $F_\nabla+2\pi i\,B_{\mathrm{fib}}|_L=0$.

This is the concrete meaning of saying that the curvature of the brane line bundle cancels the $B$-field restricted to $L$.

In contrast, if $L$ is one-dimensional, for example a line $L=\mathbb R\subset \mathbb C$, then every two-form restricts to zero on $L$. In that case $B|_L=0$, so there is nothing to cancel, and the brane local system remains flat.

In the example relevant to the quantum torus, the useful $B$-field is $B=\beta\,d\theta\wedge d\phi$. For the cotangent fiber $L=T^*_0T^2$, the coordinates $\theta,\phi$ are constant along $L$. Therefore $B|_L=0$. So the cotangent fiber itself does not need a non-flat connection for this base $B$-field. The effect of the $B$-field instead appears in the holonomy over the holomorphic triangles contributing to the wrapped product. This is what twists $xy=yx$ into $xy=qyx$.