I am reading Kontsevich' famous paper on deformation quantization of Poisson manifolds. In section 1.4.2 on page 4 he gives the general formula for the star product associated to a Poisson structure on $\mathbb{R}^n$ up to and including order 2. Now I have noticed that in the second order term, terms of the form (for example) $\partial_k(\alpha^{ij})\partial_i(\alpha^{kl})\partial_j(f)\partial_l(g)$, and I do not understand why. This term would correspond to the following graph:


I.e., there is an edge going from 1 to 2 and an edge going from 2 to 1; all such terms would be of the form $\partial_j(\alpha^{i\cdot})\partial_i(\alpha^{j\cdot})\times\text{derivatives of $f$ and $g$}$, and none of them occur in his formula in section 1.4.2. As far as I can see, such graphs would be admissible under the rules on page 5 in the paper. I have yet to find a way to evaluate the weight associated to this kind of graphs, but at least the integrand of the weight for the graph above is nonzero. So my question is:

Why do terms of this form not contribute to the second order term of the star product?


Using a remark in this recent paper by Thomas Willwacher as a hint, I found out. The problem is solved by the following: there exists a gauge transformation $D$ that transform the extra term away. To see this, we first need the second order term of a gauge transformed star product: If $\star = I + \hbar^nB_n$ and $D = I + \hbar^nD_n$, and $f\star'g = D^{-1}(Df\star Dg)$ then $$ B_2'(f,g) = B_2(f,g)+B_1(f,D_1g)+B_1(D_1f,g)-D_1(B_1(f,g))+D_1(f)D_1(g)- D_1(g D_1f)-D_1(f D_1g)+D_1(D_1(f g))+g D_2f+fD_2g-D_2(f g), $$ where, of course, $B_2'$ is the second order term of $\star'$. Setting $D_1 = 0$ reduces this to $$ B_2'(f,g) = B_2(f,g)+g D_2f+f D_2g-D_2(f g) = B_2(f,g) + d^H(D_2)(f\otimes g) $$ where $d^H$ is the Hochschild differential. (Also, $B_1' = B_1$.)

Now comes the crucial part: the term in the question $\partial_k(\alpha^{ij})\partial_i(\alpha^{kl})\partial_j\otimes\partial_l =: A$ is $d^H$-exact: indeed, $A = d^H\left(-\frac12\partial_k(\alpha^{ij})\partial_i(\alpha^{kl})\partial_j\partial_l\right)$. So if we set $D_2$ to minus the expression in the brackets, then, $A$ no longer occurs in $B_2'$ while the other terms of $B_2$ occur unmodified in $B_2'$.

Finally, expressions of the form $\partial_k(\alpha^{ij})\partial_i(\alpha^{kl})f\partial_j\partial_l(g)$ (corresponding to the diagram in the question but with both downwards arrows now pointing to $g$) also cannot occur in $B_2$ because $B_n$ is zero for any $n$ when one of its arguments is constant, and such a term would violate this rule.

Thus, the star product in section 1.4.2 in Kontsevich's paper does not equal his general formula up to order 2; instead, they are equivalent.

(By the way, the weight of the term corresponding to the diagram is, including multiplicity from similar graphs, equal to $-\frac16$ according to this paper by Giuseppe Dito. In particular, it is nonzero.)


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.