Deformation Quantization, Formality, and Star Products

Deformation quantization replaces a commutative Poisson algebra by an associative algebra over formal power series in $\hbar$. The first antisymmetric correction must reproduce the Poisson bracket; associativity then imposes a hierarchy of cohomological equations. Kontsevich’s formality theorem solves this existence and classification problem for finite-dimensional Poisson manifolds. For QFT observables, however, distributional kernels meet coincident-point singularities, so the finite-dimensional formula applies directly only to a regular subalgebra and not to arbitrary local nonlinear functionals.

Required background. Classical Observables and Poisson Factorization supplies the Poisson algebra being deformed. L∞ Algebras, Formal Moduli, and Field Equations supplies the Maurer–Cartan mechanism behind formality. Helpful background. Locally Constant Factorization Algebras and Eₙ Algebras gives a different route from local multiplication to higher algebra. Renormalization Freedom and Stückelberg–Petermann Group explains why extending products to local functionals is nonunique.

Star products and their formal classification

Let $A=C^\infty(X)$ for a finite-dimensional smooth manifold $X$ with Poisson bivector $\pi$. A star product is an $\mathbb R[[\hbar]]$-bilinear associative product

$$f\star g=fg+\sum_{r\geq1}\hbar^r B_r(f,g),$$

where each $B_r$ is bidifferential, $1$ remains a unit, and

$$B_1(f,g)-B_1(g,f)=i\{f,g\}_\pi.$$

Two products are gauge equivalent if related by a formal differential operator $T=\mathrm{id}+\hbar T_1+\cdots$ through $T(f\star g)=Tf\star' Tg$. The word “gauge” here concerns equivalent associative deformations; it is not the gauge symmetry of a field theory.

Associativity at order $\hbar$ says that $B_1$ is a Hochschild cocycle. At order $\hbar^2$, the failure of $B_1$ to associate must be a Hochschild coboundary of $B_2$. The Jacobi identity for $\pi$ is the first obstruction equation. Kontsevich constructs an $L_\infty$ quasi-isomorphism from polyvector fields with the Schouten bracket to polydifferential operators with the Gerstenhaber bracket Kontsevich 2003, Formality Theorem 4.6.2, p. 173. Applying it to Maurer–Cartan elements identifies formal Poisson structures, modulo formal diffeomorphisms, with star products modulo the equivalence above.

The theorem establishes formal existence and classification. It neither says that the $\hbar$ series converges nor selects a $*$-representation, a positive state, or a Hilbert-space completion. The quasi-isomorphism is canonical only up to higher homotopy, and different choices can yield equivalent rather than identical products.

The constant-bivector calculation

For a constant Poisson tensor $\pi^{ij}$ on $\mathbb R^d$, define

$$f\star g =m\circ\exp\!\left( \frac{i\hbar}{2}\pi^{ij}\partial_i\otimes\partial_j \right)(f\otimes g),$$

where $m$ is ordinary multiplication. Through second order,

$$\begin{aligned} f\star g={}&fg +\frac{i\hbar}{2}\pi^{ij}(\partial_i f)(\partial_j g)\\ &-\frac{\hbar^2}{8}\pi^{ij}\pi^{kl} (\partial_i\partial_k f)(\partial_j\partial_l g) +O(\hbar^3). \end{aligned}$$

This is the Moyal product Kontsevich 2003, §1.4.1, pp. 159–160. Associativity follows because the constant insertion operators acting on three tensor factors commute; the two parenthesizations exponentiate the same sum of pairwise contractions. Antisymmetry of $\pi$ gives

$$[f,g]_\star=i\hbar\{f,g\}_\pi+O(\hbar^3).$$

The absence of an $\hbar^2$ term is an immediate independent check: even powers in the exponential are symmetric under exchanging $f$ and $g$.

First application: regular free-field observables

The algebraic construction complements Canonical Quantization: Algebra, Representation, and State. Let $P=\Box_g+m^2+\xi R$ be a Green-hyperbolic Klein–Gordon operator on a globally hyperbolic spacetime, and let $\Delta=\Delta_{\mathrm R}-\Delta_{\mathrm A}$ be its causal propagator. On regular polynomial functionals of a smooth field $\phi$, define

$$\Gamma_\Delta =\int_{M^2}\Delta(x,y) \frac{\delta}{\delta\phi(x)}\otimes \frac{\delta}{\delta\phi(y)},d\mu_xd\mu_y$$

and

$$F\star G =m\circ\exp\!\left(\frac{i\hbar}{2}\Gamma_\Delta\right) (F\otimes G).$$

“Regular” means that every functional derivative used here is a smooth compactly supported density on $M^n$. Under that hypothesis the pairing with the distribution $\Delta$ is defined, and polynomial degree makes the series finite term by term. For the linear observables

$$F_f(\phi)=\int_M f\phi\,d\mu, \qquad F_g(\phi)=\int_M g\phi\,d\mu,$$

all second derivatives vanish, so

$$F_f\star F_g=F_fF_g+\frac{i\hbar}{2}\langle f,\Delta g\rangle, \qquad [F_f,F_g]_\star=i\hbar\langle f,\Delta g\rangle.$$

This is precisely the canonical commutation relation in covariant form. For quadratic regular functionals, the displayed second-order term is a double contraction, and higher terms vanish after finitely many steps. Associativity is checked exactly as in the finite-dimensional constant case because $\Delta$ is independent of the background field $\phi$.

The construction produces a formal associative algebra. Choosing a Hadamard two-point function instead of the antisymmetric propagator gives a normally ordered product related by a formal intertwiner; positivity belongs to a later choice of state, not to the star product alone.

Failure test: local nonlinear functionals

Take $F(\phi)=\int f(x)\phi(x)^4,d\mu_x$. Its second functional derivative is supported on the diagonal $x=y$, and repeated applications of $\Gamma_\Delta$ attempt to restrict or multiply singular propagators at coincident points. The regular-functional hypothesis has failed. Already Costello’s finite- versus positive-dimensional comparison makes the issue explicit: the naive BV Laplacian and bracket on all functionals cease to be defined when distributional kernels meet the diagonal Costello 2007, §10.1, pp. 43–44.

Renormalized extension can define time-ordered or star products on larger microcausal and local classes, but it requires wavefront-set conditions, extension across diagonals, scaling bounds, and finite counterterm choices. Applying the Moyal exponential before those steps is the adversarial failure. A formula that is associative on regular functionals does not, by converse, prove associativity of an unrenormalized product on local interactions.

Exercises

Compute the star commutator of coordinate functions $x^i$ and $x^j$ for a constant Poisson tensor.

Solution

All second and higher derivatives vanish. Hence $x^i\star x^j=x^ix^j+\frac{i\hbar}{2}\pi^{ij}$ and $x^j\star x^i=x^ix^j-\frac{i\hbar}{2}\pi^{ij}$. Therefore $[x^i,x^j]_\star=i\hbar\pi^{ij}$ exactly.

Why does the Moyal series terminate for two polynomial functionals of degrees $r$ and $s$?

Solution

The $n$th term differentiates each factor $n$ times. It vanishes when $n>r$ or $n>s$. Thus only $n\leq\min(r,s)$ contributes. Termination does not cure coincident-point singularities for local functionals, because even a finite contraction can be distributionally undefined.

References

• Costello 2007, Renormalisation and the Batalin–Vilkovisky Formalism, arXiv:0706.1533 [math.QA]. Open PDF
• Kontsevich 2003, Deformation Quantization of Poisson Manifolds, Letters in Mathematical Physics 66(3), 157–216. Open PDF