Symbol classes and pseudodifferential operators

Use the class S^m_(1,0) of smooth families a(x,ξ;h) whose mixed derivatives satisfy the displayed estimates for every pair of multi-indices, with constants independent of h in (0,1]. The bracket ⟨ξ⟩ means sqrt(1+|ξ|²). Each momentum derivative lowers the allowed order by one, while position derivatives preserve it. We work with globally controlled position dependence, or specify a fixed compact position support; a support that grows with one over h would require additional justification.

The order m measures high-momentum growth, not a power of h. A family can have bounded values and still fail the class through derivatives growing as h decreases. Conversely an unbounded momentum polynomial can be a legitimate positive-order symbol. The principal symbol is the leading coefficient of a stated asymptotic expansion in h, not the numerical largest term at one phase-space point. Remainder estimates must specify their symbol order and the seminorms in which the powers of h are uniform.

A standard quantization uses a(x,ξ;h) in an integral kernel with exp(i(x−y)·ξ/h). The next lesson adopts the symmetric Weyl midpoint convention. Compactly supported smooth phase-space symbols give an immediate oscillatory construction, while general symbol orders use regularization and the Schwartz-space distribution framework. The notation a(x,hD) by itself does not authorize choosing arbitrary orders for noncommuting position and momentum factors; that choice belongs to the quantization convention.

Uniform order-zero symbols in the stated class yield uniformly bounded L² operators through the Calderón–Vaillancourt theorem, controlled by finitely many derivative seminorms. The theorem is sufficient, not an equivalence: failure of one class hypothesis does not prove the operator is unbounded. Positive-order differential operators instead act naturally between semiclassical Sobolev spaces or on specified dense domains. Treating every smooth symbol as an everywhere-defined bounded operator would erase precisely the domain distinctions introduced in operator theory.

More flexible classes allow derivatives to cost h^(−δ) per differentiation. When both position and momentum derivatives lose this factor, the usual product expansion gains a small parameter h^(1−2δ), so δ<1/2 is the standard useful range for that calculus. This is a different class with a different remainder scale. One should not import its results into the uniform class without changing the hypotheses, nor claim that every rapidly oscillating family is covered simply because it is smooth for each fixed h.

Fourier multipliers provide an exact calibration. A symbol depending only on ξ quantizes as multiplication by that symbol after the semiclassical Fourier transform, independent of the position-ordering convention. Its L² operator norm is its essential supremum. Negative order describes suppression of high semiclassical momentum, while the numerical location of an ordinary frequency is rescaled by h. This elementary case lets us test symbol estimates without obscuring them behind a general oscillatory-integral proof.