Why position alone cannot describe a singularity
Objective: Why is the pair (x_0,ξ_0) more informative than singular support?
At one point, two distributions may be singular in different directions. A vertical jump edge and a horizontal jump edge share a crossing point but react differently to rapidly oscillating probes. Microlocal analysis separates position x from nonzero covector direction ξ. Before using a formula, identify the open set or manifold, the distribution or Sobolev space, the Fourier convention, the conic set, the operator order, and whether a conclusion is local, microlocal, or global.
Read the notation from this chapter's definition rather than treating symbols as decoration: For x∈R^n and ξ∈R^n∖{0}, the pair (x,ξ) records a base position and a nonzero frequency covector. Scaling ξ by t>0 keeps its direction, so microlocal singular sets are conic in ξ. The exact result established here is: Multiplying by a cutoff localizes position, while decay of the Fourier transform of that localized distribution tests directional smoothness; neither operation alone records both pieces of information.
The proof has a checkable route: 1. Choose χ∈C_c^∞ equal to one near the base point x_0, so χu forgets behavior far from x_0. 2. Compute or estimate 𝓕(χu)(ξ). Rapid decay in an open cone Γ around ξ_0 means repeated integration by parts is available in those directions. 3. Changing χ while retaining χ=1 near x_0 changes only a term supported away from x_0; convolution estimates preserve rapid decay in a slightly smaller cone. Every estimate must state which constants are uniform, which variables are large, and which lower-order or smoothing remainder is permitted.
Reconstruct the concrete calculation “Let u(x_1,x_2)=H(x_1), choose χ(x)=χ_1(x_1)χ_2(x_2) near the origin, and identify which frequency component can fail to decay rapidly.” before generalizing it. Then test the nearest failure: The zero covector is excluded because it contains no direction and compact-frequency behavior does not detect differentiability. A large value of one Fourier coefficient also does not by itself prove singularity; failure of rapid decay throughout every suitable cone is required. This separates a theorem from a formal manipulation.
Multiplying by a cutoff localizes position, while decay of the Fourier transform of that localized distribution tests directional smoothness; neither operation alone records both pieces of information.
Choose χ∈C_c^∞ equal to one near the base point x_0, so χu forgets behavior far from x_0.
Compute or estimate 𝓕(χu)(ξ). Rapid decay in an open cone Γ around ξ_0 means repeated integration by parts is available in those directions.
Changing χ while retaining χ=1 near x_0 changes only a term supported away from x_0; convolution estimates preserve rapid decay in a slightly smaller cone.
Let u(x_1,x_2)=H(x_1), choose χ(x)=χ_1(x_1)χ_2(x_2) near the origin, and identify which frequency component can fail to decay rapidly.
- The localized transform factors as 𝓕(χ_1H)(ξ_1)𝓕(χ_2)(ξ_2).
- The smooth compactly supported factor χ_2 has a rapidly decreasing transform in ξ_2.
- The jump in x_1 produces only algebraic decay in ξ_1, so singular covectors are normal to the line x_1=0 and have ξ_2=0.
Result: The possible singular directions at the origin are nonzero multiples of dx_1, not arbitrary nonzero covectors.