Distributions 15 | Support for Distributions

Support for distributions generalizes the familiar idea of where an ordinary function is nonzero, but it’s built using test functions rather than pointwise values. For a usual function f on ℝ^N, the support is the closure of the set where f(x) ≠ 0; equivalently, its complement is the largest open region where f vanishes identically. Distributions don’t allow evaluation at single points in any meaningful way, so the definition shifts from pointwise behavior to local behavior tested against smooth, compactly supported functions.

A distribution T is a linear functional acting on test functions. To say that T “vanishes on” an open set U, the transcript uses the rule: T(f) = 0 for every test function f whose support lies entirely inside U. This captures the idea that T contributes nothing when probed by test functions concentrated in that region. With this notion, the support of a distribution should again be the complement of the largest open set where T vanishes.

The delta distribution provides the guiding example. For the Dirac delta δ, applying it to a test function returns the value of that test function at the origin: δ(f) = f(0). That means δ is zero on any open set that does not contain 0, since test functions supported away from the origin evaluate to 0 under δ. The largest open set where δ vanishes is therefore ℝ^N \\ {0}, so the support of δ is exactly {0}.

For a general distribution T, the key step is showing that a maximal open set U_max exists such that T vanishes on U_max, and that the complement of U_max is a well-defined closed set called supp(T). The transcript constructs U_max by collecting all open sets U on which T vanishes and taking their union. The only nontrivial issue is whether T also vanishes on the union itself, not just on each piece.

To prove this, the argument starts with an arbitrary test function f whose support is contained in U_max. Since the support of f is compact, it can be covered by the open sets from the union; compactness then guarantees a finite subcover. That reduces the situation to a finite union of open sets U_1, …, U_m, each of which already has the property that T vanishes on it.

At this point, a partition of unity enters. For a compact set covered by finitely many open sets, one can choose test functions φ_1, …, φ_m supported in U_1, …, U_m such that their sum equals 1 on the support of f. This lets f be decomposed into a finite sum of test functions, each supported in one of the U_i. By linearity, T(f) becomes a sum of terms T(φ_i f), and each term is zero because φ_i f has support inside an open set where T vanishes. Therefore T(f)=0 for every test function supported in U_max, proving that T vanishes on U_max.

With that missing piece filled, the support of any distribution is well defined as the complement of U_max. The transcript closes by noting that this concept is needed to extend convolution to distributions in the next installment.