the principal value
/ Cauchy: koh-SHEE /
Some integrals look hopeless because the integrand blows up in the middle of the range. The function 1/x is the standard culprit: near the origin it shoots to plus and minus infinity, so the integral of 1/x times a probe seems undefined. Yet the trouble is partly an illusion — the part just left of zero is large and negative while the part just right is large and positive, and they almost cancel. The principal value is the way of taking that cancellation seriously to extract a finite, well-defined answer.
Precisely, the Cauchy principal value cuts out a symmetric little gap of width epsilon around the singular point, integrates over what remains, and then lets epsilon shrink to zero. For 1/x it means integrating over |x| > epsilon and taking the limit; because the symmetric cut treats the two sides evenly, the divergent halves cancel and the limit exists. This defines a distribution, written p.v. 1/x, whose action on a probe is the limit of the integral of phi(x)/x over |x| > epsilon. It is a genuine distribution, but emphatically not the function 1/x — it is the regularized, cancellation-respecting version of it.
The principal value matters because objects like 1/x appear constantly — in the Hilbert transform, in Fourier analysis, in fundamental solutions — and the naive integral does not converge. The principal value gives these a rigorous distributional meaning, and it sits in a precise relationship with the delta through the Sokhotski-Plemelj formula, which decomposes 1/(x minus i0) into the principal value of 1/x plus a multiple of delta.
The principal value of the integral of 1/x from -1 to 1 is 0: the contribution from -1 to -epsilon is the negative mirror image of the contribution from epsilon to 1, so they cancel exactly for every epsilon, and the limit is 0 — even though both halves diverge on their own.
Symmetric trimming around the singularity lets the diverging halves cancel into a finite value.
The principal value depends on the gap being symmetric. Trim a lopsided gap and the limit can be anything you like — so p.v. 1/x is a definite distribution only because of the even-handed cut, not a property of 1/x alone.