Cauchy principal value
/ KOH-shee /
Some integrals are divergent in the ordinary sense — they blow up at a point inside the interval, like the integral of 1/x across zero — yet the infinities on the two sides are mirror images that ought to cancel. The Cauchy principal value is a careful way of taking that cancellation seriously: you cut out a tiny symmetric window around the trouble spot and shrink it evenly from both sides at once, capturing the finite answer that an honest limit reveals.
Precisely, for a singularity at c inside [a, b], the principal value is the limit, as epsilon goes to zero, of the integral over [a, c - epsilon] plus the integral over [c + epsilon, b] — the same epsilon on both sides. The symmetry is essential. If you let the two gaps shrink at different rates, you can get any answer you like, which is exactly why the plain integral diverges; the principal value singles out the one balanced, symmetric limit. The same idea handles an infinite range: the principal value of an integral over the whole line is the limit of the integral over [-R, R] as R grows.
Principal values are indispensable in physics and signal processing: the Hilbert transform, dispersion relations (Kramers-Kronig), and the Sokhotski-Plemelj formula all live on principal-value integrals, and they appear whenever a response function has a pole on the real axis. The honest warning is that a principal value is a weaker, conditional notion of integral — it exists by virtue of cancellation, so you must always say P.V. explicitly; a principal value is not the same as the integral converging, and treating it as one can hide a genuine divergence.
The integral of 1/x from -1 to 1 diverges, but its Cauchy principal value is 0: by symmetry the integral over [-1, -epsilon] (which is ln(epsilon)) exactly cancels the integral over [epsilon, 1] (which is -ln(epsilon)) for every epsilon.
Shrinking the gap symmetrically lets equal and opposite infinities cancel, yielding a finite value.
A principal value depends on the symmetric way the limit is taken. If you remove the singular point unevenly the answer changes, so P.V. must be stated; it is a regularization, not proof that the ordinary integral converges.