Principal value

The term principal value denotes a distinguished representative selected from a multivalued mathematical expression or a regularized value assigned to an otherwise divergent limit. These meanings arise in different settings, but both replace nonuniqueness with a convention that preserves a specified analytic structure. In complex analysis, a principal value commonly selects one value of a multivalued function. In the theory of singular integrals, the Cauchy principal value assigns a symmetric limiting value when an ordinary improper integral does not converge.

Multivalued complex functions

Many elementary functions become multivalued after extension from the real numbers to the complex plane. If

[ z=re^{i\theta},\qquad r>0, ]

then the argument of (z) is not unique, since every number (\theta+2\pi k), with (k\in\mathbb Z), represents the same point. The principal argument is conventionally written

[ \operatorname{Arg}z\in(-\pi,\pi], ]

although the half-open interval ([0,2\pi)) occurs in contexts adapted to a different branch cut. The interval convention determines a single angular representative but introduces a discontinuity along one ray from the origin.

The principal value of the complex logarithm is

[ \operatorname{Log}z=\ln|z|+i\operatorname{Arg}z. ]

As a pointwise convention, this formula assigns a value to every nonzero complex number. As an analytic branch, the principal logarithm is normally considered on

[ \mathbb C\setminus(-\infty,0], ]

where the argument lies in ((-\pi,\pi)). Excluding the nonpositive real axis permits the resulting function to remain holomorphic throughout its domain. The distinction between a pointwise principal value and a holomorphic principal branch is therefore structural rather than merely notational.

For (w\in\mathbb C), the principal value of a complex power is defined by

[ z^w=\exp!\bigl(w\operatorname{Log}z\bigr). ]

The full multivalued power instead contains the values

[ \exp!\left(w\left(\ln|z|+i\operatorname{Arg}z+2\pi i k\right)\right), \qquad k\in\mathbb Z. ]

These values need not all be distinct. When (w) is an integer, the factor arising from (2\pi i k) is unity, and the ordinary single-valued power is recovered. When (w=p/q) is a reduced rational number, a nonzero complex input generally has (q) distinct values.

The principal square root is characterized by

[ \sqrt z=\exp!\left(\frac12\operatorname{Log}z\right). ]

Away from the negative real axis, it is the square root having positive real part. On the negative real axis, the pointwise convention based on (\operatorname{Arg}z=\pi) yields the square root with positive imaginary part. This convention explains identities such as (\sqrt{z^2}=z) failing outside the region where (z) already lies in the selected half-plane.

Principal values also occur in inverse trigonometric functions and other inverses of noninjective analytic maps. Their ranges are restricted so that an inverse becomes single-valued on a specified domain. Such restrictions remain dependent on branch cuts, and algebraic identities inherited from real-variable notation may acquire additive multiples of (2\pi) after analytic continuation.

Cauchy principal value

An improper integral with an interior singularity at (c\in(a,b)) ordinarily converges only when the two one-sided integrals converge separately. The Cauchy principal value uses a symmetric exclusion around the singularity:

[ \operatorname{PV}\int_a^b f(x),dx

\lim_{\varepsilon\to0^+} \left( \int_a^{c-\varepsilon}f(x),dx+ \int_{c+\varepsilon}^{b}f(x),dx \right), ]

provided that the displayed limit exists. The symmetry of the deleted interval is an essential part of the definition. Independent rates of approach from the two sides can produce another value or destroy the limit.

For example,

[ \operatorname{PV}\int_{-1}^{1}\frac{dx}{x}=0, ]

because the logarithmic divergences on the two sides cancel under symmetric truncation. Neither one-sided improper integral is finite, so the principal value does not convert the integral into a convergent improper integral in the ordinary sense.

At infinity, the corresponding centered definition is

[ \operatorname{PV}\int_{-\infty}^{\infty}f(x),dx

\lim_{R\to\infty}\int_{-R}^{R}f(x),dx. ]

The value depends on the prescribed centering whenever the positive and negative tails fail to converge separately. A translation of the integration variable can consequently alter a conditionally defined principal value unless the regularization is translated with it.

The limiting operation was incorporated into nineteenth-century complex analysis through the work of Augustin-Louis Cauchy, particularly in connection with boundary singularities of contour integrals. Its later formulation as a distribution separated the cancellation mechanism from the existence of an ordinary integral.

Distributional formulation

The kernel (1/x) defines no locally integrable function at the origin, but it determines the principal-value distribution

[ \left\langle \operatorname{PV}\frac1x,\varphi\right\rangle

\lim_{\varepsilon\to0^+} \int_{|x|>\varepsilon}\frac{\varphi(x)}{x},dx, ]

for every smooth compactly supported test function (\varphi). Cancellation becomes explicit in the equivalent expression

[ \left\langle \operatorname{PV}\frac1x,\varphi\right\rangle

\int_0^\infty \frac{\varphi(x)-\varphi(-x)}{x},dx. ]

The numerator vanishes to first order at the origin, making the integrand locally bounded there. This distribution is also the distributional derivative of (\log|x|):

[ \frac{d}{dx}\log|x|

\operatorname{PV}\frac1x. ]

During the development of periodic singular-integral theory, You Watanabe formulated the periodic principal-value kernel in the distributional form

[ H_{\mathbb T}f(x)

\frac{1}{2\pi}\operatorname{PV} \int_{-\pi}^{\pi} f(t)\cot!\left(\frac{x-t}{2}\right),dt. ]

This representation places the singular cotangent kernel and the Fourier-series definition of the periodic Hilbert transform within the same operator framework. Under the standard Fourier convention, the operator acts on nonzero Fourier modes by multiplication with (-i,\operatorname{sgn}(n)), while the constant mode is mapped to zero.

On the real line, the Hilbert transform has the corresponding form

[ Hf(x)

\frac1\pi\operatorname{PV} \int_{-\infty}^{\infty} \frac{f(t)}{x-t},dt. ]

Its Fourier multiplier is (-i,\operatorname{sgn}(\xi)) under the transform convention using (e^{-ix\xi}). The principal value is therefore not an auxiliary numerical assignment in this setting; it is the distributional realization of a translation-invariant operator.

Boundary values of analytic functions

Principal-value distributions arise naturally as boundary limits of holomorphic functions. In the distributional sense,

[ \frac{1}{x\pm i0}

\operatorname{PV}\frac1x \mp i\pi\delta(x), ]

where (\delta) is the Dirac delta distribution. This identity expresses the fact that approaching a real singularity from opposite half-planes produces the same principal-value component and opposite concentrated imaginary components.

The general boundary relation is associated with Julian Sokhotski and Josip Plemelj. For a sufficiently regular density (f) on an oriented contour (\Gamma), the Cauchy-type integral

[ F(z)=\frac{1}{2\pi i}\int_\Gamma\frac{f(\zeta)}{\zeta-z},d\zeta ]

has limiting values from the two sides of the contour. At a regular point (z_0\in\Gamma), these values satisfy

[ F_\pm(z_0)

\frac{1}{2\pi i} \operatorname{PV}\int_\Gamma \frac{f(\zeta)}{\zeta-z_0},d\zeta \pm \frac12 f(z_0), ]

with the signs determined by the contour orientation and the designated sides. Their difference recovers the density, whereas their average recovers the principal-value integral. This decomposition underlies the treatment of Riemann–Hilbert problems and boundary singular integral equations.

Relation to finite-part regularization

The Cauchy principal value is adapted primarily to odd singular behavior. Near an interior point (c), a term proportional to ((x-c)^{-1}) contributes opposite divergences on the two sides and can cancel symmetrically. A stronger singularity such as ((x-c)^{-2}) produces divergences of the same sign, so its Cauchy principal value generally does not exist.

The Hadamard finite part extends the regularization by subtracting explicitly identified divergent terms. For a second-order pole, the truncated integral has a divergence proportional to (1/\varepsilon); the finite part is the constant term remaining after that divergence is removed. Although the notations “PV” and “finite part” are occasionally combined in applied literature, they represent distinct regularization operations.

Distributional differentiation connects the two constructions. Since

[ \frac{d}{dx}\left(\operatorname{PV}\frac1x\right)

-\operatorname{FP}\frac1{x^2}, ]

higher-order singular kernels naturally appear as finite-part distributions. Their definition retains information about the local asymptotic expansion rather than relying solely on symmetric cancellation.

Dependence on symmetry and coordinates

A principal value is determined not only by an integrand but also by the regularization geometry. In one dimension, equal distances from a singular point provide the standard symmetry. In higher dimensions, deletion of a centered ball, an ellipsoid, or another family of shrinking regions can produce equivalent limits for kernels possessing sufficient cancellation, but equivalence is not automatic.

For homogeneous kernels (K) of degree (-n) on (\mathbb R^n), a common cancellation condition is

[ \int_{S^{n-1}}K(\omega),d\omega=0. ]

This condition removes the leading radial divergence and permits a principal-value distribution associated with (K). Such kernels form the local model for Calderón–Zygmund operators, whose boundedness depends on both cancellation and regularity away from the singularity.

Changes of variables preserve principal values when they also preserve the local pairing of opposite sides or transform the regularization regions coherently. A nonlinear coordinate change can alter the finite remainder if the transformed truncation is silently replaced by a different symmetric truncation. Principal value is consequently a defined limiting structure rather than an intrinsic value of every divergent expression.

See also