The sine integral is the special function conventionally defined by
The apparent singularity of the integrand at is removable: its value there is defined by continuity as . The function describes the accumulated contribution of a decaying oscillation. It is a standard example of an antiderivative that cannot be expressed using elementary functions, and it occurs naturally in Fourier transforms, wave propagation, and the analysis of band-limited signals.
Definition and conventions
For real , the defining integral is an ordinary definite integral once the integrand has been extended continuously at zero. It satisfies
Thus the sine integral is an odd function.
A related convention uses the lowercase notation
Unlike , this shifted function is not odd and tends to zero as . Distinguishing the two conventions is essential when comparing formulas or software documentation.
The integrand is the unnormalized sinc function,
Another common convention defines sinc as , which introduces corresponding factors of in integral formulas.
Derivatives and local behavior
By the fundamental theorem of calculus,
Integrating the Taylor series of term by term gives
In particular,
Consequently, is approximately near the origin.
For , differentiating again yields
On the positive real axis, stationary points occur at , where is a positive integer. Odd multiples of are local maxima, and even multiples are local minima. The function therefore oscillates around its limiting value rather than approaching it monotonically. Its first positive maximum is
An elementary indefinite-integration identity is
The Dirichlet integral and limiting values
The fundamental improper-integral identity associated with the sine integral is the Dirichlet integral:
It follows that
The improper integral converges conditionally, not absolutely:
Cancellation between positive and negative oscillations is therefore essential.
One derivation introduces exponential damping. For , let
Differentiation under the integral sign gives
Together with as , this yields
Taking , with a suitable justification relating the damped and undamped integrals, gives . This identity also connects the sine integral with the Laplace transform.
Large-argument asymptotics
For large positive , repeated integration by parts applied to
produces the asymptotic expansion
The leading terms are
where the remainder uses big-O notation.
The oscillations therefore have an envelope of order . The factorial coefficients also show that the full inverse-power expansions are divergent for fixed nonzero . They are useful as asymptotic approximations when truncated appropriately, not as convergent infinite series. Negative real arguments are handled using oddness.
Complex arguments
For a complex number , the same definition extends to
After filling in the removable singularity at zero, the integrand is an entire analytic function. The integral is consequently independent of the path joining to , and is itself entire.
The power series
converges for every complex . It also makes the identities
immediate.
Large-complex-argument asymptotics require attention to direction in the complex plane. The real-axis limiting values do not describe behavior in every direction; along imaginary directions, exponential growth occurs.
Fourier analysis and the Gibbs phenomenon
The sine integral arises whenever a sinc-shaped kernel is integrated over an interval. For example, under the inverse Fourier-transform convention
an ideal low-pass frequency response equal to for and outside that interval has impulse response
Its response to a unit step is
This follows from integrating the impulse response, equivalently from its convolution with the step.
The oscillation of explains the overshoot near the discontinuity. At ,
The first peak therefore exceeds the final level by about of the step height.
The same sine-integral profile governs the local limiting behavior of Fourier-series approximations near a jump discontinuity. This is the Gibbs phenomenon: increasing bandwidth or the number of retained Fourier modes narrows the oscillatory region, but does not eliminate its limiting relative overshoot.
Numerical evaluation
Different representations suit different argument ranges:
- Small arguments: the convergent power series is efficient and avoids evaluating the removable singularity directly.
- Moderate arguments: numerical quadrature or rational approximations can give accurate values.
- Large positive arguments: truncated asymptotic expansions efficiently approximate the deviation from .
- Negative real arguments: odd symmetry reduces evaluation to the positive case.
Although the power series converges everywhere, it can be numerically unsuitable at large arguments because large alternating terms cancel. In floating-point arithmetic, that cancellation may destroy significant digits.
A related issue arises when computing the small tail
for large positive . Direct subtraction of nearly equal quantities can lose relative accuracy. Algorithms intended to return the tail accurately can evaluate its asymptotic representation directly instead.