Inverse trigonometric functions are inverse functions of suitably restricted trigonometric functions. They recover an angle from a known sine, cosine, tangent, cotangent, secant, or cosecant. The most frequently used are arcsine, arccosine, and arctangent, written , , and . Because trigonometric functions are periodic and therefore not one-to-one on their full domains, defining their inverses requires selecting intervals on which each original function takes every relevant value exactly once.
These functions connect angle measurement with algebraic quantities. They appear in geometry, calculus, coordinate transformations, probability, and numerical computation. On the real line, their principal values describe particular angles; over the complex numbers, they extend to multivalued analytic functions whose branches are expressed through logarithms and square roots. Their definitions, identities, differentiation formulas, and computational methods all depend on handling these domain and branch choices consistently.
Definitions, notation, and principal values
An inverse exists as a single-valued function only when the original mapping is one-to-one. The sine function, for example, satisfies , so it cannot have an inverse on the entire real line. Restricting sine to makes it a bijection onto . Its inverse is the principal arcsine. Analogous restrictions define the other principal inverse trigonometric functions.
The following conventions are widely used:
| Function | Real domain | Principal range |
|---|---|---|
These definitions use radian measure, in which an angle is the corresponding arc length on a unit circle. Other angular units can be used, but the standard derivative and series formulas assume radians. For example, , equivalently .
The notation commonly means , not the reciprocal . The reciprocal is . This creates a contrast with notation such as , which means . The explicit names , , and avoid this ambiguity. The prefix “arc” refers to the circular arc whose angle has the specified trigonometric value.
Conventions are less uniform for inverse cotangent, secant, and cosecant. In particular, some definitions of arccotangent use a range centered on zero and introduce a discontinuity there, rather than taking values continuously in . Under the convention in the table,
The reciprocal-function conventions are
Identities involving these less common inverse functions must therefore be read together with their range conventions. A formula valid under one convention need not hold unchanged under another.
Geometric meaning and solutions of trigonometric equations
In a right triangle, the sine of an acute angle is the ratio of the opposite side to the hypotenuse, the cosine is the adjacent-side ratio, and the tangent is the ratio of opposite to adjacent sides. Inverse trigonometric functions reverse these relationships. If a right triangle has opposite side , adjacent side , and hypotenuse , its smaller acute angle is
This geometric interpretation initially applies to positive ratios and acute angles. The unit circle extends it to signed coordinates and angles in other quadrants.
On the unit circle, the point corresponding to an angle has coordinates . Thus selects the angle on the upper semicircle with horizontal coordinate , while selects the angle on the right semicircle with vertical coordinate . Arctangent selects the angle between and whose ray has a specified slope. A slope alone does not distinguish opposite directions, because .
A principal inverse value is not the complete solution of a trigonometric equation. For , writing , all real solutions of are
Equivalently,
For cosine and tangent, the corresponding formulas are
and
At endpoint values, some of these expressions describe the same solutions more than once.
Inverse functions therefore behave differently depending on the order of composition. The identity
holds throughout , but holds only for . For instance,
not . Outside the principal interval, the composition folds angles back into that interval. Similarly,
These folded graphs make the role of branch selection visible: the inverse returns the designated representative of an angle class, not necessarily the angle originally supplied.
Identities and branch-dependent transformations
Many identities follow from complementary angles, symmetry, and the Pythagorean theorem. On their real domains,
whereas
The principal complementary-angle identity is
Together with the range restrictions, these formulas reduce many evaluations to nonnegative arguments.
Composing ordinary trigonometric functions with inverse functions produces algebraic expressions:
and
The positive square roots are essential. The principal arcsine and arctangent lie where cosine is nonnegative or positive, respectively, while the principal arccosine lies where sine is nonnegative. Without those range facts, the Pythagorean identity would determine only an absolute value.
Useful conversions include
An analogous expression for arccosine requires a quadrant correction. For ,
but for , the angle belongs to the second quadrant, so must be added to the arctangent on the right. At , the value is . This is a typical example of an algebraically plausible identity failing when its range is ignored.
Arctangent addition has the same issue. Put . If , then
When , the sum is for positive , and for negative . Taking , for example, gives
rather than simply .
A particularly useful reciprocal identity is
It converts large arguments into small ones and plays an important role in numerical argument reduction. Such identities reflect angle addition modulo a period; the explicit multiples of restore the chosen principal values.
Derivatives, integrals, and endpoint behavior
The derivatives of inverse trigonometric functions follow from implicit differentiation. If , then , so
Because on the principal interval,
The same method gives
Under the continuous real convention for arccotangent, its derivative is .
For the reciprocal inverse functions, the absolute value in the derivative is necessary:
These expressions follow by differentiating and and using . Omitting the absolute value would produce the wrong sign on the negative part of the domain. For composite arguments, the chain rule supplies an additional factor; for example,
The same formulas identify elementary antiderivatives:
and
The first integrand measures circular geometry, while the second arises frequently after completing the square in a quadratic denominator. By integration by parts,
These show how inverse circular functions and logarithms naturally occur together in integration.
Arcsine and arccosine are continuous at their finite domain endpoints, but their derivative magnitudes grow without bound there. As ,
Thus the graphs have vertical tangents at the endpoints. The singularity is integrable:
Arctangent has no finite real singularities; its derivative approaches zero at large , and its horizontal asymptotes are .
Power series and the calculation of pi
The inverse functions admit power series that connect them with algebraic expansion and high-precision arithmetic. Expanding as a geometric series and integrating from to gives
The series also converges at . For , the error after a finite alternating sum is no larger in magnitude than the first omitted term. Its usefulness, however, depends strongly on the size of the argument.
At , the series yields
This expansion is associated with Madhava of Sangamagrama and the Kerala mathematical tradition, and with seventeenth-century work by James Gregory and Gottfried Wilhelm Leibniz. It is theoretically significant but converges slowly at this endpoint: the alternating-series error bound decreases only on the order of the reciprocal of the number of terms.
Argument reduction produces much faster formulas. In 1706, John Machin used the identity
to calculate 100 decimal places of pi. Because the arguments are small, successive terms in the two arctangent series decrease rapidly. The identity can be verified by repeated tangent addition, together with a range check that identifies the correct multiple of . Later “Machin-like” formulas use other integer combinations of arctangents of rational numbers.
Arcsine has the expansion
It follows by applying the binomial theorem to and integrating term by term. Its radius of convergence is , and it converges at both endpoints. Arccosine is then obtained from .
The restriction on the radius of convergence is explained by singularities in the complex plane. Arcsine has branch points at . Arctangent, although smooth for every real argument, has complex singularities at , each at distance from the origin. A Taylor series about zero cannot pass through those singularities. For large positive real , reciprocal reduction instead gives the convergent expansion
Complex extensions and logarithmic branches
Over the complex numbers, inverse trigonometric functions are naturally multivalued. This follows both from the periodicity of sine and cosine and from the multivalued nature of the complex logarithm. A single-valued analytic version requires choosing branches and removing suitable cuts from the plane. The resulting functions agree with the usual real principal values wherever the real definitions apply.
Using Euler’s formula,
If and , then
so
Taking logarithms yields the inverse relation. With compatible principal branches,
A standard principal branch is analytic on
Its continuous real values at may be included separately. Principal arccosine is defined compatibly by
The endpoints are branch points of square-root type. Near , for example, the leading behavior of arccosine is proportional to . Continuing an inverse value around a branch point can move it to a different sheet. The full inverse relation for sine includes
where is one selected inverse value. These are not merely different notations for the same analytic branch; they describe distinct values of the inverse relation.
Arctangent has a logarithmic representation
With principal logarithms, its cuts run along the imaginary axis from upward and from downward. Its branch points are and , consistent with
The derivative has simple poles at those points, and integration produces logarithmic branching.
In complex analysis, identities involving these functions hold on specified branches or after analytic continuation, not automatically across the entire plane. In particular, replacing a difference of principal logarithms by the principal logarithm of a quotient can change a value by an integer multiple of . Correspondingly, real inverse identities that appear purely algebraic can acquire additive multiples of or change branches when extended to complex arguments.
Two-argument arctangent and directional angles
The one-variable expression cannot by itself recover the direction of a nonzero point . It is undefined when , and it assigns the same ratio to and . The two-argument arctangent, conventionally written , uses both coordinates to determine the quadrant. It is closely related to, but not simply identical with, the inverse of the tangent function.
For real coordinates other than , a common range is . Away from boundary conventions on the axes,
Thus , whereas , even though both coordinate ratios equal . Some applications instead normalize the result to .
This function provides the angular coordinate in polar coordinates:
It also gives the principal argument of a nonzero complex number . Its discontinuity along a chosen ray is unavoidable: no continuous real-valued choice of angle exists on the entire punctured plane. Traversing a full circle changes a continuously tracked angle by , although the starting and ending points coincide. A principal angle therefore introduces a cut, while angle-tracking systems may “unwrap” successive values by adding multiples of .
Locally, away from the origin and the selected cut,
Consequently, the gradient is
Its magnitude is , showing that angular estimates become sensitive to coordinate errors near the origin. At the origin itself, direction is geometrically undefined, even if a software library returns a prescribed value for particular signed-zero inputs.
For nonzero vectors and in three-dimensional Euclidean space, their unsigned angle can be written
Here the cross product supplies the sine-related quantity, while the dot product supplies the cosine-related quantity. In two dimensions, replacing the cross-product magnitude by the signed determinant gives an oriented angle in a principal interval.
Numerical evaluation and conditioning
Practical implementations use different formulas on different argument intervals. Directly summing a Taylor series is efficient only when its argument is sufficiently small. Arctangent is therefore commonly reduced through symmetry, reciprocal identities, and tangent-subtraction formulas. If is a conveniently chosen constant, then, with the appropriate branch correction,
Choosing near makes the new argument small. A low-degree polynomial or rational approximation can then provide high accuracy over a short interval.
The CORDIC algorithm, introduced by Jack Volder in 1959, offers a different approach. It resolves a rotation into a sequence of smaller rotations with angles . In vectoring mode, a coordinate pair is rotated toward an axis while these angles are accumulated, yielding a direction estimate. The method is attractive in hardware because its basic iterations use shifts and additions rather than general multiplication or division. Polynomial and rational approximations are common in software, while CORDIC remains important in specialized digital implementations.
The input sensitivity of arcsine and arccosine follows from their derivatives. For a small perturbation ,
Near , even a well-designed algorithm cannot prevent input uncertainty from being magnified. This is a property of the mathematical problem, described by its conditioning, rather than necessarily a defect in the algorithm. Numerical stability concerns how closely the implementation respects that underlying sensitivity.
Equivalent formulas can nevertheless have different rounding behavior. For real ,
Factoring as can avoid losing information when a rounded square is subtracted from . Implementations may also use the half-angle identity
especially for near . This avoids subtracting two nearly equal angles in .
In floating-point arithmetic, mathematically equivalent expressions need not give identical results. An angle obtained from a normalized dot product may encounter a computed cosine slightly outside , even when exact arithmetic would keep it inside. Robust implementations account for rounding, scaling, overflow, underflow, infinities, and signed zeros. Computing directly also avoids explicitly forming , which may overflow or underflow although the direction is well defined.
Applications in geometry, probability, and physical models
Inverse trigonometric functions recover geometrical information from measured lengths and coordinates. For a triangle with side lengths , the law of cosines gives the angle opposite :
The principal range matches the possible interior angles of a nondegenerate triangle. By contrast, recovering an angle from its sine can leave two possibilities, and . This is the source of the ambiguous case in triangle reconstruction from two sides and an angle opposite one of them.
In three-dimensional geometry and robotics, inverse functions convert coordinates or vector relationships into orientation parameters. If a point has distance from the origin and vertical coordinate , its polar angle from the positive vertical axis is . Its azimuth is obtained from . In a planar two-link mechanism with link lengths , a target position satisfies
The inverse cosine determines the magnitude of a joint angle, while different signs and subsequent directional calculations distinguish the two usual configurations. An argument outside indicates that the target lies outside the ideal mechanism’s reachable region.
In probability, arctangent appears in the cumulative distribution function of the standard Cauchy distribution:
Its inverse is
Arcsine similarly describes the distribution of a circular coordinate. If is uniform on and , then
and
The larger density near the endpoints reflects the fact that sine changes slowly near its extrema.
In oscillatory models, a sinusoidal signal’s amplitude alone does not determine its phase: the same amplitude occurs at multiple points in a cycle. Two quadrature components, proportional to sine and cosine, allow phase recovery with . This principle appears in signal processing and measurements based on rotating or oscillating quantities. A first-order response with complex denominator , for example, has phase
for nonnegative angular frequency and positive time constant . The inverse function translates a ratio of imaginary and real components into an angle, while its branch convention fixes the reported phase interval.