Integration by substitution is a method in calculus that transforms an integral by replacing its variable with a suitable function of another variable. It reverses the chain rule for differentiation: a composite expression together with the derivative of its inner function can be integrated by treating the inner function as a new variable. The method applies to both indefinite and definite integrals and extends, through change-of-variables theorems, to integration in several dimensions.
Mathematical basis
Suppose is a continuous function on an interval , and is continuously differentiable on an interval , with . If is an antiderivative of , so that , the chain rule gives
Consequently,
Introducing
expresses the same rule as
The right-hand side is evaluated in , after which is substituted back. The constant represents an arbitrary additive constant on the interval under consideration.
The notation , associated with the differential notation of Gottfried Wilhelm Leibniz, records the derivative factor required by the chain rule. In elementary calculus, the justification is the differentiation identity above, rather than treating and as ordinary numbers.
Connection with definite integration
Under the same continuity and differentiability assumptions, the fundamental theorem of calculus gives
Indeed, both sides equal . This formula shows that a substitution changes not only the integrand and differential but also the endpoints.
These assumptions are convenient sufficient conditions. More general substitution theorems exist for the Riemann integral and the Lebesgue integral, but their hypotheses must be stated separately.
Choosing and carrying out a substitution
A useful substitution usually simplifies function composition, removes a troublesome expression, or changes the integrand into a known form. Typical candidates include:
- an expression inside a power, root, exponential, or trigonometric function;
- a denominator whose derivative occurs elsewhere in the integrand;
- a transformation that removes a square root or exploits symmetry;
- a rescaling or translation that reduces an expression to a standard form.
For an indefinite integral, the usual procedure is to choose , compute , rewrite the entire integral in , integrate, and then return to . For a definite integral, the endpoints can instead be converted to -values and the calculation completed entirely in the new variable.
A composite function and its derivative
Consider
With and ,
The factor is what makes the substitution directly useful. A visually complicated inner expression is not, by itself, enough to guarantee a simpler integral.
A derivative present up to a constant
For
take . Since ,
A constant mismatch between the differential and the original integrand is handled by a compensating factor.
Logarithmic form
If is continuously differentiable and nonzero on an interval, then
This follows by taking and using the antiderivative of . The absolute value allows the real natural logarithm to handle either sign of , provided the interval does not cross a zero of . Different connected intervals may have different constants of integration.
Definite integrals, orientation, and invertibility
For example,
becomes, under ,
Here the transformed endpoints happen to have the same numerical values as the original endpoints. In general, they do not:
Decreasing substitutions
The order of the transformed endpoints records orientation. With ,
and therefore
The negative derivative and the reversal of limits are consistent parts of the same transformation.
Why one-to-one behavior is not always required
The identity
does not require to be an injective function. It follows directly from an antiderivative and the chain rule. If reverses direction, the signed contributions from retraced intervals cancel.
For instance,
This does not mean that every integral involving can be transformed over without further care. When the calculation requires an inverse function, a non-injective transformation may need to be restricted to separate monotone branches.
The distinction is between an oriented substitution, which carries the signed derivative, and a transformation of lengths, areas, or densities, which generally uses an absolute derivative and accounts for multiple branches.
Substitutions that introduce a new parameter
A substitution can also be written as . The rule then takes the form
with an appropriate inverse relation or interval restriction when expressing the result back in .
For example, for , set with . Then , and
Thus,
The restriction ensures that , rather than .
Trigonometric substitution
Identities among trigonometric functions make certain substitutions useful for square roots of quadratic expressions. For , standard choices include
Intervals must be chosen carefully to determine signs and inverse branches.
For example, setting , with , gives
Hence, on ,
The substitution removes the radical, although a trigonometric identity is still needed to complete the integration.
Improper integrals and convergence
For an improper integral, substitution is justified through the limit defining the integral. It is first performed on a truncated proper integral, and then the endpoint limit is taken.
For , the substitution gives
The infinite interval becomes a finite interval with a potentially singular endpoint.
A formal change of variable does not establish convergence by itself. Interior singularities must also be handled separately. Cancellation across a singularity can produce a Cauchy principal value even when the ordinary improper integral does not exist; these are different notions of integration.
Higher-dimensional change of variables
In several dimensions, the derivative factor is replaced by a determinant built from the Jacobian matrix.
Let be open sets, and let be a continuously differentiable bijection with a continuously differentiable inverse. For an integrable function , the change-of-variables theorem gives
The absolute Jacobian determinant describes local volume scaling. Its absolute value appears because ordinary volume is nonnegative, unlike the oriented differential in the one-dimensional chain-rule formula.
For polar coordinates,
and
Consequently,
The parameter range must avoid unintended multiple coverage. Boundary identifications and the degeneracy at can be handled by excluding sets of measure zero or using an appropriate version of the theorem.
Applications and limitations
Transforming probability densities
Substitution underlies the transformation rule for a probability density function. If a random variable has density , and for a continuously differentiable, strictly monotone function with nonzero derivative on the relevant interval, then
For a non-injective transformation, contributions from the inverse branches must be added. Under suitable regularity conditions,
at values where the inverse branches are regular. The absolute derivatives prevent negative densities, while the sum accounts for distinct original values mapped to the same outcome.
Differential equations and numerical integration
Substitution appears in separating variables in a differential equation, although division by a function of the unknown can exclude solutions where that function vanishes.
In numerical analysis, changes of variable can map an infinite interval to a finite one, reduce endpoint singularities, or redistribute resolution. The transformed integrand may also become more singular or vary more rapidly, so a formally correct substitution is not necessarily numerically advantageous.
Common errors and limits of the method
Frequent errors include omitting the derivative factor, keeping old endpoints after changing variables, mixing old and new variables in the transformed integrand, and ignoring signs or branch restrictions.
Substitution also does not guarantee an elementary antiderivative. Transforming
with on yields
which does not turn the problem into an elementary integral. A successful substitution must simplify the complete transformed expression, not merely rename part of it.
The method may be combined with algebraic manipulation, trigonometric identities, or integration by parts. A proposed antiderivative can always be checked by taking its derivative on the stated domain.