Skip to content
Topos Uranos

Differentiation under the integral sign

Summary

Many integrals depend on a number that is not the variable of integration: the integral of e−txe^{-tx} between 00 and 11, for example, is a function of tt. This lesson studies these integrals with a parameter. If the integrand is continuous on a rectangle, the integral is continuous in the parameter, and if its derivative with respect to the parameter is continuous as well, the integral is differentiated by differentiating inside it: this is the Leibniz rule. The two arguments rest on uniform continuity on a rectangle, which is established with sequences, like the Heine-Cantor theorem of the unit on sequences, and with the bare minimum about functions of two variables. For improper integrals, passing to the limit requires the tail to become small for all values of the parameter at once: this is uniform convergence, which is obtained from a common bound for the integrand (the Weierstrass M-test) or, when there is none, from the tail bound of Dirichlet’s test from the preceding lesson. With these tools one computes the Frullani integrals and, as the lesson on conditional convergence announced, the value π2\frac{\pi}{2} of the integral of sin⁡xx\frac{\sin x}{x} on [0,∞)\ico{0}{\infty}.

Learning objectives

  1. Prove that a function continuous on a rectangle is uniformly continuous, and deduce continuity and the Leibniz rule for integrals with a parameter over a closed interval.
  2. Extend continuity and differentiation to improper integrals with a parameter by means of uniform convergence, and recognize through examples why convergence for each value of the parameter is not enough.
  3. Compute Frullani integrals under their exact hypotheses.
  4. Compute definite and improper integrals by differentiating with respect to a parameter, among them ∫0∞sin⁡xx dx=π2\int_0^{\infty} \frac{\sin x}{x} \, \mathrm{d}x = \frac{\pi}{2}.

Functions of two variables and uniform continuity

In an integral with a parameter, F(t)=∫abf(x,t) dxF(t) = \int_a^b f(x, t) \, \mathrm{d}x, the integrand is a function of two variables: it assigns a real number f(x,t)f(x, t) to each point (x,t)(x, t) of a set in the plane. The Calculus III course studies these functions in general; this lesson needs only their continuity on products of two intervals, I×JI \times J, made up of the points (x,t)(x, t) with xx in II and tt in JJ, and the derivative with respect to one variable with the other held fixed. The definitions copy those of Calculus I and that of the lesson on sequences, functions and uniform continuity, with a square in place of an interval.

DefinitionContinuity of a function of two variables

Let DD be a set of points in the plane and ff a function that assigns to each point (x,t)(x, t) of DD a real number f(x,t)f(x, t). We say that ff is continuous at the point (x0,t0)(x_0, t_0) of DD if for every ε>0\varepsilon > 0 there exists a δ>0\delta > 0 such that ∣f(x,t)−f(x0,t0)∣<ε\abs{f(x, t) - f(x_0, t_0)} < \varepsilon for every point (x,t)(x, t) of DD with ∣x−x0∣<δ\abs{x - x_0} < \delta and ∣t−t0∣<δ\abs{t - t_0} < \delta; that it is continuous on DD if it is continuous at each point of DD; and that it is uniformly continuous on DD if for every ε>0\varepsilon > 0 there exists a δ>0\delta > 0, depending only on ε\varepsilon, such that:

(∀(x,t)∈D)(∀(y,s)∈D)(∣x−y∣<δ∧∣t−s∣<δ→∣f(x,t)−f(y,s)∣<ε)(\forall (x, t) \in D)(\forall (y, s) \in D)\left(\abs{x - y} < \delta \wedge \abs{t - s} < \delta \rightarrow \abs{f(x, t) - f(y, s)} < \varepsilon\right)

The points close to (x0,t0)(x_0, t_0) are those of a square of side 2δ2\delta centered at it. Calculus III measures closeness with the Euclidean distance, and the two definitions are equivalent, because the disk of radius δ\delta lies inside that square and the square lies inside the disk of radius 2δ2\delta; nothing in this lesson depends on that equivalence.

LemmaOperations with continuous functions of two variables

Let II and JJ be intervals and D=I×JD = I \times J. Then:

  1. if φ\varphi is continuous on II and ψ\psi is continuous on JJ, the functions (x,t)↦φ(x)(x, t) \mapsto \varphi(x) and (x,t)↦ψ(t)(x, t) \mapsto \psi(t) are continuous on DD; and if ff is continuous on DD and t0t_0 lies in JJ, the function x↦f(x,t0)x \mapsto f(x, t_0) is continuous on II (likewise with x0x_0 fixed);
  2. if ff and gg are continuous on DD, so are f+gf + g and fgfg;
  3. if ff is continuous on DD, with values in an interval KK, and hh is continuous on KK, the function (x,t)↦h(f(x,t))(x, t) \mapsto h(f(x, t)) is continuous on DD.

The arguments for the theorem on operations with continuous functions and for the one on composition, from the lesson on limit and continuity, work with the new definition; they are repeated here.

Proof

  1. ∣x−x0∣<δ∧∣t−t0∣<δ→∣φ(x)−φ(x0)∣<ε\abs{x - x_0} < \delta \wedge \abs{t - t_0} < \delta \rightarrow \abs{\varphi(x) - \varphi(x_0)} < \varepsilon

    Let (x0,t0)(x_0, t_0) be a point of DD and ε>0\varepsilon > 0. If φ\varphi is continuous at x0x_0, the δ\delta corresponding to it works for the first function, because the condition on tt plays no role; likewise with ψ\psi. If ff is continuous at (x0,t0)(x_0, t_0), the points (x,t0)(x, t_0) with xx in II lie in DD and satisfy ∣t0−t0∣<δ\abs{t_0 - t_0} < \delta, and the δ\delta of ff works for the restriction.

  2. ∣fg−f0g0∣=∣f0 (g−g0)+g (f−f0)∣≤∣f0∣ ε2(∣f0∣+1)+(∣g0∣+1) ε2(∣g0∣+1)<ε\abs{fg - f_0 g_0} = \abs{f_0 \, (g - g_0) + g \, (f - f_0)} \leq \frac{\abs{f_0} \, \varepsilon}{2(\abs{f_0} + 1)} + \frac{(\abs{g_0} + 1) \, \varepsilon}{2(\abs{g_0} + 1)} < \resaltar{\varepsilon}

    Let f0=f(x0,t0)f_0 = f(x_0, t_0) and g0=g(x0,t0)g_0 = g(x_0, t_0). For the sum, take the smaller of the δ\delta of ff and of gg for ε2\frac{\varepsilon}{2}, and the triangle inequality. For the product, let δ\delta be the smallest of three: that of gg for 11, which gives ∣g(x,t)∣<∣g0∣+1\abs{g(x, t)} < \abs{g_0} + 1; that of gg for ε2(∣f0∣+1)\frac{\varepsilon}{2(\abs{f_0} + 1)}; and that of ff for ε2(∣g0∣+1)\frac{\varepsilon}{2(\abs{g_0} + 1)}. At the points of DD in the square of side 2δ2\delta:

  3. ∣x−x0∣<δ∧∣t−t0∣<δ→∣f(x,t)−y0∣<η→∣h(f(x,t))−h(y0)∣<ε\abs{x - x_0} < \delta \wedge \abs{t - t_0} < \delta \rightarrow \abs{f(x, t) - y_0} < \eta \rightarrow \resaltar{\abs{h(f(x, t)) - h(y_0)} < \varepsilon}

    Let y0=f(x0,t0)y_0 = f(x_0, t_0), which lies in KK. Since hh is continuous at y0y_0, there is an η>0\eta > 0 such that ∣h(y)−h(y0)∣<ε\abs{h(y) - h(y_0)} < \varepsilon for every yy in KK with ∣y−y0∣<η\abs{y - y_0} < \eta; and since ff is continuous at (x0,t0)(x_0, t_0), there is a δ\delta corresponding to η\eta. Since f(x,t)f(x, t) lies in KK:

For example, (x,t)↦e−tx(x, t) \mapsto e^{-tx} is continuous on every product of two intervals: the functions xx and tt are continuous by the first item, their product and the product of the latter by the constant −1-1 by the second, and the composite exponential by the third.

TheoremHeine-Cantor theorem on a rectangle

Let a<ba < b, p<qp < q and ff be a function continuous on the rectangle [a,b]×[p,q]\icc{a}{b} \times \icc{p}{q}. Then ff is uniformly continuous on it. If it were not, there would be pairs of points ever closer together with values far apart; the Bolzano-Weierstrass theorem, applied to each coordinate, would make those pairs converge to a single point of the rectangle, at which ff would not be continuous.

Proof

  1. ∣xk−yk∣<1k,∣tk−sk∣<1k,∣f(xk,tk)−f(yk,sk)∣≥ε0\dato{par}{\abs{x_k - y_k} < \frac{1}{k}, \qquad \abs{t_k - s_k} < \frac{1}{k}, \qquad \abs{f(x_k, t_k) - f(y_k, s_k)} \geq \varepsilon_0}

    If ff is not uniformly continuous, there is an ε0>0\varepsilon_0 > 0 for which no δ\delta works. With δ=1k\delta = \frac{1}{k}, for each natural number kk, choose two points of the rectangle, (xk,tk)(x_k, t_k) and (yk,sk)(y_k, s_k), that show it (one choice for each kk, as is customary in analysis).

  2. xni→x∗,tni→t∗x_{n_i} \to x^{*}, \qquad t_{n_i} \to t^{*}

    Since a≤xk≤ba \leq x_k \leq b, the Bolzano-Weierstrass theorem, from the lesson on subsequences, gives a subsequence (xkj)(x_{k_j}) with limit x∗x^{*} in [a,b]\icc{a}{b}; applied to (tkj)(t_{k_j}), which lies in [p,q]\icc{p}{q}, it gives a subsequence of it with limit t∗t^{*} in [p,q]\icc{p}{q}. Let nin_i be its indices, which increase strictly and satisfy ni≥in_i \geq i, by the lemma on the indices of a subsequence. By the theorem on the subsequences of a sequence with a limit, from the same lesson, (xni)(x_{n_i}) also tends to x∗x^{*}.

  3. yni=xni+(yni−xni)→x∗,sni→t∗y_{n_i} = x_{n_i} + (y_{n_i} - x_{n_i}) \to x^{*}, \qquad s_{n_i} \to t^{*}

    By the first inequality, ∣yni−xni∣<1ni≤1i\abs{y_{n_i} - x_{n_i}} < \frac{1}{n_i} \leq \frac{1}{i}, which tends to 00, by the proposition on powers of the reciprocal, from the lesson on sequences and their limit. The theorem stating that the limit theorems hold for sequences, from the same lesson (the squeeze theorem and the algebra of limits), gives the limits of the other two coordinates.

  4. ∣f(xni,tni)−f(yni,sni)∣≤∣f(xni,tni)−f(x∗,t∗)∣+∣f(x∗,t∗)−f(yni,sni)∣<ε0\abs{f(x_{n_i}, t_{n_i}) - f(y_{n_i}, s_{n_i})} \leq \abs{f(x_{n_i}, t_{n_i}) - f(x^{*}, t^{*})} + \abs{f(x^{*}, t^{*}) - f(y_{n_i}, s_{n_i})} < \resaltar{\varepsilon_0}

    Since ff is continuous at (x∗,t∗)(x^{*}, t^{*}), there is a δ\delta corresponding to ε02\frac{\varepsilon_0}{2}, and by the definition of the limit of a sequence there is an index from which on the four coordinates are at distance less than δ\delta from their limits (the largest of the four indices). From that index on, the triangle inequality contradicts the third inequality of the pairs.

TheoremContinuity of an integral with a parameter

Let a<ba < b, p<qp < q and ff be a function continuous on [a,b]×[p,q]\icc{a}{b} \times \icc{p}{q}. Then, for each tt in [p,q]\icc{p}{q}, the function x↦f(x,t)x \mapsto f(x, t) is integrable on [a,b]\icc{a}{b}, and the function FF defined by the following formula is uniformly continuous on [p,q]\icc{p}{q}:

F(t)=∫abf(x,t) dxF(t) = \int_a^b f(x, t) \, \mathrm{d}x

If tt is close to t0t_0, f(x,t)f(x, t) is close to f(x,t0)f(x, t_0) for all xx at once, and so are the integrals. The figure shows the graphs of x↦f(x,t0)x \mapsto f(x, t_0) and of x↦f(x,t)x \mapsto f(x, t); the second, inside a strip of height 2η2\eta around the first; and the absolute value of the difference of the two, below the height η\eta, whose integral bounds ∣F(t)−F(t0)∣\abs{F(t) - F(t_0)}.

Proof

  1. t∈[p,q]→F(t)=∫abf(x,t) dxt \in \icc{p}{q} \rightarrow F(t) = \int_a^b f(x, t) \, \mathrm{d}x

    By the first item of the lemma on operations with continuous functions of two variables, each function x↦f(x,t)x \mapsto f(x, t) is continuous on [a,b]\icc{a}{b}, and integrable by the theorem stating that continuous functions are integrable, from the lesson on integrable functions: FF is defined on [p,q]\icc{p}{q}.

  2. ∣t−t0∣<δ→∣f(x,t)−f(x,t0)∣<η,a≤x≤b\dato{eta}{\abs{t - t_0} < \delta \rightarrow \abs{f(x, t) - f(x, t_0)} < \eta, \qquad a \leq x \leq b}

    Given ε>0\varepsilon > 0, let η=ε2(b−a)\eta = \frac{\varepsilon}{2(b - a)}. The Heine-Cantor theorem on a rectangle gives a δ\delta corresponding to η\eta and depending only on ε\varepsilon. If tt and t0t_0 lie in [p,q]\icc{p}{q} and ∣t−t0∣<δ\abs{t - t_0} < \delta, then for each xx in [a,b]\icc{a}{b} the points (x,t)(x, t) and (x,t0)(x, t_0) have the same first coordinate and second coordinates at distance less than δ\delta.

  3. ∣F(t)−F(t0)∣≤∫ab∣f(x,t)−f(x,t0)∣ dx≤η (b−a)=ε2<ε\abs{F(t) - F(t_0)} \leq \int_a^b \abs{f(x, t) - f(x, t_0)} \, \mathrm{d}x \leq \eta \, (b - a) = \frac{\varepsilon}{2} < \resaltar{\varepsilon}

    By the theorem on linearity of the integral, the one on the absolute value of the integral and the corollary on bounds for the integral, from the lesson on properties of the definite integral, the difference is bounded. Since δ\delta does not depend on t0t_0, FF is uniformly continuous, and in particular continuous.

The closed and bounded interval of integration cannot be dropped: the solved problem “An integral with a parameter that jumps” shows an improper integral of a function continuous on [0,∞)×[0,1]\ico{0}{\infty} \times \icc{0}{1} that equals 11 if t>0t > 0 and 00 if t=0t = 0.