Tutorials › Real Analysis › Comprehensive Core Real Analysis Proof Examination

Riemann Integration · Tutorial 500 of 1000

Comprehensive Core Real Analysis Proof Examination

Learn to prove when transforming an integrable function preserves Riemann integrability, and how Darboux-gap estimates organize a rigorous argument.

Advanced 10 min read

What You'll Learn

  • Use a Lipschitz bound to control the Darboux gap of a transformed function
  • Prove that a continuous transformation of a bounded Riemann integrable function is Riemann integrable
  • Separate intervals of small oscillation from intervals of large oscillation in an integrability proof
  • Apply the composition theorem to squares, reciprocals, and step-function transformations
  • Identify common gaps in proofs involving boundedness, ranges, and endpoint values

A Proof Examination in Riemann Integration

The previous tutorial examined how uniform limits interact with Riemann integration. This tutorial turns to a related proof question: if \(f\) is Riemann integrable and \(\varphi\) is continuous, must the transformed function \(\varphi\circ f\) also be Riemann integrable? A proof needs more than the observation that \(\varphi\) is continuous. It must connect continuity of \(\varphi\) to the Darboux gap of the composition.

We will build that connection in two stages. First, a Lipschitz bound gives a direct quantitative estimate. Then uniform continuity, together with a division of partition intervals into two classes, proves the result for every continuous transformation on the bounded range of \(f\). These arguments also provide a useful examination method: state the hypotheses, control the quantities used in the criterion, and check that each estimate applies on every relevant interval.

Throughout, let \(a<b\), and let \(f:[a,b]\to\mathbb{R}\) be bounded. We use the Darboux Criterion and the Oscillation Formula for the Darboux Gap established earlier in this course. For a partition \(P=\{x_0,\ldots,x_n\}\), write \(\omega_i(f)\) for the oscillation of \(f\) on \([x_{i-1},x_i]\). The formula says that

$$ U(f,P)-L(f,P) = \sum_{i=1}^{n}\omega_i(f)(x_i-x_{i-1}). $$

Thus a small Darboux gap means that the oscillations, weighted by interval lengths, have a small total.

First Proof: A Quantitative Lipschitz Estimate

Theorem (Lipschitz Transformations Preserve Integrability): Suppose \(f:[a,b]\to[m,M]\) is Riemann integrable, and \(\varphi:[m,M]\to\mathbb{R}\) satisfies
$$ |\varphi(u)-\varphi(v)|\leq L|u-v| \quad\text{for all }u,v\in[m,M], $$
for some \(L\geq0\). Then \(\varphi\circ f\) is Riemann integrable. Moreover, for every partition \(P\),
$$ U(\varphi\circ f,P)-L(\varphi\circ f,P) \leq L\bigl(U(f,P)-L(f,P)\bigr). $$

Proof. For any two points \(x,y\) in a partition subinterval, the Lipschitz hypothesis gives

$$ |\varphi(f(x))-\varphi(f(y))| \leq L|f(x)-f(y)|. $$

Taking the supremum over pairs \(x,y\) in that subinterval shows that its oscillation satisfies \(\omega_i(\varphi\circ f)\leq L\omega_i(f)\). The composition is bounded because \(f(x)\in[m,M]\) and the Lipschitz condition bounds \(\varphi\) on that interval: for any fixed \(u_0\in[m,M]\),

$$ |\varphi(u)|\leq|\varphi(u_0)|+L|u-u_0| \leq|\varphi(u_0)|+L(M-m). $$

Multiplying the oscillation inequality on each subinterval by its length and summing, the Oscillation Formula gives

$$ U(\varphi\circ f,P)-L(\varphi\circ f,P) \leq L\sum_{i=1}^{n}\omega_i(f)(x_i-x_{i-1}) = L\bigl(U(f,P)-L(f,P)\bigr). $$

If \(L=0\), then \(\varphi\) is constant on \([m,M]\), so \(\varphi\circ f\) is a constant function and is integrable. If \(L>0\), let \(\varepsilon>0\). Since \(f\) is integrable, the Darboux Criterion gives a partition \(P\) with \(U(f,P)-L(f,P)<\varepsilon/L\). The displayed estimate makes the Darboux gap of \(\varphi\circ f\) smaller than \(\varepsilon\). Applying the Darboux Criterion proves integrability. \(\square\)

The quantitative estimate is stronger than an existence argument: it shows exactly how the partition gap changes. A small Lipschitz constant can make the transformed gap smaller; a large one may enlarge it, but by at most the stated factor.

Worked Example: Squaring a Piecewise-Defined Integrable Function

Define \(f:[0,1]\to\mathbb{R}\) by \(f(x)=x\) for \(0\leq x<1/2\), and \(f(x)=x+1\) for \(1/2\leq x\leq1\). This is the sum of the continuous function \(x\) and a step function, so linearity of Riemann integrability shows that \(f\) is integrable. Its range lies in \([0,2]\). The function \(\varphi(t)=t^2\) is Lipschitz on \([0,2]\), since

$$ |u^2-v^2|=|u-v||u+v|\leq4|u-v| \quad\text{for }u,v\in[0,2]. $$

The theorem therefore proves that \(f^2\) is Riemann integrable. We can also calculate its integral by splitting at \(1/2\); the value at the splitting point does not affect the integral:

$$ \begin{aligned} \int_0^1 f(x)^2\,dx &=\int_0^{1/2}x^2\,dx+\int_{1/2}^{1}(x+1)^2\,dx\\ &=\frac{1}{24} +\left[\frac{x^3}{3}+x^2+x\right]_{1/2}^{1}\\ &=\frac{1}{24}+\left(\frac{7}{3}-\frac{19}{24}\right) =\frac{19}{12}. \end{aligned} $$

For the second integral, the antiderivative has value \(7/3\) at \(1\) and \(19/24\) at \(1/2\); their difference is \(37/24\). Adding \(1/24\) gives \(38/24=19/12\), as displayed.

Continuous Transformations: Separate the Difficult Intervals

A continuous function on a closed bounded interval is uniformly continuous, but it need not be Lipschitz. To handle this broader case, use uniform continuity on partition intervals where \(f\) has small oscillation. The intervals where \(f\) has large oscillation need not be controlled individually; integrability ensures their total length is small.

Theorem (Continuous Transformations Preserve Integrability): Suppose \(f:[a,b]\to[m,M]\) is Riemann integrable and \(\varphi:[m,M]\to\mathbb{R}\) is continuous. Then \(\varphi\circ f\) is Riemann integrable on \([a,b]\).

Proof. If \(m=M\), then \(f\) is constant, and so is \(\varphi\circ f\). Suppose \(m<M\). Since \(\varphi\) is continuous on the closed bounded interval \([m,M]\), it is bounded and uniformly continuous there. Let \(C\geq0\) satisfy \(|\varphi(t)|\leq C\) for all \(t\in[m,M]\).

If \(C=0\), then \(\varphi\circ f\) is zero and is integrable. Assume \(C>0\), and let \(\varepsilon>0\). Uniform continuity gives a number \(\delta>0\) such that, for \(u,v\in[m,M]\),

$$ |u-v|<\delta \quad\Longrightarrow\quad |\varphi(u)-\varphi(v)|< \frac{\varepsilon}{2(b-a)}. $$

Because \(f\) is Riemann integrable, choose a partition \(P=\{x_0,\ldots,x_n\}\) such that

$$ U(f,P)-L(f,P) < \eta, \qquad \eta=\frac{\varepsilon\delta}{4C}. $$

Call a subinterval good if its oscillation \(\omega_i(f)\) is less than \(\delta\), and bad if \(\omega_i(f)\geq\delta\). On each good interval, any two values of \(f\) differ by less than \(\delta\); uniform continuity therefore bounds the oscillation of \(\varphi\circ f\) there by \(\varepsilon/(2(b-a))\). The total contribution of all good intervals to the Darboux gap is at most

$$ \frac{\varepsilon}{2(b-a)} \sum_{\text{good }i}(x_i-x_{i-1}) \leq\frac{\varepsilon}{2}. $$

On any bad interval, the oscillation of \(\varphi\circ f\) is at most \(2C\), because its values lie between \(-C\) and \(C\). Also, since \(\omega_i(f)\geq\delta\), the total length of the bad intervals satisfies

$$ \delta\sum_{\text{bad }i}(x_i-x_{i-1}) \leq \sum_{\text{bad }i}\omega_i(f)(x_i-x_{i-1}) \leq U(f,P)-L(f,P) <\eta. $$

Their contribution to the Darboux gap of \(\varphi\circ f\) is consequently less than

$$ 2C\sum_{\text{bad }i}(x_i-x_{i-1}) < \frac{2C\eta}{\delta} = \frac{\varepsilon}{2}. $$

The Oscillation Formula now bounds the full Darboux gap by the sum of the good-interval and bad-interval contributions, which is less than \(\varepsilon\). The composition is bounded because \(|\varphi(f(x))|\leq C\). The Darboux Criterion proves that \(\varphi\circ f\) is Riemann integrable. \(\square\)

The central proof technique is the split into good and bad intervals. Uniform continuity controls the transformed oscillation where the original oscillation is small. The Darboux gap of \(f\) then ensures that the intervals where this control is unavailable occupy little total length.

Worked Example: Taking Reciprocals Away from Zero

On \([0,1]\), define \(f(x)=1+x\) for \(0\leq x<1/2\), and \(f(x)=3+x\) for \(1/2\leq x\leq1\). As in the earlier example, this is a sum of a continuous function and a step function, so it is Riemann integrable. Its values lie in \([1,4]\). The reciprocal transformation is Lipschitz on this range because

$$ \left|\frac{1}{u}-\frac{1}{v}\right| = \frac{|u-v|}{uv} \leq |u-v| \quad\text{for }u,v\in[1,4]. $$

Thus \(1/f\) is integrable. The lower bound \(f(x)\geq1\) matters: without it, the reciprocal transformation could be undefined or unbounded on the range. Splitting at \(1/2\) also gives an explicit integral:

$$ \begin{aligned} \int_0^1\frac{1}{f(x)}\,dx &=\int_0^{1/2}\frac{1}{1+x}\,dx +\int_{1/2}^{1}\frac{1}{3+x}\,dx\\ &=\ln\left(\frac{3}{2}\right)+\ln\left(\frac{8}{7}\right) =\ln\left(\frac{12}{7}\right). \end{aligned} $$

In the second term, the endpoint denominators are \(7/2\) and \(4\), so the logarithmic difference is \(\ln(4)-\ln(7/2)=\ln(8/7)\).

Worked Example: Transforming a Step Function by a Continuous Map

Let \(f(x)=\lfloor3x\rfloor\) on \([0,1]\), where \(\lfloor y\rfloor\) denotes the greatest integer no larger than \(y\). This is a step function: it equals \(0\) on \([0,1/3)\), \(1\) on \([1/3,2/3)\), and \(2\) on \([2/3,1)\), with \(f(1)=3\). In particular, it is Riemann integrable and takes values in \([0,3]\). The continuous transformation \(\varphi(t)=\cos(\pi t)\) therefore gives an integrable composition.

On the three subintervals of length \(1/3\), the composition takes values \(1,-1,1\), respectively. At the single endpoint \(x=1\), its value is \(\cos(3\pi)=-1\); this isolated value does not affect the integral. Hence

$$ \int_0^1\cos\bigl(\pi\lfloor3x\rfloor\bigr)\,dx = \frac{1}{3}(1)+\frac{1}{3}(-1)+\frac{1}{3}(1) = \frac{1}{3}. $$

The example is simple enough to calculate directly, but the continuous-transformation theorem applies just as well when an integrable function has many discontinuities and no convenient finite step description.

How to Audit a Proof

These results suggest a practical examination routine for proofs about transformed functions:

1
Check the range.
Identify a closed bounded interval containing every value of \(f\). Continuity or a Lipschitz condition for \(\varphi\) must apply on that interval.
2
Check boundedness.
Establish that the transformed function is bounded before invoking the Darboux Criterion. For continuous \(\varphi\), boundedness on the compact range supplies this step.
3
Control the Darboux gap.
Use the oscillation formula. A Lipschitz map gives a direct bound; a merely continuous map requires the good-interval and bad-interval argument.
4
Conclude with the criterion.
For every positive tolerance, exhibit a partition whose transformed Darboux gap is smaller than that tolerance.

A common error is to say only that \(\varphi\) is continuous and conclude that \(\varphi\circ f\) is integrable. Continuity alone does not explain how the Darboux gap is controlled. The proof must use uniform continuity on a bounded closed range and account for intervals on which \(f\) oscillates too much for the uniform-continuity estimate to apply. Another frequent omission is the range check: \(1/t\), for example, is continuous only on intervals that avoid zero.

The broader lesson is a useful proof habit: identify the criterion that will finish the argument, express its key quantity in terms of the hypotheses, and treat every exceptional case explicitly. Here, the oscillation formula turns integrability into a weighted estimate, while continuity of the transformation transfers that estimate to the composed function.

Check Your Understanding

Answer each question by identifying the estimate or hypothesis that makes the proof work.

  1. What Darboux-gap estimate follows when \(\varphi\) is Lipschitz with constant \(L\)?
  2. In the proof for a continuous transformation, why can the total length of the bad intervals be bounded using the Darboux gap of \(f\)?
  3. Why must the composition be shown to be bounded before applying the Darboux Criterion?
  4. What range condition makes the reciprocal transformation safe in the second worked example?
  5. For \(f(x)=\lfloor3x\rfloor\), what values does \(\cos(\pi f(x))\) take on the three subintervals of length \(1/3\), and what is its integral?