A function can fail to be useful long before it becomes discontinuous or non-measurable. The simple formula $f(x)=1/x$ on $(0,1)$ is smooth, measurable, and easy to integrate on compact subintervals away from $0$, yet it has no global ceiling. If a later argument wants to estimate $|f(x)g(x)|$ by a fixed multiple of $|g(x)|$, or wants a uniform error estimate that holds for every $x$ in the domain, this function refuses to cooperate. The idea of a bounded function records the difference between pointwise control and uniform control. Pointwise, every finite value is harmless. Uniformly, the whole graph must live inside one horizontal strip, or more generally inside one ball of the codomain. This small distinction is one of the first places where analysis stops asking what happens at a point and starts asking what can be controlled at once. [example: A Smooth Function with No Global Bound] Let $f:(0,1)\to\mathbb R$ be given by $f(x)=1/x$. For each fixed $x\in(0,1)$, we have $x>0$, so $f(x)=1/x$ is a finite real number. The function is differentiable at every $x\in(0,1)$ because the reciprocal function is differentiable away from $0$, with derivative $f'(x)=-1/x^2$. We show that no number $M>0$ satisfies $|f(x)|\le M$ for every $x\in(0,1)$. Let $M>0$. Since $M>0$, the number \begin{align*} x=\frac{1}{M+1} \end{align*} satisfies $x>0$. Also $M+1>1$, so \begin{align*} \frac{1}{M+1}<1 \end{align*} and therefore $x\in(0,1)$. For this choice of $x$, \begin{align*} |f(x)|=\left|\frac{1}{x}\right|=\frac{1}{x} \end{align*} because $x>0$. Substituting $x=1/(M+1)$ gives \begin{align*} \frac{1}{x}=\frac{1}{1/(M+1)}=M+1. \end{align*} Since $M+1>M$, we have $|f(x)|>M$. Thus every proposed global bound $M$ is exceeded by some point of $(0,1)$, so $f$ is not bounded on $(0,1)$. [/example] This example is not pathological in its formula; the failure comes from the domain. On every closed interval $[a,1]$ with $0<a<1$, the same function is bounded by $1/a$. Boundedness is therefore a global assertion about a chosen domain, not just a local feature of a formula. ## Definition The basic real-valued definition asks for one number that controls every output. The point is not that the function avoids infinity as a value, but that its finite values are uniformly trapped. [definition: Bounded Real-Valued Function] Let $X$ be a set. A function $f:X\to\mathbb R$ is a bounded real-valued function if there exists $M\ge 0$ such that \begin{align*} |f(x)|\le M \end{align*} for every $x\in X$. [/definition] The definition uses absolute value because the codomain is ordered and centred at $0$. In order-based arguments, however, one often receives separate estimates $m\le f(x)$ and $f(x)\le M$ rather than a symmetric estimate around zero. The obstruction is to turn those two one-sided bounds into a single absolute-value bound, and conversely to recover two-sided bounds from one number. [quotetheorem:9148] This theorem explains why boundedness is sometimes phrased as being bounded above and below. The absolute-value formulation is symmetric around $0$; the order formulation permits an interval whose centre is chosen by the function. [example: Finding the Sharp Bound] Let $f:[0,2]\to\mathbb R$ be given by $f(x)=x(2-x)$. For $x\in[0,2]$, expanding around $1$ gives \begin{align*} x(2-x)=2x-x^2=1-(x^2-2x+1)=1-(x-1)^2. \end{align*} Also $0\le x$ and $0\le 2-x$, so \begin{align*} f(x)=x(2-x)\ge 0. \end{align*} Since $x\in[0,2]$, subtracting $1$ gives $-1\le x-1\le 1$, hence \begin{align*} 0\le (x-1)^2\le 1. \end{align*} Therefore \begin{align*} f(x)=1-(x-1)^2\le 1. \end{align*} Thus $0\le f(x)\le 1$ for every $x\in[0,2]$, so $|f(x)|=f(x)\le 1$ for every $x\in[0,2]$. At $x=1$, \begin{align*} f(1)=1(2-1)=1, \end{align*} so the upper bound $1$ is actually attained. Hence the least possible uniform bound is \begin{align*} \sup_{x\in[0,2]}|f(x)|=1. \end{align*} The notation for this supremal size will be introduced below as the supremum norm. The example shows that the best uniform bound is not merely an estimate here; it is reached at the midpoint of the interval. [/example] ## Metric and Normed Viewpoints The real-valued definition is the model case, but many functions in analysis are not scalar-valued: curves take values in $\mathbb R^n$, solutions may be vector-valued, and abstract maps may land in a [metric space](/page/Metric%20Space). To keep boundedness from depending on an absolute value that may not exist, the metric version asks whether the entire image fits inside one ball. As usual for this page, metric spaces are assumed to be nonempty. [definition: Bounded Map into a Metric Space] Let $X$ be a set and let $(Y,d)$ be a metric space. A function $f:X\to Y$ is bounded if there exist $y_0\in Y$ and $R\ge 0$ such that \begin{align*} f(X)\subset \overline{B}(y_0,R):=\{y\in Y:d(y,y_0)\le R\}. \end{align*} [/definition] For real-valued functions, taking $Y=\mathbb R$ with its usual metric and choosing $y_0=0$ gives the absolute-value bound $|f(x)|\le R$. Conversely, if the image of $f:X\to\mathbb R$ lies in some closed ball $\overline{B}(y_0,R)$, then the triangle inequality gives $|f(x)|\le R+|y_0|$ for every $x\in X$. Thus the metric definition and the scalar definition describe the same real-valued functions. Once a function is known to be bounded, the next question is quantitative: among all possible bounds, which number records the true uniform size of the function? For scalar functions, that number is obtained by taking the supremum of the absolute values. [definition: Supremum Norm] Let $X$ be a set, and use the convention that the supremum of the empty set of nonnegative [real numbers](/page/Real%20Numbers) is $0$. The supremum norm is the function from the set of bounded functions $f:X\to\mathbb R$ to $[0,\infty)$ that assigns to $f$ the number \begin{align*} \sup_{x\in X}|f(x)|. \end{align*} For a bounded function $f:X\to\mathbb R$, write \begin{align*} \|f\|_\infty:=\sup_{x\in X}|f(x)|. \end{align*} [/definition] This norm is finite exactly for bounded real-valued functions. It turns pointwise inequalities into norm estimates, and those estimates are the language used throughout the rest of the chapter. ## Uniform Control and Pointwise Control Boundedness is a uniform condition. It is stronger than saying that every value is finite, and it is different from saying that every small part of the domain behaves well. The distinction matters whenever an estimate must survive a supremum over all inputs. ### Local Boundedness A useful intermediate notion is boundedness near each point. This captures functions such as $1/x$ on $(0,1)$: no single global bound exists, but each point has a neighbourhood on which the function is controlled. [definition: Locally Bounded Function] Let $(X,\tau)$ be a [topological space](/page/Topological%20Space). A function $f:X\to\mathbb R$ is locally bounded if for every $x_0\in X$ there exist an [open set](/page/Open%20Set) $U\in\tau$ with $x_0\in U$ and a number $M\ge 0$ such that \begin{align*} |f(x)|\le M \end{align*} for every $x\in U$. [/definition] Local boundedness is enough for many pointwise constructions, but it cannot replace boundedness in estimates over the whole domain. The example below is the standard warning: the function behaves well around every point, while the domain contains a direction in which the values escape. [example: Locally Bounded but Not Bounded] Let $f:\mathbb R\to\mathbb R$ be given by $f(x)=x$. Fix $x_0\in\mathbb R$ and set $U=(x_0-1,x_0+1)$. Then $U$ is open and contains $x_0$, because \begin{align*} x_0-1<x_0<x_0+1. \end{align*} If $x\in U$, then $x_0-1<x<x_0+1$, so $-1<x-x_0<1$. Hence $|x-x_0|<1$, and the triangle inequality gives \begin{align*} |f(x)|=|x|=|x_0+(x-x_0)|\le |x_0|+|x-x_0|<|x_0|+1. \end{align*} Thus on the neighbourhood $U$ the function is bounded by $|x_0|+1$, and since $x_0$ was arbitrary, $f$ is locally bounded. We now show that no global bound exists on $\mathbb R$. Let $M>0$ be any proposed bound and choose \begin{align*} x=M+1. \end{align*} Then $x\in\mathbb R$, and since $M>0$ we have $M+1>0$. Therefore \begin{align*} |f(x)|=|x|=|M+1|=M+1. \end{align*} Because $M+1>M$, this point satisfies $|f(x)|>M$. Every proposed bound is exceeded somewhere on $\mathbb R$, so $f$ is not bounded on $\mathbb R$. [/example] ### Uniformly Bounded Families Sometimes the question is not whether one function is controlled, but whether a whole collection can be controlled by the same constant. This occurs in convergence questions, approximation schemes, and compactness arguments. [definition: Uniformly Bounded Family] Let $X$ be a set and let $\mathcal F$ be a family of functions $f:X\to\mathbb R$. The family $\mathcal F$ is uniformly bounded if there exists $M\ge 0$ such that \begin{align*} |f(x)|\le M \end{align*} for every $f\in\mathcal F$ and every $x\in X$. [/definition] The word uniform now refers to two variables at once: the same bound works for every input and every function in the family. This is much stronger than requiring each member of the family to be bounded separately. [example: Bounded Members Without a Uniform Bound] For each $n\in\mathbb N$, define $f_n:[0,1]\to\mathbb R$ by $f_n(x)=nx$. Fix $n\in\mathbb N$. If $x\in[0,1]$, then $0\le x\le 1$, so multiplying by $n\ge 0$ gives \begin{align*} 0\le nx\le n. \end{align*} Thus \begin{align*} |f_n(x)|=|nx|=nx\le n. \end{align*} At the endpoint $x=1$, \begin{align*} |f_n(1)|=|n\cdot 1|=n. \end{align*} Therefore the least upper bound of the set $\{|f_n(x)|:x\in[0,1]\}$ is $n$, and hence \begin{align*} \|f_n\|_\infty=n. \end{align*} So every individual function $f_n$ is bounded on $[0,1]$. However, the family $(f_n)_{n\in\mathbb N}$ is not uniformly bounded. If a common bound $M\ge 0$ existed, then for every $n\in\mathbb N$ we would have \begin{align*} |f_n(1)|\le M. \end{align*} But $|f_n(1)|=n$, so this would imply $n\le M$ for every $n\in\mathbb N$. Since the natural numbers are not bounded above in $\mathbb R$, we can choose $n\in\mathbb N$ with $n>M$, contradicting $n\le M$. Thus the functions are bounded one at a time, but their bounds grow without any single uniform bound for the whole family. [/example] Uniform boundedness of a sequence can interact with pointwise convergence. Pointwise convergence alone gives information at each fixed input, but it does not by itself produce a bound that is independent of the input. When every function in the sequence lies under the same envelope, the natural question is whether the limiting function can escape that envelope at some point. [quotetheorem:9149] This theorem is often used without ceremony in analysis: a uniform envelope passes to a pointwise limit. What it does not provide is control of the error $|f_n(x)-f(x)|$ uniformly in $x$. ## Algebra of Bounded Functions Bounded functions are stable under the operations that make real-valued functions useful. If two functions each live inside a horizontal strip, then their sum, scalar multiple, and product also live inside a horizontal strip, with bounds computed from the old ones. ### The Function Space $B(X)$ To turn this stability into a named object, fix the domain and collect all bounded scalar functions on it. This gives a natural home for uniform estimates, because all functions in the space have finite supremum norm. [definition: Space of Bounded Functions] Let $X$ be a set. The space of bounded real-valued functions on $X$ is \begin{align*} B(X):=\{f:X\to\mathbb R: f \text{ is bounded}\}. \end{align*} [/definition] The notation $B(X)$ is common in elementary analysis, while some texts use $\ell^\infty(X)$ for arbitrary sets. Once this collection is named, the next question is whether the usual pointwise operations stay inside it and interact well with the supremum norm. [quotetheorem:9150] The product estimate is one reason bounded functions appear constantly as multipliers. Multiplying by a bounded function cannot enlarge another function by more than a fixed factor at any point. [example: A Bounded Multiplier] Let $m:\mathbb R\to\mathbb R$ be given by $m(x)=\sin x$, and let $g:\mathbb R\to\mathbb R$ be any function. For every $x\in\mathbb R$ we have $-1\le \sin x\le 1$, so $|m(x)|=|\sin x|\le 1$. Hence \begin{align*} |m(x)g(x)|=|m(x)|\,|g(x)|\le 1\cdot |g(x)|=|g(x)|. \end{align*} If $g$ is bounded, then $|g(x)|\le \|g\|_\infty$ for every $x\in\mathbb R$, and therefore \begin{align*} |m(x)g(x)|\le |g(x)|\le \|g\|_\infty. \end{align*} Taking the supremum over $x\in\mathbb R$ gives \begin{align*} \|mg\|_\infty=\sup_{x\in\mathbb R}|m(x)g(x)|\le \|g\|_\infty. \end{align*} Thus the single uniform bound $|\sin x|\le 1$ makes multiplication by $\sin x$ norm-nonincreasing for the supremum norm. The same pointwise idea later reappears in measure-theoretic function spaces, where one compares integrals rather than suprema. [/example] ### Completeness in the Supremum Norm The supremum norm is not only a bookkeeping device. Approximation arguments often construct a candidate function indirectly, by producing functions that get uniformly close to one another. The obstruction is that a [Cauchy sequence](/page/Cauchy%20Sequence) could converge pointwise to something outside the space unless boundedness and the supremum norm keep the limit under control. [quotetheorem:9151] Completeness is the bridge from estimates to existence. If an approximation procedure produces a Cauchy sequence in the supremum norm, then it converges uniformly to a bounded function on the same domain. ## Compactness and Continuity Boundedness becomes more powerful when it is forced by structural assumptions instead of checked by hand. The fundamental source is compactness: continuous functions cannot escape to arbitrarily large values while moving inside a compact domain. ### Extrema on Compact Sets Before stating the [compactness theorem](/theorems/2748), it is worth naming the extremal property that continuous functions enjoy on compact sets. Boundedness gives control, while attainment says the best bounds are reached by actual points of the domain. [definition: Maximum and Minimum of a Function] Let $X$ be a set and let $f:X\to\mathbb R$. A point $x_{\max}\in X$ is a maximum point of $f$ if \begin{align*} f(x)\le f(x_{\max}) \end{align*} for every $x\in X$. A point $x_{\min}\in X$ is a minimum point of $f$ if \begin{align*} f(x_{\min})\le f(x) \end{align*} for every $x\in X$. [/definition] The existence of a supremum bound does not imply the existence of a maximum point. A function can approach its best possible bound without ever attaining it, or it can escape to larger values along a non-compact domain. This raises the basic extremal-value question: which hypotheses force the upper and lower bounds of a real-valued function to be achieved by points of the domain? The compactness of the domain and the continuity of the function are exactly the conditions that close off the two escape routes described above. [quotetheorem:182] The compactness hypothesis is doing real work. Continuity by itself does not prevent escape on a non-compact domain, and boundedness by itself does not force an extremum to be attained. [example: Missing Compactness and Missing Attainment] The function $f:\mathbb R\to\mathbb R$ given by $f(x)=x$ is continuous: if $\varepsilon>0$, choose $\delta=\varepsilon$, and whenever $|x-y|<\delta$ we have \begin{align*} |f(x)-f(y)|=|x-y|<\varepsilon. \end{align*} It is not bounded on $\mathbb R$. Indeed, if $M>0$ is proposed as a bound, take $x=M+1$. Then $x\in\mathbb R$ and \begin{align*} |f(x)|=|M+1|=M+1>M. \end{align*} Thus continuity alone does not force boundedness when the domain is not compact. Now let $g:(0,1)\to\mathbb R$ be given by $g(x)=x$. If $x\in(0,1)$, then $0<x<1$, so \begin{align*} |g(x)|=|x|=x<1. \end{align*} Hence $g$ is bounded on $(0,1)$, for example by $1$. However, $g$ has no maximum point: if $x_0\in(0,1)$, then \begin{align*} x_0<\frac{x_0+1}{2}<1, \end{align*} so $\frac{x_0+1}{2}\in(0,1)$ and \begin{align*} g\left(\frac{x_0+1}{2}\right)=\frac{x_0+1}{2}>x_0=g(x_0). \end{align*} It also has no minimum point: if $x_0\in(0,1)$, then \begin{align*} 0<\frac{x_0}{2}<x_0, \end{align*} so $\frac{x_0}{2}\in(0,1)$ and \begin{align*} g\left(\frac{x_0}{2}\right)=\frac{x_0}{2}<x_0=g(x_0). \end{align*} Thus boundedness alone does not force a largest or smallest value to be attained. [/example] ### From Local Bounds to Global Bounds Compactness also explains why local boundedness can become global boundedness. Local estimates may use different neighbourhoods and different constants, and on a large domain there might be infinitely many of them with no common maximum. Compactness removes that obstruction by reducing the domain to finitely many local pieces whose bounds can be compared. [quotetheorem:9152] This theorem is a compactness principle in miniature. Local information becomes global information precisely because compactness reduces an [open cover](/page/Open%20Cover) to finitely many pieces. ## Measure, Essential Bounds, and Integration In measure theory, [changing a function on a null set](/theorems/4915) should not change its analytic size in $L^p$ spaces. Ordinary boundedness is too sensitive for this purpose: a single exceptional point can destroy it. The replacement is essential boundedness, which ignores behaviour on sets of measure zero. ### Essential Boundedness The measure-theoretic version asks for a bound that may fail on a null set. This is the right condition for spaces where functions are identified up to a.e. equality. [definition: Essentially Bounded Function] Let $(E,\mathcal E,\mu)$ be a [measure space](/page/Measure%20Space). A [measurable function](/page/Measurable%20Function) $f:E\to\mathbb R$ is essentially bounded if there exists $M\ge 0$ such that \begin{align*} |f(x)|\le M \end{align*} for $\mu$-a.e. $x\in E$. [/definition] Essential boundedness belongs to measure theory rather than pointwise analysis. Once null sets are ignored, the relevant size is not the supremum of a representative but the least a.e. bound. Since $L^p$ spaces identify functions that agree a.e., we first name the space whose elements are those equivalence classes. [definition: $L^\infty$ Space] Let $(E,\mathcal E,\mu)$ be a measure space. The space $L^\infty(E,\mathcal E,\mu)$ is the set of a.e.-equivalence classes of [measurable functions](/page/Measurable%20Functions) $f:E\to\mathbb R$ that are essentially bounded. [/definition] An ordinary supremum cannot be a norm on $L^\infty(E,\mathcal E,\mu)$, because an element of $L^\infty$ is an a.e.-equivalence class rather than a chosen pointwise representative. If two representatives differ only on a null set, assigning different sizes to them would contradict the way the space identifies functions. The essential supremum solves this problem by recording the least bound that works after discarding a null exceptional set. [definition: Essential Supremum Norm] Let $(E,\mathcal E,\mu)$ be a measure space. The essential supremum norm is the function \begin{align*} \|\cdot\|_{L^\infty}:L^\infty(E,\mathcal E,\mu)\to[0,\infty) \end{align*} defined by \begin{align*} \|[f]\|_{L^\infty}:=\operatorname{ess\,sup}_{x\in E}|f(x)|. \end{align*} For an equivalence class represented by $f$, write \begin{align*} \|f\|_{L^\infty}:=\operatorname{ess\,sup}_{x\in E}|f(x)|. \end{align*} [/definition] This is the norm used in $L^\infty(E,\mathcal E,\mu)$. It measures the smallest a.e. bound, not the largest literal value of a chosen representative. [example: Unbounded at One Point but Essentially Bounded] Let $f:[0,1]\to\mathbb R$ be given by $f(0)=1000$ and $f(x)=0$ for $x\in(0,1]$. The set of absolute values is \begin{align*} \{|f(x)|:x\in[0,1]\}=\{|f(0)|\}\cup\{|f(x)|:x\in(0,1]\}=\{1000\}\cup\{0\}. \end{align*} Since $0\le 1000$ and $1000$ belongs to this set, its supremum is $1000$. Hence, as an ordinary bounded function, \begin{align*} \|f\|_\infty=\sup_{x\in[0,1]}|f(x)|=1000. \end{align*} With respect to [Lebesgue measure](/page/Lebesgue%20Measure) $\mathcal L^1$, the exceptional set on which $f$ is not zero is exactly $\{0\}$, and \begin{align*} \mathcal L^1(\{0\})=0. \end{align*} Therefore $f(x)=0$ for $\mathcal L^1$-a.e. $x\in[0,1]$. On the full-measure set $(0,1]$, \begin{align*} |f(x)|=|0|=0. \end{align*} Thus $0$ is an a.e. bound for $|f|$. Since every essential bound is nonnegative and no number smaller than $0$ can bound the nonnegative function $|f|$, the least a.e. bound is $0$, so \begin{align*} \|f\|_{L^\infty}=0. \end{align*} The point value $f(0)=1000$ changes the ordinary supremum, but it disappears from the essential supremum because the singleton $\{0\}$ has Lebesgue measure zero. [/example] ### Bounded Height and Finite Mass On finite measure spaces, essential boundedness should imply membership in every finite $L^p$ space: bounded height over finite measure has finite total $p$-mass. The point that needs control is the integral of $|f|^p$: an essential bound controls the height almost everywhere, while finiteness of the measure controls the size of the region being integrated over. [quotetheorem:9153] The finite measure hypothesis cannot simply be removed. A function with bounded height can still have infinite total mass if it persists over an infinite domain. [example: Bounded but Not Integrable on an Infinite Measure Space] Let $f:\mathbb R\to\mathbb R$ be given by $f(x)=1$. For every $x\in\mathbb R$, \begin{align*} |f(x)|=|1|=1. \end{align*} Thus $|f(x)|\le 1$ for every $x\in\mathbb R$, so $f$ is bounded. The number $1$ is also the least essential bound: if $0\le M<1$, then $|f(x)|=1>M$ for every $x\in\mathbb R$, and the exceptional set where the inequality $|f(x)|\le M$ fails is all of $\mathbb R$, whose Lebesgue measure is infinite rather than zero. Hence \begin{align*} \|f\|_{L^\infty}=1. \end{align*} For the $L^1$ norm, the absolute value is again constant: \begin{align*} \int_{\mathbb R}|f(x)|\,d\mathcal L^1(x)=\int_{\mathbb R}1\,d\mathcal L^1(x). \end{align*} For each $R>0$, monotonicity of the [Lebesgue integral](/page/Lebesgue%20Integral) gives \begin{align*} \int_{\mathbb R}1\,d\mathcal L^1(x)\ge \int_{[-R,R]}1\,d\mathcal L^1(x). \end{align*} Since the integral of the constant function $1$ over an interval is its Lebesgue measure, \begin{align*} \int_{[-R,R]}1\,d\mathcal L^1(x)=\mathcal L^1([-R,R])=2R. \end{align*} Because $2R$ can be made arbitrarily large as $R\to\infty$, the integral over $\mathbb R$ is infinite: \begin{align*} \int_{\mathbb R}|f(x)|\,d\mathcal L^1(x)=\infty. \end{align*} Therefore $f\notin L^1(\mathbb R)$: bounded height does not imply finite integral on a space of infinite measure. [/example] ## Boundedness Under Operations and Composition Many estimates in analysis are built by combining functions. It is therefore useful to know which constructions preserve boundedness and which require extra hypotheses. The safe operations are those that do not allow a hidden denominator or an unbounded outer function to magnify values. ### Composition A first principle is that bounded input into a bounded-on-the-relevant-range function gives bounded output. This is the composition viewpoint: the range of the inner function becomes the domain on which the outer function must be controlled. [quotetheorem:9154] This theorem is intentionally stated with boundedness of $\varphi$ on the interval that contains the image of $f$. The outer function may be unbounded elsewhere without affecting the composition. [example: An Unbounded Formula That Gives a Bounded Composition] Let $g:\mathbb R\to\mathbb R$ be given by $g(x)=2+\sin x$, and let $\varphi:(0,\infty)\to\mathbb R$ be given by $\varphi(t)=1/t$. The formula $\varphi$ is unbounded on $(0,\infty)$: if $M>0$, set \begin{align*} t=\frac{1}{M+1}. \end{align*} Then $t>0$, so $t\in(0,\infty)$. At this point, \begin{align*} |\varphi(t)|=\left|\frac{1}{t}\right|=\frac{1}{1/(M+1)}=M+1>M. \end{align*} Thus no single bound works for $\varphi$ on all of $(0,\infty)$. However, for every $x\in\mathbb R$, the sine function satisfies $-1\le \sin x\le 1$, so adding $2$ throughout gives \begin{align*} 1\le g(x)=2+\sin x\le 3. \end{align*} Thus $g(\mathbb R)\subset[1,3]\subset(0,\infty)$, and the composition $h=\varphi\circ g$ is defined by \begin{align*} h(x)=\frac{1}{2+\sin x}. \end{align*} Taking reciprocals of the positive inequalities reverses the order, giving \begin{align*} \frac{1}{3}\le \frac{1}{2+\sin x}\le 1. \end{align*} Therefore \begin{align*} \frac{1}{3}\le h(x)\le 1 \end{align*} for every $x\in\mathbb R$. The reciprocal formula is unbounded near $0$, but this composition only feeds it values in $[1,3]$, safely away from the singular point. [/example] ### Reciprocals Division is the common place where boundedness fails under algebraic operations. A bounded numerator divided by a bounded denominator need not be bounded if the denominator approaches zero. [definition: Bounded Away from Zero] Let $X$ be a set. A function $g:X\to\mathbb R$ is bounded away from zero if there exists $c>0$ such that \begin{align*} |g(x)|\ge c \end{align*} for every $x\in X$. [/definition] Taking reciprocals turns small denominators into large values, so a reciprocal estimate needs a uniform lower bound rather than an upper bound. The obstruction is not that $g$ might be large, but that values of $g$ might approach zero and make $1/g$ arbitrarily large. A positive lower bound on $|g|$ is exactly the condition that prevents this blow-up. [quotetheorem:9155] The hypothesis concerns the lower bound of $|g|$, not the upper bound of $g$. A bounded denominator can still be dangerous if it comes arbitrarily close to zero. [example: Bounded Denominator with Unbounded Reciprocal] Let $g:(0,1)\to\mathbb R$ be given by $g(x)=x$. If $x\in(0,1)$, then $0<x<1$, so \begin{align*} |g(x)|=|x|=x<1. \end{align*} Thus $g$ is bounded on $(0,1)$, for example by the constant $1$. Now consider the reciprocal function $1/g:(0,1)\to\mathbb R$, which is given by \begin{align*} \frac{1}{g(x)}=\frac{1}{x}. \end{align*} We show that no global bound exists. Let $M>0$ be any proposed bound. Since $M+1>1$, the number \begin{align*} x=\frac{1}{M+1} \end{align*} satisfies $0<x<1$, so $x\in(0,1)$. For this choice of $x$, \begin{align*} \left|\frac{1}{g(x)}\right|=\left|\frac{1}{x}\right|. \end{align*} Because $x>0$, this becomes \begin{align*} \left|\frac{1}{x}\right|=\frac{1}{x}. \end{align*} Substituting $x=1/(M+1)$ gives \begin{align*} \frac{1}{x}=\frac{1}{1/(M+1)}=M+1. \end{align*} Since $M+1>M$, we have found a point $x\in(0,1)$ with \begin{align*} \left|\frac{1}{g(x)}\right|>M. \end{align*} Therefore $1/g$ is not bounded on $(0,1)$, even though the denominator $g$ itself is bounded there. [/example] ## Boundedness, Variation, and Regularity Boundedness controls height, not oscillation. A function may remain trapped in a finite interval while oscillating rapidly, and a function may have finite total variation for reasons stronger than mere boundedness. Keeping these notions separate prevents a common mistake: bounded output does not imply regular behaviour. To measure oscillation rather than height, we add up the changes along partitions of an interval. This answers a different question from boundedness: not how far the graph rises, but how much cumulative motion it can make. [definition: Total Variation on an Interval] Let $[a,b]\subsetneq\mathbb R$ be a compact interval. The total variation on $[a,b]$ is the functional \begin{align*} V_a^b:\{f:[a,b]\to\mathbb R\}\to[0,\infty] \end{align*} defined by \begin{align*} V_a^b(f):=\sup\left\{\sum_{i=1}^n |f(x_i)-f(x_{i-1})|: n\in\mathbb N,\ a=x_0<x_1<\cdots<x_n=b\right\}. \end{align*} We also write $\operatorname{Var}_{[a,b]}(f)$ for this same quantity when emphasizing the interval rather than the endpoint notation. [/definition] Analysis often needs a class of functions whose total oscillation is finite, because such functions admit stronger compactness and structure than arbitrary bounded functions. The total variation functional can take the value $\infty$, so the useful dividing line is whether the supremum over all partition sums is finite. That finiteness condition is the property being named here. [definition: Function of Bounded Variation on an Interval] Let $[a,b]\subsetneq\mathbb R$ be a compact interval. A function $f:[a,b]\to\mathbb R$ is of bounded variation on $[a,b]$ if \begin{align*} V_a^b(f)<\infty. \end{align*} [/definition] A finite variation bound controls how much the function can move away from any one value. Once one point value is fixed, every other value can differ from it by no more than the total accumulated variation along the interval. This observation raises the first basic structural question about bounded variation: does controlling total oscillation automatically control the ordinary size of the function? The only possible source of unboundedness would be values escaping far from a fixed reference value, but any such escape would force a large variation sum along a partition containing the two points. Thus finite total variation should prevent the graph from attaining arbitrarily large heights on the interval. The quoted theorem records this boundedness consequence before we separate it from the strictly weaker condition of being merely bounded. [quotetheorem:9156] The converse fails in a strong way. A function can remain between $-1$ and $1$ while accumulating infinite oscillation near a point. [example: Bounded Function with Infinite Oscillation] Define $f:(0,1]\to\mathbb R$ by $f(x)=\sin(1/x)$. Since $-1\le \sin t\le 1$ for every $t\in\mathbb R$, substituting $t=1/x$ gives \begin{align*} -1\le \sin(1/x)\le 1. \end{align*} Hence \begin{align*} |f(x)|=|\sin(1/x)|\le 1 \end{align*} for every $x\in(0,1]$, so $f$ is bounded. For each $k\in\mathbb N\cup\{0\}$, set \begin{align*} x_k=\frac{1}{\frac{\pi}{2}+k\pi}. \end{align*} Because $\frac{\pi}{2}+k\pi>0$, each $x_k$ belongs to $(0,1]$. Also, \begin{align*} \frac{1}{x_k}=\frac{1}{1/(\frac{\pi}{2}+k\pi)}=\frac{\pi}{2}+k\pi. \end{align*} Therefore \begin{align*} f(x_k)=\sin\left(\frac{1}{x_k}\right)=\sin\left(\frac{\pi}{2}+k\pi\right)=(-1)^k. \end{align*} Thus the values along this sequence alternate between $1$ and $-1$. On each fixed interval $[\delta,1]$ with $0<\delta<1$, the function has finite variation. Indeed, for $x\in[\delta,1]$, \begin{align*} f'(x)=-\frac{\cos(1/x)}{x^2}. \end{align*} Since $|\cos(1/x)|\le 1$ and $x\ge \delta$, we have \begin{align*} |f'(x)|=\frac{|\cos(1/x)|}{x^2}\le \frac{1}{\delta^2}. \end{align*} If $\delta=t_0<t_1<\cdots<t_n=1$ is any partition of $[\delta,1]$, then the [mean value theorem](/theorems/186) gives, for each $i$, a point $c_i\in(t_{i-1},t_i)$ such that \begin{align*} |f(t_i)-f(t_{i-1})|=|f'(c_i)|(t_i-t_{i-1}). \end{align*} Using the derivative bound, \begin{align*} |f(t_i)-f(t_{i-1})|\le \frac{1}{\delta^2}(t_i-t_{i-1}). \end{align*} Summing over the partition gives \begin{align*} \sum_{i=1}^n |f(t_i)-f(t_{i-1})|\le \frac{1}{\delta^2}\sum_{i=1}^n(t_i-t_{i-1})=\frac{1-\delta}{\delta^2}. \end{align*} Taking the supremum over all partitions shows $V_\delta^1(f)\le (1-\delta)/\delta^2<\infty$. The variation becomes unbounded as the interval approaches $0$. Fix $N\in\mathbb N$. Since the denominators $\frac{\pi}{2}+k\pi$ increase with $k$, the points satisfy \begin{align*} x_N<x_{N-1}<\cdots<x_1<x_0. \end{align*} So $x_N,x_{N-1},\ldots,x_0$ form a partition of the interval $[x_N,x_0]$. Along this partition, \begin{align*} |f(x_{k-1})-f(x_k)|=|(-1)^{k-1}-(-1)^k|=2 \end{align*} for each $k=1,\ldots,N$. Therefore \begin{align*} \sum_{k=1}^N |f(x_{k-1})-f(x_k)|=\sum_{k=1}^N 2=2N. \end{align*} Since $[x_N,x_0]\subset(0,1]$, the total variation on $(0,1]$ in the improper sense is at least $2N$ for every $N\in\mathbb N$. No finite number can dominate all the values $2N$, so this improper total variation is infinite. Thus boundedness controls the height of $f$, but not its accumulated oscillation near $0$. [/example] ## Beyond and Connected Topics Bounded functions are a first model for uniform estimates. In [Cambridge IA Analysis Notes](/page/Cambridge%20IA%20Analysis%20Notes), this idea appears through suprema, continuous functions on compact intervals, and the extreme value theorem. The conceptual move is from checking values one at a time to controlling a whole set at once; boundedness is the simplest form of that move. The next layer is topological. In [Cambridge IB Analysis and Topology](/page/Cambridge%20IB%20Analysis%20and%20Topology), boundedness interacts with compactness, metric spaces, and [uniform convergence](/page/Uniform%20Convergence). The same supremum norm that measures boundedness also defines uniform convergence of functions, so a bounded-function estimate often becomes the language in which convergence is measured. The function-space point of view becomes more systematic in [Cambridge II Analysis of Functions](/page/Cambridge%20II%20Analysis%20of%20Functions). The notation $C_b(X)$ denotes bounded continuous functions on a topological space $X$, while $L^\infty(E,\mathcal E,\mu)$ denotes the essentially bounded measurable functions on a measure space, modulo a.e. equality. These spaces, together with normed algebras of functions, are natural settings for approximation and operator estimates. Here boundedness is not just a property of examples; it is part of the ambient normed structure. In [Geometric Measure Theory III: BV Functions and Sets of Finite Perimeter](/page/Geometric%20Measure%20Theory%20III%3A%20BV%20Functions%20and%20Sets%20of%20Finite%20Perimeter), boundedness is no longer the main regularity condition. It is compared with variation, perimeter, and measure-theoretic control, where bounded height is useful but not enough to describe geometric structure. This is the natural warning at the edge of the chapter: boundedness controls size, while deeper theories control oscillation, compactness, or geometry. ## References Androma, [Cambridge IA Analysis Notes](/page/Cambridge%20IA%20Analysis%20Notes). Androma, [Cambridge IB Analysis and Topology](/page/Cambridge%20IB%20Analysis%20and%20Topology). Androma, [Cambridge II Analysis of Functions](/page/Cambridge%20II%20Analysis%20of%20Functions). Androma, [Geometric Measure Theory III: BV Functions and Sets of Finite Perimeter](/page/Geometric%20Measure%20Theory%20III%3A%20BV%20Functions%20and%20Sets%20of%20Finite%20Perimeter). Rudin, *Principles of Mathematical Analysis* (1976). Folland, *Real Analysis* (1999). Royden and Fitzpatrick, *Real Analysis* (2010).