The Central Idea
In elementary mathematics, a function is often studied one at a time. Functional analysis instead studies entire families of functions.
The central shift is
\[ \boxed{ \text{individual functions} \longrightarrow \text{spaces whose points are functions} }. \]In such a space, we can ask how far apart two functions are, whether a sequence of functions converges, whether limits remain inside the space, and how linear transformations act on the functions.
Functions as Vectors
Let \(X\) be a set and let
\[ f,g:X\to\mathbb{R}. \]Define addition pointwise:
\[ (f+g)(x)=f(x)+g(x), \]and scalar multiplication by
\[ (cf)(x)=cf(x). \]Under these operations, many collections of functions form vector spaces.
Closure Is Essential
A collection of functions forms a vector space only when the required operations keep us inside the collection.
For example, the set of continuous functions on \([0,1]\) is closed under addition and scalar multiplication:
\[ f,g\in C([0,1]) \Longrightarrow f+g\in C([0,1]), \]and
\[ c\in\mathbb{R} \Longrightarrow cf\in C([0,1]). \]The Space \(C([a,b])\)
The notation
\[ C([a,b]) \]denotes the vector space of continuous real-valued functions on the closed interval \([a,b]\).
A natural norm on this space is the supremum norm:
\[ \boxed{ \|f\|_\infty = \max_{x\in[a,b]}|f(x)| }. \]The maximum exists because \([a,b]\) is compact and \(|f|\) is continuous.
The Supremum Distance
The supremum norm induces the metric
\[ \boxed{ d_\infty(f,g) = \|f-g\|_\infty = \max_{x\in[a,b]}|f(x)-g(x)| }. \]This distance measures the largest vertical separation between two functions across the entire interval.
Uniform Convergence Is Norm Convergence in \(C([a,b])\)
A sequence \(f_n\) converges uniformly to \(f\) on \([a,b]\) exactly when
\[ \|f_n-f\|_\infty\to0. \]Thus
\[ \boxed{ f_n\to f\text{ uniformly} \iff f_n\to f\text{ in the supremum norm} }. \]Completeness of \(C([a,b])\)
The normed space
\[ \bigl(C([a,b]),\|\cdot\|_\infty\bigr) \]is complete.
That means every Cauchy sequence in the supremum norm converges uniformly to a continuous function in \(C([a,b])\).
Therefore,
\[ \boxed{ C([a,b]) \text{ is a Banach space} }. \]Why Completeness Matters
Suppose an approximation procedure produces functions
\[ f_1,f_2,f_3,\ldots \]that form a Cauchy sequence in a normed function space.
If the space is complete, then the approximations converge to an element that still belongs to the same space.
Completeness prevents limiting procedures from leaving the analytical environment in which they were constructed.
Continuous Functions on Compact Spaces
More generally, if \(K\) is a compact topological space, then
\[ C(K) \]denotes the continuous scalar-valued functions on \(K\).
The supremum norm is
\[ \|f\|_\infty = \sup_{x\in K}|f(x)|. \]Compactness guarantees boundedness, and \(C(K)\) is complete in this norm.
The Space of Bounded Functions
Let \(X\) be any set. The space
\[ \ell^\infty(X) \]can be used for bounded scalar-valued functions on \(X\):
\[ \ell^\infty(X) = \left\{ f:X\to\mathbb{R}: \sup_{x\in X}|f(x)|<\infty \right\}. \]Equipped with the supremum norm,
\[ \|f\|_\infty = \sup_{x\in X}|f(x)|, \]this is a Banach space.
Sequence Spaces
An infinite sequence
\[ x=(x_1,x_2,x_3,\ldots) \]can be viewed as a function
\[ x:\mathbb{N}\to\mathbb{R}. \]Therefore sequence spaces are special examples of function spaces.
The Space \(\ell^1\)
Define
\[ \ell^1 = \left\{ x=(x_n): \sum_{n=1}^{\infty}|x_n|<\infty \right\}. \]Its norm is
\[ \boxed{ \|x\|_1 = \sum_{n=1}^{\infty}|x_n| }. \]The space \(\ell^1\) is complete, so it is a Banach space.
The Space \(\ell^2\)
Define
\[ \ell^2 = \left\{ x=(x_n): \sum_{n=1}^{\infty}|x_n|^2<\infty \right\}. \]The norm is
\[ \boxed{ \|x\|_2 = \left( \sum_{n=1}^{\infty}|x_n|^2 \right)^{1/2} }. \]The Inner Product on \(\ell^2\)
For real sequences \(x,y\in\ell^2\), define
\[ \langle x,y\rangle = \sum_{n=1}^{\infty}x_ny_n. \]Then
\[ \|x\|_2 = \sqrt{\langle x,x\rangle}. \]Since \(\ell^2\) is complete in this inner-product norm,
\[ \boxed{ \ell^2 \text{ is a Hilbert space} }. \]The General Sequence Space \(\ell^p\)
For
\[ 1\le p<\infty, \]define
\[ \ell^p = \left\{ x=(x_n): \sum_{n=1}^{\infty}|x_n|^p<\infty \right\}. \]The norm is
\[ \boxed{ \|x\|_p = \left( \sum_{n=1}^{\infty}|x_n|^p \right)^{1/p} }. \]Every \(\ell^p\) space with \(1\le p<\infty\) is a Banach space.
The Space \(\ell^\infty\)
Define
\[ \ell^\infty = \left\{ x=(x_n): \sup_n|x_n|<\infty \right\}. \]The norm is
\[ \boxed{ \|x\|_\infty = \sup_n|x_n| }. \]This space is also Banach.
The Space \(c_0\)
Define
\[ c_0 = \{x=(x_n):x_n\to0\}. \]Every convergent-to-zero sequence is bounded, so
\[ c_0\subseteq\ell^\infty. \]With the supremum norm,
\[ \boxed{ c_0 \text{ is a closed subspace of }\ell^\infty } \]and is therefore Banach.
Finite-Support Sequences
Let
\[ c_{00} \]denote the sequences having only finitely many nonzero terms.
This is a vector space, but it is not complete in the usual \(\ell^p\) norms.
For \(1\le p<\infty\),
\[ \overline{c_{00}}^{\,\ell^p} = \ell^p. \]From Sequence Spaces to \(L^p\) Spaces
The sequence-space norm
\[ \left( \sum |x_n|^p \right)^{1/p} \]suggests the integral norm
\[ \left( \int_X|f|^p\,d\mu \right)^{1/p}. \]Replacing counting measure by a general measure turns \(\ell^p\) into the broader family of \(L^p\) spaces.
The Space \(L^p\)
For
\[ 1\le p<\infty, \]define
\[ L^p(X,\mu) = \left\{ f: \int_X|f|^p\,d\mu<\infty \right\}, \]where functions equal almost everywhere are identified.
The norm is
\[ \boxed{ \|f\|_p = \left( \int_X|f|^p\,d\mu \right)^{1/p} }. \]Why We Identify Functions Almost Everywhere
Suppose
\[ f=g \quad\text{a.e.} \]Then
\[ |f-g|^p=0 \quad\text{a.e.}, \]so
\[ \|f-g\|_p=0. \]A genuine norm must satisfy
\[ \|u\|=0 \Longrightarrow u=0. \]Therefore \(L^p\) treats almost-everywhere equal functions as the same vector.
The Space \(L^\infty\)
A measurable function belongs to \(L^\infty(X,\mu)\) if it is essentially bounded.
The norm is the essential supremum:
\[ \boxed{ \|f\|_\infty = \operatorname*{ess\,sup}_{x\in X}|f(x)| }. \]Values attained only on null sets do not affect this norm.
Essential Supremum
The essential supremum of \(|f|\) is the smallest number \(M\) such that
\[ |f(x)|\le M \]almost everywhere.
For example, if \(f=0\) except at one point where \(f=1000\), then with respect to Lebesgue measure,
\[ \|f\|_\infty=0. \]The exceptional point has measure zero.
Hölder's Inequality
Let
\[ 1Then
\[ \boxed{ \int_X|fg|\,d\mu \le \|f\|_p\|g\|_q }. \]The exponents \(p\) and \(q\) are called conjugate exponents.
Cauchy-Schwarz as the \(p=2\) Case
When
\[ p=q=2, \]Hölder's inequality becomes
\[ \boxed{ \int_X|fg|\,d\mu \le \|f\|_2\|g\|_2 }. \]This is the Cauchy-Schwarz inequality for \(L^2\).
Minkowski's Inequality
For
\[ 1\le p<\infty, \]Minkowski's inequality states
\[ \boxed{ \|f+g\|_p \le \|f\|_p+\|g\|_p }. \]This is the triangle inequality required for \(\|\cdot\|_p\) to be a norm.
Completeness of \(L^p\)
For every
\[ 1\le p\le\infty, \]the space
\[ L^p(X,\mu) \]is complete under its standard norm.
Therefore,
\[ \boxed{ L^p \text{ is a Banach space for }1\le p\le\infty }. \]The Special Role of \(L^2\)
The space \(L^2(X,\mu)\) has the inner product
\[ \langle f,g\rangle = \int_X f(x)g(x)\,d\mu \]for real-valued functions, with the complex conjugate inserted in the complex case.
Since
\[ \|f\|_2 = \sqrt{\langle f,f\rangle} \]and \(L^2\) is complete,
\[ \boxed{ L^2 \text{ is a Hilbert space} }. \]Not Every \(L^p\) Space Is a Hilbert Space
For \(p\ne2\), the standard \(L^p\) norm generally does not arise from an inner product.
Thus
\[ L^1, \qquad L^3, \qquad L^\infty \]are Banach spaces, but not Hilbert spaces under their standard norms.
Function-Space Convergence Depends on the Norm
The same sequence of functions can behave differently under different norms.
For example,
\[ f_n(x)=x^n \]on \([0,1]\) satisfies
\[ f_n\to0 \quad\text{in }L^1, \]because
\[ \|f_n\|_1 = \frac1{n+1}\to0. \]But
\[ \|f_n\|_\infty=1 \]for every \(n\), so the convergence is not uniform.
Norm Choice Encodes What Counts as Small
Under the supremum norm, a function is small only when it is small everywhere except possibly nowhere.
Under an \(L^p\) norm, large errors may be tolerated on sufficiently small sets.
Choosing a function space and norm therefore determines what type of approximation matters.
Inclusions on Finite Measure Spaces
Suppose
\[ \mu(X)<\infty. \]If
\[ 1\le pOn finite-measure spaces, stronger integrability implies weaker integrability.
The Inclusion Can Fail on Infinite Measure Spaces
On spaces of infinite measure, there is no general inclusion
\[ L^q\subseteq L^p \]
merely from \(p
Functions may decay enough to belong to one \(L^p\) space but not another,
and behavior near zero and at infinity can produce different outcomes.
Differentiable Function Spaces
The notation
\[ C^1([a,b]) \]denotes functions with a continuous first derivative.
More generally,
\[ C^k([a,b]) \]consists of functions whose derivatives through order \(k\) are continuous.
A Norm on \(C^1([a,b])\)
One useful norm is
\[ \boxed{ \|f\|_{C^1} = \|f\|_\infty+\|f'\|_\infty }. \]This norm controls both the function and its derivative.
With this norm, \(C^1([a,b])\) is complete.
Polynomial Subspaces
Let
\[ \mathcal{P} \]denote the vector space of all real polynomials.
Restricted to \([a,b]\), polynomials form a subspace of
\[ C([a,b]). \]However, the polynomial subspace is not closed in the supremum norm.
Weierstrass Approximation Theorem
The Weierstrass Approximation Theorem states that every continuous real-valued function on \([a,b]\) can be approximated uniformly by polynomials.
Symbolically,
\[ \boxed{ \overline{\mathcal{P}}^{\,\|\cdot\|_\infty} = C([a,b]) }. \]Thus polynomials are dense in \(C([a,b])\).
Dense Subspaces
A subset \(D\) of a normed space \(X\) is dense if
\[ \overline{D}=X. \]Equivalently, every element of \(X\) can be approximated arbitrarily closely by elements of \(D\).
Dense subspaces are valuable because simpler objects can represent or approximate more complicated ones.
Separability
A metric space is separable if it contains a countable dense subset.
For example, \(C([a,b])\) is separable in the supremum norm.
For \(1\le p<\infty\), standard \(L^p\) spaces on familiar sigma-finite measure spaces such as intervals in \(\mathbb{R}\) are also separable.
A Caution about \(L^\infty\)
Unlike \(L^p\) for finite \(p\), \(L^\infty\) is often nonseparable.
This illustrates an important lesson: spaces that look similar by notation may have significantly different structural properties.
Closed Subspaces
If \(Y\) is a closed linear subspace of a Banach space \(X\), then \(Y\) is itself Banach under the inherited norm.
This principle is frequently used to prove completeness of function spaces by embedding them into a larger known Banach space.
Finite-Dimensional versus Infinite-Dimensional Spaces
In finite-dimensional vector spaces, all norms are equivalent and closed bounded sets are compact.
Infinite-dimensional function spaces behave differently.
For example, the closed unit ball in an infinite-dimensional Banach space is generally not compact in the norm topology.
Why Infinite Dimension Changes Geometry
A function space may require infinitely many independent directions to describe its elements.
This allows bounded sequences to spread across new directions indefinitely without producing a norm-convergent subsequence.
Many finite-dimensional geometric intuitions therefore require revision in functional analysis.
Direct Sums and Product Spaces
Function spaces can be combined.
If \(X\) and \(Y\) are normed spaces, then ordered pairs
\[ (x,y)\in X\times Y \]form a vector space.
One possible norm is
\[ \|(x,y)\| = \|x\|_X+\|y\|_Y. \]Product constructions are useful when a system contains several coupled function-valued quantities.
Weighted Function Spaces
Sometimes different regions of a domain should contribute differently to a norm.
A weighted \(L^p\) expression has the form
\[ \|f\|_{L^p(w)} = \left( \int_X|f(x)|^p w(x)\,d\mu \right)^{1/p}, \]where \(w\) is a nonnegative weight.
Appropriate conditions on \(w\) determine the resulting function-space structure.
Function Spaces in Differential Equations
A differential equation is often solved not by searching among all possible functions, but by choosing a function space whose elements possess the required regularity and integrability.
For example, one may ask for solutions in
\[ C^1, \qquad C^2, \qquad L^2, \]or more advanced Sobolev spaces.
The choice of space is part of the mathematical model.
Function Spaces in Approximation
Approximation theory asks how closely a complicated function can be represented by simpler functions.
But “closely” depends on the chosen norm.
Uniform approximation uses
\[ \|\cdot\|_\infty, \]while mean-square approximation uses
\[ \|\cdot\|_2. \]Different norms produce different best approximations.
Function Spaces in Fourier Analysis
Fourier methods naturally live in function spaces.
In \(L^2\), orthogonality and projection permit functions to be decomposed into frequency components much as vectors are decomposed along orthogonal coordinate directions.
This is one reason Hilbert-space structure is so important.
Function Spaces in Probability
Random variables are measurable functions on a probability space.
If a random variable \(X\) satisfies
\[ \mathbb{E}|X|^p<\infty, \]then it belongs to an \(L^p\) space over that probability space.
Thus expected-value and variance calculations fit naturally into the same function-space framework.
Common Error: Confusing a Function with Its Function Space
A function is one element.
A function space is a collection of functions equipped with additional structure.
The distinction is analogous to the difference between one vector and the vector space containing it.
Common Error: Ignoring the Norm
Saying
\[ f_n\to f \]“in a function space” is incomplete unless the relevant topology or norm is understood.
The same sequence may converge in \(L^1\) but fail to converge in the supremum norm.
Common Error: Treating \(L^p\) Elements as Strict Pointwise Functions
In \(L^p\), two measurable functions that differ only on a null set represent the same element.
Pointwise values on null sets therefore cannot be used to distinguish \(L^p\) elements.
Common Error: Assuming Every Banach Space Is Hilbert
A Banach space is complete with respect to a norm.
A Hilbert space is complete with respect to a norm induced by an inner product.
Every Hilbert space is Banach, but not every Banach space is Hilbert.
Common Error: Assuming Closed and Bounded Means Compact
The Heine-Borel characterization is a finite-dimensional phenomenon.
In infinite-dimensional normed spaces, a set may be closed and bounded without being compact.
Common Error: Assuming All \(L^p\) Inclusions Hold Everywhere
On finite-measure spaces,
\[ q>p \]gives
\[ L^q\subseteq L^p. \]On infinite-measure spaces, this implication may fail.
The measure of the underlying space is part of the hypothesis.
A Reliable Procedure for Identifying a Function Space
- Identify the domain and codomain of the functions.
- State the regularity or integrability condition defining membership.
- Check closure under addition and scalar multiplication.
- Identify the norm, metric, or inner product being used.
- Determine the associated convergence notion.
- Check completeness, separability, or other structural properties when relevant.
A Reliable Procedure for Checking \(L^p\) Membership
- Verify measurability of \(f\).
- Form \(|f|^p\).
- Compute or estimate \[ \int_X|f|^p\,d\mu. \]
- If the integral is finite, conclude \(f\in L^p\).
- For \(p=\infty\), check essential boundedness instead.
A Reliable Procedure for Comparing Two Function Norms
- Write both norms explicitly.
- Identify what each norm measures: maximum error, average error, square error, or another quantity.
- Use an inequality such as Hölder, Cauchy-Schwarz, or a finite-measure estimate when available.
- Check whether the domain has finite or infinite measure.
- Do not infer equivalence of norms from finite-dimensional intuition.
A Reliable Procedure for Testing Completeness
- Take an arbitrary Cauchy sequence in the stated norm.
- Identify a candidate limit using pointwise, coordinatewise, or measure-theoretic arguments.
- Show the candidate belongs to the same function space.
- Show convergence occurs in the original norm.
- Alternatively, realize the space as a closed subspace of a known Banach space.
Example: Comparing \(L^1\) and \(L^\infty\)
On \([0,1]\), let
\[ f_n(x)=x^n. \]Then
\[ \|f_n\|_1 = \int_0^1x^n\,dx = \frac1{n+1}\to0. \]But
\[ \|f_n\|_\infty=1. \]Therefore,
\[ \boxed{ f_n\to0\text{ in }L^1 \quad\text{but not in }L^\infty }. \]Example: An \(\ell^p\) Membership Test
Consider
\[ x_n=\frac1n. \]For \(p=1\),
\[ \sum_{n=1}^{\infty}\frac1n \]diverges, so
\[ x\notin\ell^1. \]For \(p=2\),
\[ \sum_{n=1}^{\infty}\frac1{n^2} \]converges, so
\[ \boxed{ x\in\ell^2 \text{ but }x\notin\ell^1 }. \]Example: An \(L^p\) Membership Test
On \((0,1)\), consider
\[ f(x)=x^{-\alpha} \]with \(\alpha>0\).
We have
\[ |f(x)|^p=x^{-\alpha p}. \]The integral
\[ \int_0^1x^{-\alpha p}\,dx \]is finite exactly when
\[ \alpha p<1. \]Therefore,
\[ \boxed{ x^{-\alpha}\in L^p(0,1) \iff \alpha p<1 }. \]Practice 1: Vector-Space Structure
Explain why
\[ C([0,1]) \]is closed under addition and scalar multiplication.
Practice 2: Supremum Norm
For
\[ f(x)=x^2-1 \]on \([-1,1]\), compute
\[ \|f\|_\infty. \]Practice 3: \(\ell^1\) Membership
Determine whether
\[ x_n=\frac1{2^n} \]defines an element of \(\ell^1\).
Practice 4: \(\ell^2\) Membership
Determine whether
\[ x_n=\frac1{\sqrt n} \]belongs to \(\ell^2\).
Practice 5: \(L^2\) Norm
Let
\[ f(x)=x \]on \([0,1]\). Compute
\[ \|f\|_2. \]Practice 6: \(L^\infty\) Norm
Let \(f(x)=x^2\) on \([-2,1]\). Compute
\[ \|f\|_\infty. \]Practice 7: Almost-Everywhere Equivalence
Let \(f=0\) on \([0,1]\), and let
\[ g(x)= \begin{cases} 10, & x=\frac12,\\ 0, & x\ne\frac12. \end{cases} \]Compare \(f\) and \(g\) as pointwise functions and as elements of \(L^2([0,1])\).
Practice 8: Convergence Depends on the Norm
Let
\[ f_n(x)=x^n \]on \([0,1]\). Determine whether \(f_n\to0\) in \(L^1\) and whether \(f_n\to0\) in the supremum norm.
Practice 9: \(L^p\) Membership
Determine for which \(p\ge1\) the function
\[ f(x)=x^{-1/4} \]belongs to \(L^p(0,1)\).
Practice 10: Identify the Error
A student says:
Since every closed and bounded subset of \(\mathbb{R}^n\) is compact, every closed and bounded subset of a Banach space is compact.
Explain the error.
Practice Answers
Practice 1
If \(f,g\in C([0,1])\), then the sum \(f+g\) is continuous. If \(c\in\mathbb{R}\), then \(cf\) is continuous. Therefore the space is closed under the two vector-space operations.
Practice 2
On \([-1,1]\),
\[ f(x)=x^2-1 \]ranges from \(-1\) to \(0\). Hence
\[ |f(x)|\le1, \]with equality at \(x=0\). Therefore,
\[ \boxed{\|f\|_\infty=1}. \]Practice 3
We compute
\[ \sum_{n=1}^{\infty}\frac1{2^n}=1. \]Therefore,
\[ \boxed{x\in\ell^1}. \]Practice 4
We have
\[ \sum_{n=1}^{\infty} \left|\frac1{\sqrt n}\right|^2 = \sum_{n=1}^{\infty}\frac1n. \]The harmonic series diverges. Therefore,
\[ \boxed{x\notin\ell^2}. \]Practice 5
We have
\[ \|f\|_2 = \left( \int_0^1x^2\,dx \right)^{1/2} = \left(\frac13\right)^{1/2}. \]Thus
\[ \boxed{\|f\|_2=\frac1{\sqrt3}}. \]Practice 6
On \([-2,1]\),
\[ x^2\le4, \]with equality at \(x=-2\). Therefore,
\[ \boxed{\|f\|_\infty=4}. \]Practice 7
Pointwise, \(f\ne g\) because they differ at \(x=1/2\).
However, the singleton \(\{1/2\}\) has Lebesgue measure zero, so
\[ f=g \quad\text{a.e.} \]Therefore they represent the same element of \(L^2([0,1])\).
Practice 8
We have
\[ \|f_n\|_1 = \int_0^1x^n\,dx = \frac1{n+1}\to0. \]Thus \(f_n\to0\) in \(L^1\).
But
\[ \|f_n\|_\infty=1 \]for every \(n\), so \(f_n\) does not converge to \(0\) in the supremum norm.
Practice 9
We need
\[ \int_0^1 x^{-p/4}\,dx < \infty. \]This occurs exactly when
\[ \frac p4<1. \]Therefore,
\[ \boxed{ f\in L^p(0,1) \iff 1\le p<4 }. \]Practice 10
The statement “closed and bounded implies compact” is valid in finite-dimensional Euclidean spaces, but not in general infinite-dimensional Banach spaces.
Functional analysis therefore requires separate compactness arguments rather than direct transfer of the finite-dimensional Heine-Borel theorem.
What to Remember
Function spaces treat functions as vectors. Once a norm is chosen, distance and convergence become structural:
\[ d(f,g)=\|f-g\|. \]Important examples include
\[ C([a,b]), \qquad \ell^p, \qquad L^p, \qquad L^\infty. \]For
\[ 1\le p<\infty, \]the \(L^p\) norm is
\[ \boxed{ \|f\|_p = \left( \int_X|f|^p\,d\mu \right)^{1/p} }. \]All standard \(L^p\) spaces for \(1\le p\le\infty\) are Banach spaces, while \(L^2\) is additionally a Hilbert space.
The most important structural habit is: always identify the function class, the norm or inner product, and the resulting notion of convergence before drawing conclusions.
Where We Go Next
Once functions are organized into normed and inner-product spaces, the next question is how transformations act on those spaces.
The next lesson, Operators and Linear Functionals, studies linear maps between function spaces, boundedness, continuity, operator norms, kernels, ranges, dual spaces, and the special role of linear functionals in functional analysis.