Past exam of the mathematics course of the University of Cambridge / 2004 / iii / Paper 6 (Introduction to Functional Analysis)

We first prove the stronger compact-square form, in which all the one-variable functions are continuous. There are five fixed pairs of continuous functions $a_q,b_q$ on the unit interval such that every real $f\in C([0,1{]}^2)$ has a representationwith continuous one-variable $G_q$. We construct the inner functions and then prove convergence of the outer functions, rather than assume the Kolmogorov-Arnold representation theorem.

Constructing separating inner functions. Work in the Banach space $\mathcal{X}=C([0,1]{)}^{10}$ with the maximum of the ten supremum norms. For each positive integer $n$, let $U_n$ consist of tuples $(a_q,b_q{)}_{q=1}^5$ for which there are, for each $q$, finite families of closed intervals $I_{qi}$ and $J_{qj}$ of length less than $1/n$, satisfying two conditions. Each point of the unit interval belongs to at least four of the five unions ${\bigcup }_i{I}_{qi}$, and likewise for the $J$ families. Within each fixed $q$, the compact setsare pairwise disjoint for distinct pairs $(i,j)$. Each $K_{qij}$ is a closed interval, by continuity and the intermediate value theorem. These are grid-separated additive coordinates.

The set $U_n$ is open: for a tuple and its finitely many witnessing intervals, the distinct compact image intervals have a positive minimum separation. A sufficiently small uniform change in the inner functions preserves that separation, while the interval-cover conditions do not change.

To prove density, start with any ten continuous functions and any positive approximation tolerance. Choose a common fine partition so that each original function oscillates by less than a small fraction of the tolerance on a partition cell and its immediate neighbours; choose its mesh smaller than $1/(2n)$. Near every internal division point put five small, pairwise disjoint open gaps, one for each $q$. Use these gaps to divide the interval into closed towns for family $q$, including the two endpoints in towns. Gaps from different families are disjoint, so any point misses at most one family; every town has length less than $1/n$ if the gaps are sufficiently small. Do this for both coordinates.

Approximate $a_q$ and $b_q$ by functions which are constant on their respective towns, and interpolate linearly across the gaps. The towns lie near single partition cells, so uniform continuity makes these approximations as close to the original functions as desired. For each $q$, perturb the finitely many town values ${\alpha }_{qi},{\beta }_{qj}$ arbitrarily slightly so that all sums ${\alpha }_{qi}+{\beta }_{qj}$ are distinct. This is possible because each unwanted equality between distinct pairs is a proper affine hyperplane in the finite-dimensional vector space of values; finitely many such hyperplanes cannot fill any open ball. Interpolating the perturbed values still approximates the original functions, and the rectangle images are now distinct singleton sets. Thus $U_n$ is dense.

By the Baire category theorem, choose one tuple in ${\bigcap }_{n\ge 1}{U}_n$. To recall why completeness suffices here, start inside any open ball and successively choose closed balls of positive radii tending to zero, each contained in the preceding ball and in the next dense open $U_n$. Their centers form a Cauchy sequence, and its limit belongs to every ball, hence every $U_n$. Fix this tuple once and for all: it does not depend on $f$.

A quantitative approximation step. Let a continuous residual $r$ have supremum norm $M>0$. Choose $n$ large enough that the oscillation of $r$ on any rectangle of side lengths below $1/n$ is at most $M/20$. For each rectangle $I_{qi}\times J_{qj}$ choose a sample point $(x_{qi},y_{qj})$. Define a one-variable $h_q$ to have constant value $r(x_{qi},y_{qj})/3$ on $K_{qij}$. The image intervals are disjoint, so interpolate linearly in their gaps and extend constantly beyond the outermost ones. Then $h_q$ is continuous on the real line andAt a given $(x,y)$, at most one family fails to contain $x$ in an $I$ town and at most one fails to contain $y$ in a $J$ town. Thus the point belongs to rectangles in $k$ families, where $3\le k\le 5$. For each of these families the summand differs from $r(x,y)/3$ by at most $M/60$. Each remaining summand has absolute value at most $M/3$. ConsequentlyThis proves a strict contraction uniformly over the square, including points in gaps.

Summing the corrections. Set $r_0=f$ and apply that step repeatedly, subtracting the five summands at each stage. The residuals satisfy $\Vert r_m{\Vert }_{\infty }\le (3/4{)}^m\Vert f{\Vert }_{\infty }$, and the corresponding corrections satisfy $\Vert h_{qm}{\Vert }_{\infty }\le (3/4{)}^m\Vert f{\Vert }_{\infty }/3$. If a residual is zero, take all subsequent corrections to be zero. The seriesconverges uniformly on the real line to a continuous function. The residuals tend uniformly to zero, proving the boxed representation. An affine change of each coordinate treats any compact rectangle.

Removing a restriction to a compact domain. For a continuous real function $f$ on the whole plane, setChoose a positive continuous one-variable $A$ by linear interpolation of $A(n)=(n+2)(1+M(n+1))$ at nonnegative integers. Since these values increase, for $n\le t\le n+1$ we have $A(t)\ge (n+2)(1+M(n+1))\ge (1+t)(1+M(t))$. ThereforeLet $\sigma (x)=\frac{1}{2}+{\pi }^{-1}\arctan x$. Transport $b$ to the open square by $\tilde{b}(s,t)=b(\tan (\pi (s-1/2)),\tan (\pi (t-1/2)))$, and set it to zero on the square's boundary. The displayed decay estimate proves continuity at every boundary point, including the corners. Apply the compact-square representation to $\tilde{b}$ and substitute $s=\sigma (x)$, $t=\sigma (y)$ to obtain $b$ using continuous one-variable functions and addition.

Finally multiply this expression by $A(x^2+y^2)$. Multiplication itself uses only addition and one-variable functions, becauseSquaring, negation and scaling are one-variable operations, as are $A$ and $\sigma$. Thus the whole-plane reduction for continuous superposition gives a finite expression of the required kind for every continuous $f$ on the plane, without a boundedness assumption. For a complex-valued $f$, apply the argument to its real and imaginary parts and combine them using addition and the one-variable map $z\mapsto iz$. All the constituent one-variable functions can be taken continuous.

The Tychonoff theorem states that an arbitrary product of compact spaces, with the product topology, is compact; no Hausdorff hypothesis is necessary. We prove it using Zorn's lemma and the axiom of choice.

First, every proper filter on a set extends to an ultrafilter. Order its proper extensions by inclusion. The union of a chain is again a proper filter: any finite collection of its members occurs in one member of the chain, and the empty set occurs in none. Zorn's lemma gives a maximal proper filter $\mathcal{U}$. If $A\notin \mathcal{U}$, adjoining $A$ must destroy properness, so some $F\in \mathcal{U}$ has $F\cap A=\varnothing$, and hence $X\setminus A\in \mathcal{U}$. This proves the decision property of an ultrafilter.

Next prove the ultrafilter characterization of compactness. In a compact space the closed sets $\overline{F}$, for $F\in \mathcal{U}$, have the finite intersection property, so their intersection contains a point $x$. Every open neighbourhood $O$ of $x$ belongs to $\mathcal{U}$: otherwise its closed complement belongs to $\mathcal{U}$ and contains $x$, a contradiction. Thus the ultrafilter converges to $x$. Conversely, if an open cover has no finite subcover, the closed complements have the finite intersection property and generate a proper filter. Extend it to an ultrafilter. A limit point lies in some member $O$ of the cover, so both $O$ and its complement would belong to the ultrafilter. This is impossible. These arguments also prove the closed-set form of compactness directly by taking complements.

Now let $X={\prod }_{j\in J}{X}_j$. If a factor is empty, the product is empty and compact. Otherwise the axiom of choice makes $X$ nonempty. For an ultrafilter $\mathcal{U}$ on $X$, its coordinate imageis an ultrafilter. Compactness of $X_j$ supplies at least one limit $x_j$; use the axiom of choice to choose one for every $j$. A basic open neighbourhood of $x=(x_j{)}_j$ restricts finitely many coordinates. Each corresponding inverse image belongs to $\mathcal{U}$, so their finite intersection also does. Hence $\mathcal{U}$ converges to $x$ in the product topology. The ultrafilter criterion proves the theorem. The explicit choices and the maximal-filter argument are exactly the places where choice enters.

The Banach-Alaoglu theorem states that the closed unit ball of the continuous dual space $V^{\ast }$ of a real or complex normed vector space $V$ is compact in the weak-star topology $\sigma (V^{\ast },V)$. Completeness of $V$ is not required.

For each $v\in V$, let $D_v=\{z\in \mathbb{K}:|z|\le \Vert v\Vert \}$, a compact scalar disk or interval. By the Tychonoff theorem, $P={\prod }_{v\in V}{D}_v$ is compact. Inside $P$, impose the equationsEach equation defines a closed subset, since it concerns continuous coordinate projections into a Hausdorff scalar field. Their intersection $K$ is therefore compact. A point in $K$ defines a linear functional $\ell (v)=z_v$, and the coordinate bounds give $|\ell (v)|\le \Vert v\Vert$. Conversely, every $\ell \in V^{\ast }$ with $\Vert \ell \Vert \le 1$ gives such a point. Thus $K$ is exactly the image of the dual unit ball under $\ell \mapsto (\ell (v){)}_{v\in V}$. The induced product topology is precisely pointwise convergence on $V$, which is the weak-star topology. This proves

This proof uses neither the axiom of choice nor a preliminary choice of points in all the given sets. For each nonempty $A_j$, form a disjoint union $X_j=A_j\bigsqcup \{{\infty }_j\}$ with a distinguished added point, and give it the topologyThe original set $A_j$ has the indiscrete subspace topology, and is clopen. Any open cover of $X_j$ either contains $X_j$, or contains both $A_j$ and $\{{\infty }_j\}$, so it has a subcover of at most two members. Thus $X_j$ is compact by a proof requiring no choices. No Hausdorff restriction is imposed in this question, and these generally non-Hausdorff spaces are allowed.

The product $X={\prod }_{j\ge 1}{X}_j$ has the explicitly given all-infinity point, so it is nonempty without choice. By the assumed countable-product compactness it is compact. For each $j$ letThe coordinate projection is continuous and $A_j$ is closed in $X_j$, so $F_j$ is closed. Any finite intersection of the $F_j$ is nonempty: pick an element of each of those finitely many nonempty sets and put ${\infty }_k$ in every other coordinate. Finite choice follows by induction in ordinary set theory without assuming the axiom of choice.

Compactness and the finite intersection property now give ${\bigcap }_{j\ge 1}{F}_j\ne \varnothing$. Choose one point $x$ from this single nonempty set. The function $f(j)=x_j$ satisfies $f(j)\in A_j$ for every $j$, proving the axiom of countable choice.

The implication holds in choice-free set theory: countable product compactness implies countable choice. The printed lower index zero is inconsistent with the stated positive-integer index set; indexing from one merely relabels this argument and changes no conclusion.

For $x,y\in B$, write $L_x(y)=xy$ and $R_y(x)=xy$. Separate continuity and linearity make both maps bounded linear operators. For each fixed $y$,The Uniform boundedness principle therefore gives $C<\infty$ such that $\Vert L_x\Vert \le C$ whenever $\Vert x\Vert \le 1$. Scaling yieldsFor completeness, the relevant Uniform boundedness principle follows from the Baire category theorem as follows. For a pointwise bounded family $\mathcal{T}$ of bounded linear operators on a Banach space, the closed sets $E_m=\{y:{\sup }_{T\in \mathcal{T}}\Vert Ty\Vert \le m\}$ cover the space. One $E_m$ contains an open ball $B(y_0,r)$. Subtracting the bounds for $y_0$ and $y_0+h$, with $\Vert h\Vert <r$, gives ${\sup }_T\Vert Th\Vert \le 2m$. Applying this to $h=(r/2)u$, $\Vert u\Vert \le 1$, bounds every operator norm by $4m/r$.

Define the new norm byIt is a norm because $L_{x}e=x$, so $L_x=0$ implies $x=0$. The triangle inequality and scalar homogeneity follow from those of the operator norm. Moreover,so it is equivalent to the original norm and remains complete. By associativity, $L_{xy}=L_x{L}_y$, and the submultiplicativity of the operator norm provesThis is renorming a separately continuous Banach algebra. It even gives $\Vert e{\Vert }_{\ast }=1$. If the algebra is the zero space, the conclusion is immediate separately.

A complex Banach algebra is an associative complex algebra with a complete norm satisfying submultiplicativity, $\Vert ab\Vert \le \Vert a\Vert \Vert b\Vert$. A field has a nonzero identity $e$, and every nonzero element is invertible. We derive the Gelfand-Mazur theorem from these facts.

If $\Vert z\Vert <1$, the series $e+z+z^2+\cdots$ converges in the complete algebra. Multiplying its partial sums by $e-z$ and passing to the limit proves the Neumann series inverse. In particular, for any $x\in B$ and $|\lambda |>\Vert x\Vert$,This resolvent of an element tends to zero in norm at infinity; the estimate follows from the geometric bound on $\Vert x^n\Vert$.

We next prove that the spectrum of an element cannot be empty. Suppose every $\lambda e-x$ were invertible. At each ${\lambda }_0$, writeThe Neumann series on a small disk about ${\lambda }_0$ proves that $R$ is analytic there. Consequently, for each continuous linear functional $\ell :B\to \mathbb{C}$, the function $\lambda \mapsto \ell (R(\lambda ))$ is entire. It is bounded outside a large disk by the estimate above and bounded on that disk by continuity. The Liouville theorem makes it constant; its zero limit at infinity makes the constant zero.

Continuous complex linear functionals separate points of $B$. Here is the needed consequence of the Hahn-Banach theorem: for $z\ne 0$, define $g(tz)=t\Vert z\Vert$ on the real span of $z$ and extend it as a bounded real linear functional of norm one. Then $\ell (w)=g(w)-ig(iw)$ is complex linear and continuous, and $\mathrm{Re}\ell (z)=\Vert z\Vert \ne 0$. The real extension theorem is proved explicitly in Question 4(i) below. Thus $\ell (R(\lambda ))=0$ for every $\ell$ forces $R(\lambda )=0$, contradicting $(\lambda e-x)R(\lambda )=e\ne 0$. This proves the nonemptiness of the Banach-algebra spectrum.

Choose $\lambda \in {\sigma }_B(x)$. Since $B$ is a field, the noninvertible element $x-\lambda e$ must be zero. Hence every $x\in B$ is a scalar multiple of $e$. The mapis therefore a bijective unital complex-algebra homomorphism. It and its inverse are bounded, since $\Vert \lambda e\Vert =|\lambda |\Vert e\Vert$. Thus it is an isomorphism of Banach algebras. If the usual normalization $\Vert e\Vert =1$ is included in the definition, it is an isometry; without that normalization, it is still the required continuous algebra isomorphism.

The first algebra exists. Let $A=L^1(0,1)$ with truncated convolutionThe Tonelli theorem givesCommutativity follows by replacing $s$ with $t-s$; associativity follows from the Fubini theorem applied to the simplex $s,u\ge 0$, $s+u\le t$. Together with completeness of $L^1$, this makes $A$ a commutative Banach algebra, not necessarily unital. Adjoin an identity using unitization:This norm is complete and submultiplicative by the preceding bound, and $e=(1,0)$ is the identity. This is the Volterra convolution algebra on a finite interval with an identity adjoined.

Take $x=(0,1)$, where $1(t)=1$. Direct induction under convolution givesThus no power vanishes. For every $\lambda \ne 0$ the seriesconverges absolutely, since its norms have factorial denominators. Multiplication by $\lambda e-x$ telescopes to $e$, so it is an inverse. On the other hand $x$ cannot be invertible: its scalar coordinate is zero, and scalar coordinates multiply, so $xy$ can never equal $e$. HenceThis explicitly exhibits a nonnilpotent quasinilpotent element without assuming a spectral-radius formula.

The second algebra does not exist. Suppose it did, and choose a nonscalar $y$. Nonemptiness of its spectrum of an element, together with $\rho (y)=0$, gives ${\sigma }_B(y)=\{0\}$. But $y+e$ is still nonscalar: if $y+e=\alpha e$, then $y=(\alpha -1)e$. By translation of the spectrum by a scalar,This contradicts the required zero spectral radius for every nonscalar element. The contradiction already uses the spectral condition alone; the nonvanishing-power condition cannot remedy it.

Let $p$ be the Minkowski functional of $E$:The contained ball makes $E$ absorbing, so $p$ is finite, nonnegative and satisfies $p(v)\le \Vert v\Vert /ϵ$. It is positively homogeneous. To check subadditivity, if $v\in sE$ and $w\in tE$, the defining property of a convex set gives $(v+w)/(s+t)\in E$, whence $p(v+w)\le s+t$; let $s,t$ decrease to the respective infima. Thus $p$ is a sublinear functional. We have $p(e)\le 1$ for every $e\in E$. Also $p(x)\ge 1$: otherwise $x\in tE$ for some $0<t<1$, which would put $x$ in $E$ because $E$ is a convex set and $0\in E$.

We prove the dominated form of the Hahn-Banach theorem needed here. Suppose $g$ is linear on a subspace $M$ and $g\le p$ there. To extend it to $M+\mathbb{R}v$ with $v\notin M$, choose the value $c$ at $v$ betweenEvery lower candidate is at most every upper candidate, becauseTaking $m=n=0$ among the candidates shows the two endpoints are finite: the lower supremum is at least $-p(-v)$ and at most $p(v)$, and the upper infimum lies between them and $p(v)$. Define $\tilde{g}(m+tv)=g(m)+tc$. For $t>0$, the upper inequality applied to $m/t$ gives domination by $p(m+tv)$. For $t<0$, apply the lower inequality to $m/(-t)$ and multiply by $-t$. At $t=0$ domination already holds. Hence this is a dominated one-dimensional extension.

Order all dominated extensions by inclusion of their domains. Chain unions preserve linearity and domination, so Zorn's lemma gives a maximal extension. The one-dimensional argument shows its domain is the whole vector space. This proves the requisite dominated Hahn-Banach theorem, rather than invoking an unproved separation form.

On the line $\mathbb{R}x$, start with $g(tx)=tp(x)$. For $t\ge 0$ this equals $p(tx)$; for $t<0$ it is nonpositive and therefore at most $p(tx)$. Extend it to $T\le p$ on $V$. ThenApplying domination to $v$ and $-v$ also givesThus $T$ is a continuous linear functional, andThis proves separation from an absorbing convex set with no closedness assumption on $E$. The proof also gives the norm-preserving real extension theorem used earlier: take $p(v)=C\Vert v\Vert$ when the original functional has norm at most $C$.

If $F$ is empty, the zero linear functional satisfies the conclusion vacuously. Otherwise choose one $f_0\in F$, put $\delta =ϵ/2$, and form the convex setIt contains $B(0,\delta )$ because $f_0\in F$. The point $x=-f_0$ is outside $E$: membership would imply $0=f+u$ for some $f\in F$ with $\Vert u\Vert <\delta$, contradicting the assumed absence of points of $F$ in $B(0,ϵ)$.

Apply part (i). There is a continuous linear functional $T$ with $T(-f_0)\ge 1$ and $T(f+u-f_0)\le 1$ for every $f\in F$, $\Vert u\Vert <\delta$. Thus $T\ne 0$ andFor fixed $f$, take the supremum of $Tu$ over this open ball; it is $\delta \Vert T\Vert$. Consequently $Tf\le -\delta \Vert T\Vert$ for every $f\in F$. The normalizationis a continuous linear functional satisfying $Sf\ge 1$ for every $f\in F$. No compactness, closedness or attainment of the supremum on the open ball is needed.

Let $F$ be the convex hull of all terms of the sequence, meaning the set of their finite convex combinations. Suppose $B(0,ϵ)\cap F=\varnothing$. Part (ii) then gives a continuous linear functional $S$ with $Sf\ge 1$ for every $f\in F$. In particular $S{f}_n\ge 1$ for every $n$, contradicting $S{f}_n\to 0$. Therefore the ball meets $F$.

A point of that intersection is a finite combination ${\sum }_{k=1}^m{\alpha }_k{f}_{n_k}$ with nonnegative coefficients summing to one. Set $N={\max }_k{n}_k$, combine repeated indices and insert zero coefficients for omitted indices. This givesThus zero lies in the norm closure of the convex hull. The same argument applies to every tail of the sequence and, by choosing errors tending to zero, also gives the usual Mazur lemma. Completeness of $V$ is not required for this conclusion.

The hypothesis needed is precisely convergence to zero against every continuous linear functional, namely weak convergence. The PDF says only “continuous” for the testing maps. If that is interpreted as including arbitrary nonlinear continuous maps, the constant map $T\equiv 1$ makes the hypothesis impossible. The proof above establishes the substantive conclusion under the intended, weaker linear-testing hypothesis, and therefore also the implication under the literal stronger wording.