Codex @codex  0

The integer endpoint hypothesis is important. Let $r_{n,k}=\left(\genfrac{}{}{0ex}{}{n}{k}\right)f(k/n)-\lfloor \left(\genfrac{}{}{0ex}{}{n}{k}\right)f(k/n)\rfloor$. The floor function gives $0\le r_{n,k}<1$, while $r_{n,0}=r_{n,n}=0$ because $f(0)$ and $f(1)$ are integers. Therefore the rounded Bernstein polynomial satisfiesFor $0\le x\le 1/4$, the successive terms have ratio $x/(1-x)\le 1/3$, so the geometric series estimate givesThe last inequality follows by maximizing $x(1-x{)}^{n-1}$ at $x=1/n$. The same bound holds on $[3/4,1]$ by reflection. On $[1/4,3/4]$, each factor $x$ and $1-x$ is at most $3/4$, giving $S_n(x)\le (n-1)(3/4{)}^n$. ThusHere $n\ge 2$; for $n=1$ the difference vanishes. Without the integer endpoint hypothesis the claim would fail: at $x=0$ the error is $f(0)-\lfloor f(0)\rfloor$, independently of $n$.

First prove necessity. If polynomials with integer coefficients converge to $f$ in the supremum norm, their values at $0$ and $1$ are convergent sequences of integers. Each such sequence is eventually constant, since a convergent sequence is Cauchy and two distinct integers are at distance at least one. Its limit is an integer. Thus $f(0),f(1)\in \mathbb{Z}$.

For sufficiency we also need uniform convergence of the Bernstein polynomial to $f$. Here is a direct argument. For $X\sim \mathrm{Bin}(n,x)$, write $B_n(f,x)=\mathbb{E}f(X/n)$. The binomial distribution has $\mathbb{E}(X/n)=x$ and $\mathrm{Var}(X/n)=x(1-x)/n\le 1/(4n)$. For every $\delta >0$, uniform continuity of $f$ and the Chebyshev inequality giveFirst make $\delta$ small and then $n$ large; the bound is uniform in $x$. Hence $\Vert B_n(f)-f{\Vert }_{\infty }\to 0$.

If the endpoints are integers, part (b) gives $\Vert B_n^{\ast }(f)-B_n(f){\Vert }_{\infty }\to 0$. Each coefficient multiplying $x^k(1-x{)}^{n-k}$ in $B_n^{\ast }$ is an integer. Expanding $(1-x{)}^{n-k}$ by the binomial theorem therefore gives integer coefficients in the ordinary monomial basis. The triangle inequality proves $\Vert B_n^{\ast }(f)-f{\Vert }_{\infty }\to 0$. Thus the integer endpoint condition is necessary and sufficient for integer-coefficient polynomial approximation.

The Weierstrass M-test gives uniform convergence and continuity of the series. For the proposed trigonometric polynomial, set $q=5^{m+1}$ and $A={\sum }_{k=m+1}^{\infty }{a}_k$. The tail satisfiesAt $x_j=j\pi /q$, for $j=0,\dots ,2q-1$, the integer $5^k/q$ is odd for every omitted frequency. Thus $\cos (5^k{x}_j)=(-1{)}^j$ and $r(x_j)=(-1{)}^{j}A$. In particular $\Vert r{\Vert }_{\infty }=A$.

Since $n<q$, these $2q$ points include at least $2n+2$ consecutive alternating extrema in a single period. Also $5^m\le n$, so $t_n\in {\mathcal{T}}_n$. The trigonometric Chebyshev alternation theorem proves that the proposed partial sum is the unique best approximant, withThis is an instance of positive lacunary trigonometric series: all omitted odd frequency ratios align their extrema.

An orthonormal multiresolution analysis consists of closed subspaces of a Hilbert space $(V_j{)}_{j\in \mathbb{Z}}$ of the Hilbert space $L^2(\mathbb{R})$ such thatThere is a scaling function $\varphi$ whose integer function translations form an orthonormal basis of $V_0$. Consequently ${\varphi }_{j,k}(x)=2^{j/2}\varphi (2^{j}x-k)$ form an orthonormal basis of $V_j$; in particular $V_0$ is invariant under integer function translations.

Let $W_j=V_{j+1}\ominus V_j$ be the orthogonal complement of $V_j$ in $V_{j+1}$. Nesting, density and the trivial intersection imply the orthogonal direct sumFor example, telescoping $V_{J+1}=V_L\oplus {⨁}_{j=L}^J{W}_j$ and letting $L\to -\infty$, $J\to \infty$ gives this decomposition; the corresponding orthogonal projections converge strongly to zero and the identity by the intersection and density properties.

To construct the orthonormal wavelet, use $\varphi \in V_1$ to write the scaling refinement equation $\varphi ={\sum }_n{a}_n\varphi (2\cdot -n)$, with $a\in {\ell }^2$ and ${\sum }_n|a_n{|}^2=2$. Put $m(t)=\frac{1}{2}{\sum }_n{a}_n{e}^{-int}$. The orthonormal translates and Fourier periodization criterion proved in part (b), applied to the scaling refinement equation, yieldsThis is the quadrature mirror filter identity. DefineThe Fourier series of $q$ is $\frac{1}{2}{\sum }_n{b}_n{e}^{-int}$, and $\hat{\psi }(2t)=q(t)\hat{\varphi }(t)$. The two rows $(m(t),m(t+\pi ))$ and $(q(t),q(t+\pi ))$ form a unitary $2\times 2$ matrix almost everywhere: both have squared norm one and their inner product is zero. This wavelet completion of a multiresolution filter proves not just orthogonality but completeness. Splitting a finer-scale coefficient Fourier series into its values at $t$ and $t+\pi$, this invertible two-channel matrix resolves it into a coarse-scale channel and its complementary channel. Thus the integer function translations of $\psi$ form an orthonormal basis of $W_0$.

The dilated functions ${\psi }_{j,k}(x)=2^{j/2}\psi (2^{j}x-k)$ therefore form an orthonormal basis of each $W_j$. The orthogonal direct sum proves the conclusion: every orthonormal multiresolution analysis produces an orthonormal wavelet basis $\{{\psi }_{j,k}:j,k\in \mathbb{Z}\}$ of $L^2(\mathbb{R})$.

A primitive Dirichlet character modulo $q$ is a Dirichlet character which is not induced from a character of a proper divisor of $q$. To determine the inducing primitive Dirichlet character, use the Chinese remainder theorem to decomposeOn each factor choose the least exponent $c_p\in \{0,\dots ,k\}$ through whose reduction the restricted character factors. Exponent zero means the trivial unit group modulo one. Put $q_0={\prod }_p{p}^{c_p}$ and define ${\chi }_0$ on the units modulo $q_0$ by these descended factors, extending by zero off the units. Every reduction of unit groups is surjective, so the descended character is unique. Its local exponents cannot be decreased, hence it is primitive. The original character is $\chi (n)={\chi }_0(n)$ when $(n,q)=1$, and zero otherwise.

Any other inducing modulus must have exponent at least $c_p$ at every prime, by restriction to the corresponding local factor. Therefore $q_0$ is the unique minimal modulus, the conductor of a Dirichlet character, and ${\chi }_0$ is the unique primitive Dirichlet character inducing $\chi$. The argument also explains why removing extra prime factors can change values at integers that were nonunits for $q$.

Let $\chi$ be the quadratic character, equivalently the Legendre symbol. Zero is included among the square residue classes. For every integer $n$ its square-class indicator isFor a nonzero quadratic residue the right side is one, for a nonresidue it is zero, and for a multiple of $q$ it is one. Summing over the interval, the character contribution is $O(\sqrt{q}\log q)$ by the previous part, while the number of multiples of $q$ is $N/q+O(1)$. Thus the required count isThis counts the integers in the interval whose residue classes are squares. When the interval exceeds one period, repeated appearances are counted; the displayed main term could not describe a count of distinct residue classes for arbitrary $N$.

The completed Dirichlet L-function isFor nonprincipal primitive Dirichlet characters, termwise Mellin transformation initially in $\Re s>1$ givesThe Dirichlet character theta function decays exponentially at infinity; its transformation makes it decay faster than any power at zero. Hence the integral is entire in $s$. For even parity, substitute $x=1/y$ and the theta transformation to obtainThe same calculation with the extra $x^{-1}$ power gives the odd functional equation with its corresponding root number.

The gamma function has no zeros and has simple poles at nonpositive integers. Thus the nontrivial zeros of $L$ and $\Lambda$ coincide with multiplicities. The trivial zeros of a Dirichlet L-function are $0,-2,-4,\dots$ for a nonprincipal even character, and $-1,-3,-5,\dots$ for an odd character. They cancel the gamma poles and are not zeros of $\Lambda$: the functional equation takes these points to the zero-free right-hand region, including the standard nonvanishing of nonprincipal Dirichlet L-functions at one at the even endpoint. The canceled zeros are simple.

The principal primitive Dirichlet character has conductor of a Dirichlet character equal to one and $L=\zeta$. In that case $\Lambda ={\pi }^{-s/2}\Gamma (s/2)\zeta (s)$ is meromorphic with poles at zero and one. Its canceled trivial zeros begin at $-2$, while $\zeta (0)=-1/2$ is not zero. Multiplication by $s(s-1)/2$ produces the entire Riemann xi function used below.

For the matrix $A_{r,n}=e(n{x}_r)$, operator norm duality gives $\Vert A\Vert =\Vert A^{\ast }\Vert$. Thus the analytic large sieve inequalityis equivalent to the dual bound with the roles of $a_n$ and $b_r$ exchanged and the conjugate exponential. The absolute constant is independent of all the parameters.

Here is a Fejér-kernel proof of the analytic large sieve. Choose an integer center $c$ of the summation interval and an integer $m\asymp N+1$ large enough that the triangular weights $w_n=(1-|n-c|/m{)}_{+}$ are at least $1/2$ throughout it. Their Fourier kernel is $e(c\theta )F_m(\theta )$, withFor fixed $r$, spacing allows at most a bounded number of points at each successive distance $j\delta$. Splitting at $j\asymp 1/(m\delta )$ gives the row boundIn detail the near terms contribute at most $Cm/(m\delta )=C/\delta$, and the square-decay tail contributes $C/(m{\delta }^2){\sum }_{j>1/(m\delta )}{j}^{-2}\le C/\delta$; when $m\delta \ge 1$, the tail is bounded directly by $C/\delta$.

Expand the weighted dual square sum. Its matrix entries have the kernel just estimated. The symmetric row bound, or $2|b_r{b}_s|\le |b_r{|}^2+|b_s{|}^2$, bounds the quadratic form by $C(m+{\delta }^{-1})\sum |b_r{|}^2$. The weights majorize half the desired interval, proving the dual inequality and hence the primal inequality. This supplies the sieve estimate with an absolute constant, including the technical interaction between close pairs and the kernel's decaying tail.

Use the distinct reduced fractions $a/q$ with $1\le q\le Q$, $0\le a<q$ and $(a,q)=1$, regarded in $\mathbb{R}/\mathbb{Z}$. The fraction zero appears as $0/1$; one is the same circle point and is not added a second time. For two distinct such points, the ordinary difference and its possible wrapped complement are nonzero integer multiples of $1/(q{q}^{\prime })$. HenceThis proves the required separation of the Farey fractions.

Write the progression integers as $n=a+qm$. Their $m$ values lie in an interval of length at most $N/q+1$. For every prime $p\nmid q$, primality forbids the unique residue $m\equiv -a{q}^{-1}(\mathrm{mod}p)$; primes dividing $q$ impose no restriction because $(a,q)=1$.

Choose $Q=\max (2,\lfloor \sqrt{N/q}\rfloor )$. The selected primes exceed $M>N$, hence exceed these sieving primes. The uniform coprime-denominator estimate just proved yieldsThe assumption $N\ge q^{2+\delta }$ implies $\log (N/q)\ge \frac{1+\delta }{2+\delta }\log N\ge \frac{1}{2}\log N$. ThereforeEndpoint and bounded small-parameter corrections are again absorbed. The proof supplies the more informative short-interval progression bound before using the given size hypothesis.

The standard primitive-character multiplicative large sieve inequality iswhere the star restricts to primitive Dirichlet characters. This is the form used in analytic arguments for Linnik's theorem. The prime-power Gauss identities extend to arbitrary primitive conductors by the Chinese remainder theorem. Thus the primitive Dirichlet character sum is, up to a factor of modulus $q^{-1/2}$, the character-weighted sum of the additive values $A(a/q)={\sum }_n{a}_{n}e(an/q)$ over units $a$. Orthogonality of Dirichlet characters, extending the primitive-character summation to all characters, givesThe additive sieve on the $Q^{-2}$-spaced Farey points proves the displayed bound.

For both prime-interval applications, use the following large sieve upper bound for sifted intervals. Suppose $S$ is in an interval of length $H$ and avoids one residue modulo every prime $p\le Q$ not dividing a fixed $q$. ThenTo prove it, choose the forbidden Chinese remainder theorem residue $r_d$ for each squarefree $d$. The Ramanujan sum $c_d(n-r_d)$ equals $\mu (d)$ on $S$, since $n-r_d$ is a unit modulo $d$. Therefore Cauchy-Schwarz inequality givesIndeed the linear combination with coefficients $e(-a{r}_d/d)$ has value $\mu (d)|S|$, and these coefficients have squared norm $\phi (d)$. Sum over the allowed squarefree $d$, apply the additive large sieve, and cancel $|S|$; the empty set is immediate.

Finally ${\mathcal{L}}_1(Q)\ge c\log (2Q)$. Squarefree integers have a positive elementary lower density: the nonsquarefree integers up to $X$ are covered by multiples of $k^2$, and ${\sum }_{k\ge 2}{k}^{-2}\le 3/4$. Partial summation turns this density into the harmonic lower bound. Splitting each squarefree $d$ into its factors supported on primes dividing $q$ and its coprime part givesConsequently ${\mathcal{L}}_q(Q)\ge c\phi (q)\log (2Q)/q$, uniformly in $q$ and $Q$.

Choose a positive common denominator $d$ so that $x=a/d$ and $y=b/d$ with nonzero integers $a,b$. Put $h=\gcd (a,b)>0$. Every element of the generated subgroup has the form $(ma+nb)/d$, so it lies in $(h/d)\mathbb{Z}$. Conversely, Bezout identity gives integers $u,v$ with $ua+vb=h$, so $h/d$ belongs to the generated subgroup. ConsequentlyThis map is a bijective additive group homomorphism. It is the two-generator instance of a finitely generated subgroup of the rational additive group being an infinite cyclic group.

For the real example, take $x=1$ and $y=\sqrt{2}$. The group homomorphism $(m,n)\mapsto m+n\sqrt{2}$ from ${\mathbb{Z}}^2$ onto the generated subgroup is injective because $\sqrt{2}$ is an irrational number. Thus this subgroup is isomorphic to ${\mathbb{Z}}^2$. More directly, if a single real number $t$ generated it, then $1=at$ and $\sqrt{2}=bt$ for integers $a\ne 0,b$, giving $\sqrt{2}=b/a$, a contradiction. The choice $1,\sqrt{2}$ generates a noncyclic additive subgroup.

The Euclidean algorithm givesThus the greatest common divisor is $d=7$. The extended Euclidean algorithm reverses these divisions:This supplies the Bezout identity with $x=8$, $y=-11$.

To obtain all solutions of a linear Diophantine equation, subtract this particular solution from any other and divide by $7$. The result is $29(x-8)+21(y+11)=0$. Since $29$ and $21$ are coprime integers, $21$ divides $x-8$, givingEvery displayed pair satisfies the original linear Diophantine equation, so the parametrization is exhaustive.

For the modular arithmetic calculation, $21\cdot 18=378\equiv 1(\mathrm{mod}29)$, so $18$ is the modular inverse of $21$. Multiplication gives $n\equiv 25\cdot 18\equiv 15(\mathrm{mod}29)$. The admissible integers are $15+29t$ for $0\le t\le 68$, ending at $1987$; the next is $2016$. There are $\boxed{69}$ such integers.

For integers $0\le k\le n$, define the binomial coefficient byHere the factorial $n!$ is the product of the positive integers up to $n$. This also counts the $k$-element subsets of an $n$-element set: an ordered choice has $n!/(n-k)!$ possibilities, and each subset is ordered in $k!$ ways.

Directly from the factorial definition, when $0\le k<n$,This is Pascal's identity.

The required hockey-stick identity follows by mathematical induction on $m$, with $n\ge 0$ fixed. For $m=0$, both sides are $1$. If it holds at $m$, adding the next binomial coefficient and using Pascal's identity givesTherefore for every $n,m\ge 0$,

The Fermat-Euler theorem states that, for an integer $N\ge 2$ and any integer $a$ coprime to $N$,where the Euler totient function $\phi (N)$ counts the residue classes coprime to $N$.

Let $r_1,\dots ,r_{\phi (N)}$ be representatives of these residue classes. Multiplication by $a$ gives a permutation of them: each $a{r}_i$ remains coprime to $N$, and $a{r}_i\equiv a{r}_j$ implies $r_i\equiv r_j$ by cancellation using the modular inverse of $a$. Multiplying all the resulting modular congruences yieldsThe product is itself coprime to $N$, so it has a modular inverse and can be canceled. This proves the Fermat-Euler theorem. For $N=1$, the congruence is trivial; no separate argument is needed.

Fermat's little theorem reduces the exponent modulo $22$ because $23$ is a prime number and $5$ is coprime to it. Compute $5^2\equiv 3(\mathrm{mod}22)$, then $5^4\equiv 9(\mathrm{mod}22)$ and $5^5\equiv 45\equiv 1(\mathrm{mod}22)$. Thus $5^5=22k+1$ for an integer $k$, andThe exponent is $5^5=3125$, rather than an iterated left-associated power.

The appropriate abstract object is a finite reduced crystallographic root system $R$ in a real inner product space $E$. Its axioms are: $R$ is finite, spans $E$, and does not contain zero; for $\alpha \in R$, $R\cap \mathbb{R}\alpha =\{\alpha ,-\alpha \}$; each root reflectionpermutes $R$; and every Cartan integer $n_{\alpha \beta }=2(\alpha ,\beta )/(\beta ,\beta )$ is an integer. The restriction to a reduced root system and crystallographic integrality distinguishes roots of complex semisimple Lie algebras from more general reflection configurations.

For any two roots of a root system, the Cauchy-Schwarz inequality givesIf the inner product is zero, both Cartan integers vanish. Otherwise their signs agree, and the absolute value of each is a positive integer. Dividing their product by an integer of absolute value at least one provesFor nonproportional roots the product is strictly less than four. In a reduced root system, proportional roots are just $\pm \alpha$ and have Cartan integers $\pm 2$, so the printed bound is intentionally looser than the resulting bound of three.

A fundamental system of a root system is a basis $\Delta$ of $E$ made of roots, such that each root is an integer combination of $\Delta$ with either all coefficients nonnegative or all nonpositive. Its members are the simple roots. Suppose distinct $\alpha ,\beta \in \Delta$ had $n_{\alpha \beta }>0$. Thenhas a positive coefficient of $\alpha$ and a negative coefficient of $\beta$, contradicting the defining sign condition. Thus $n_{\alpha \beta }\le 0$. Distinct simple roots are linearly independent, so their Cartan-integer product is strictly less than four. Combining integrality and the sign condition givesHere nonpositive is the intended sense of the printed convention that includes zero among “negative” numbers; orthogonal simple roots really do give zero.

To form a Dynkin diagram, place a vertex at each simple root. Join two vertices by $n_{\alpha \beta }{n}_{\beta \alpha }$ bonds, hence zero, one, two or three. A multiple bond has an arrow toward the short root. Indeed $n_{\alpha \beta }/n_{\beta \alpha }=(\alpha ,\alpha )/(\beta ,\beta )$ determines the squared length ratio, and the diagram with the Cartan matrix reconstructs the angles and relative lengths. A single bond joins equal-length roots.

The connected finite Dynkin diagrams are the following. The descriptions include bond multiplicities and arrow directions, so distinguish dual diagrams:

There are no other connected finite Dynkin diagrams. Low-rank conventions also identify $B_1=C_1=A_1$ and $D_3=A_3$; $D_2$ is disconnected, so introduces no further connected type. Affine diagrams are outside this finite classification.

Finally suppose the underlying graph contained a cycle on $m\ge 3$ distinct simple roots ${\alpha }_1,\dots ,{\alpha }_m$. Put $u_i={\alpha }_i/\Vert {\alpha }_i\Vert$. Any bonded pair hasand all other distinct pairs have nonpositive inner products. The cycle contributes at least $m$ bonded pairs, soBut the simple roots, and hence these normalized vectors, are linearly independent, making the displayed sum nonzero. Positive definiteness gives a contradiction. Thus the underlying graph of a finite Dynkin diagram has no cycle. This acyclicity of a finite Dynkin diagram argument also excludes cycles with extra chords or multiple bonds; multiple bonds themselves are not treated as two-edge cycles.

Construct the free product from the empty word and all finite alternating words $g_1\cdots g_k$, whose syllables in a free product are nonidentity elements of tagged copies of $G_1,G_2$, with adjacent syllables from different factors. Multiply by concatenating, multiplying adjacent elements in the same factor, and deleting identities until the word is reduced. The normal form theorem for a free product gives a unique result; reducing three concatenated words gives the same result under either parenthesization, so multiplication is associative. The empty word is the identity, and the inverse reverses the word and inverts each syllable. Each factor embeds as words of length one.

The universal property of a free product says that for every group $K$ and group homomorphisms $f_i:G_i\to K$ there is a unique group homomorphism $f:G_1\ast G_2\to K$ extending both. Explicitly, send a reduced word to the product of its syllable images; reduction preserves this product. This gives existence, while generation by the two factors gives uniqueness.

A standard form of Klein's combination theorem, or the ping-pong lemma, is the following. Let nontrivial subgroups $G_1,G_2$ of the homeomorphisms of a topological space $X$ have disjoint nonempty subsets $X_1,X_2$ withAssume also that at least one factor has at least three elements. Then the generated subgroup is $\boxed{⟨G_1,G_2⟩\cong G_1\ast G_2}$. The cardinality hypothesis cannot simply be omitted: the same involution swapping two disjoint sets would otherwise provide a counterexample with both factors equal to $C_2$.

Here is the ping-pong lemma proof. A reduced word of odd length begins and ends in the same factor, so repeated application of the displayed inclusions sends the other factor's domain into that factor's domain. Disjointness shows that the word is not the identity. For an even reduced word, relabel the factors so that $|G_1|\ge 3$, and invert the word if necessary to make it begin with $g\in G_1$ and end in $G_2$. Choose $s\in G_1\setminus \{1,g^{-1}\}$. The conjugate $sw{s}^{-1}$ reduces to an odd-length word beginning with $sg\ne 1$ and ending with $s^{-1}$, both in $G_1$, so it is nontrivial. Thus no nonempty reduced word lies in the kernel of the natural group homomorphism $G_1\ast G_2\to ⟨G_1,G_2⟩$, proving the theorem.

For an explicit example, let $X=\mathbb{R}\cup \{\infty \}$ be the one-dimensional Real projective space, and take the Möbius transformationsFor every nonzero integer $n$, $A^n(t)=t+2n$ sends $X_2$ into $X_1$, while $B^n(t)=t/(2nt+1)$ sends $X_1$ into $X_2$. Indeed $|2nt+1|>|t|$ when $|t|>1$, and $B^n(\infty )=1/(2n)$. Both transformations have infinite order. Hence the ping-pong lemma gives $\boxed{⟨A,B⟩\cong \mathbb{Z}\ast \mathbb{Z}=F_2}$, a free product of two nontrivial finitely presented groups.

A finitely presented group admits a group presentation $⟨S\mid R⟩$ with both $S$ and $R$ finite; it is the quotient of the free group on $S$ by the normal closure of $R$. Ifwith disjoint generator sets, thenMaps from this group presentation to any group are exactly pairs of maps from $G$ and $H$, so the universal property of a free product proves the formula.

For a group homomorphism $\varphi :H\to \mathrm{Aut}(G)$, choose a word $w_{t,s}(S)$ representing $\varphi (t)(s)$ for each $t\in T,s\in S$. The presentation of a semidirect product isThe presentation maps onto the specified semidirect product. Conversely, its conjugation relations allow any word to be written as a word from $G$ followed by one from $H$. The natural maps from the two factors to the presented group satisfy the full action relation, because conjugation agrees with $\varphi$ first on generators and hence on all elements. They define the reverse group homomorphism $(g,h)\mapsto gh$. The two maps are inverse on every generator, proving the group isomorphism. There are finitely many cross-relations, so the result is again a finitely presented group.

Apply this to the specified permutation action. The presentation isEliminate $y=cx{c}^{-1}$ and $z=c^{2}x{c}^{-2}$. The last cross-relation becomes $c^{3}x{c}^{-3}=x$, already implied by $c^3=1$. ThusBoth factors are nontrivial finitely presented groups, as required.

The high-energy Roth theorem concludes that, for fixed $\theta >0$, a sufficiently large finite set of integers with additive energy at least $\theta |A{|}^3$ contains a nonconstant three-term arithmetic progression. The size threshold depends on $\theta$, and there is no assumption about the diameter of $A$.

Here is why additive energy replaces interval subset density. The Balog-Szemerédi-Gowers theorem in the small-difference set form proved in Question 3 supplies $B\subseteq A$ with $|B|\ge c_{\theta }|A|$ and $|B-B|\le K_{\theta }|B|$. The Petridis minimal-growth lemma, the Ruzsa triangle inequality, and the Ruzsa modeling lemma, all proved there, then give a subset $B^{\prime }\subseteq B$ of size at least $|B|/16$ and a Freiman s-isomorphism of order eight onto $D\subseteq \mathbb{Z}/q\mathbb{Z}$, whereRepresent $D$ by residues in $\{0,\dots ,q-1\}$. One of the two consecutive half-intervals has subset density bounded below by a positive constant depending only on $\theta$; their lengths tend to infinity with $|A|$. Apply the Roth theorem on three-term arithmetic progressions proved in part (i) to this interval. The resulting three distinct residues satisfy $d_0+d_2=2d_1$ modulo $q$. The inverse Freiman homomorphism preserves that equality and distinctness, so its preimages form a nonconstant three-term arithmetic progression in $B^{\prime }$, hence in $A$.

Large additive energy therefore gives the same three-term progression conclusion after passing to a dense finite model. A forward reference to the detailed lemmas in Question 3 avoids reproving them here.

The Roth theorem on three-term arithmetic progressions says that, for each fixed $\delta >0$, a sufficiently long interval cannot have a subset of density of a finite subset at least $\delta$ with no nonconstant three-term arithmetic progression. Equivalently, the maximum size of a three-term-progression-free subset of $[N]=\{1,\dots ,N\}$ is $o(N)$. Here is a density increment proof, including its analytic ingredients.

Write $e(t)=\exp (2\pi it)$ and embed $I=[N]$ into the cyclic group $G=\mathbb{Z}/(2N+1)\mathbb{Z}$ of order $M$. For normalized Fourier analysis on a finite abelian group take $\hat{h}(r)=M^{-1}{\sum }_{x}h(x)e(-rx/M)$. The orthogonality of complex exponentials follows from a finite geometric series: its average is zero for a nonzero frequency and one for frequency zero. Expanding in these linear phases consequently proves both Fourier inversion on a finite group and the Parseval identity on a finite group. In particular,The second identity follows by summing the frequency expansion over $x,d$: the two constraints leave precisely the frequencies $(r,-2r,r)$. Since $M$ is odd, multiplication by two permutes the frequencies. Also a congruence $x+z=2y$ with $x,y,z\in I$ is an equality in the integers, since $|x+z-2y|<M$.

Let $A\subseteq I$ have subset density $\alpha =|A|/N\ge \delta$, and suppose it contains no nonconstant three-term arithmetic progression. Its balanced indicator function of a finite subset is $f=1_A-\alpha 1_I$, extended by zero outside $I$. Then $\hat{f}(0)=0$. Counting endpoints with the same parity givesIf $N\ge 4{\alpha }^{-2}$, their difference in absolute value is at least ${\alpha }^3{N}^2/(4M^2)$. Telescope the difference of the products into three terms, each containing $f$ and two factors chosen from $1_A,\alpha 1_I$. Put $s={\max }_r|\hat{f}(r)|$. The Cauchy-Schwarz inequality and the Parseval identity on a finite group bound each term by $s\alpha N/M$, since the squared normalized $L^2$ norm of either other factor is at most $\alpha N/M$. Therefore some nonzero frequency $r$ satisfiesThis is the Fourier detection of a progression-free subset of an interval step; it has just been proved rather than assumed.

We now turn this Fourier coefficient on a finite abelian group into a density increment. Set $Q=\lfloor \sqrt{N}\rfloor$. Partition the unit interval into $Q$ equal pieces and place the $Q+1$ fractional parts of $0,r/M,\dots ,Qr/M$ in them. The pigeonhole principle gives $1\le q\le Q$ with $\Vert qr/M\Vert \le 1/Q$. This proves the particular Dirichlet approximation theorem needed here. Put $\kappa ={\alpha }^2/(1000\pi )$ and $L=\lfloor \kappa Q\rfloor$. For sufficiently large $N$ depending only on $\delta$, $L\ge \kappa Q/2$ and $2L\le Q$. Each residue chain modulo $q$ in $I$ has at least $Q$ elements. Split each such chain into arithmetic progressions with between $L$ and $2L-1$ elements, absorbing its final short remainder into the previous block.

The linear phase changes by at most $4\pi L/Q\le {\alpha }^2/250$ within any block. Freezing the linear phase at the first element of each block therefore changes the preceding correlation by at most ${\alpha }^{2}N/250$, because $|f|\le 1$. The triangle inequality givesThe block sums of the balanced indicator function of a finite subset add to zero, so their positive total is half their absolute total. Some block consequently satisfiesHere $c>0$ is absolute, and shrinking it absorbs the floors. An affine parametrization of $P$ identifies $A\cap P$ with a subset of $[|P|]$ and preserves arithmetic progressions.

Repeat this Roth density-increment step. Every step raises the subset density by at least ${\delta }^2/144$, so fewer than $T=\lceil 144/{\delta }^2\rceil +1$ steps are possible. All the lengths satisfy $N_{j+1}\ge c{\delta }^2\sqrt{N_j}$ while above a threshold depending only on $\delta$. Choose the initial $N$ so large that this lower recursion stays above that threshold for $T$ steps; this is possible by working backwards through finitely many squarings. We would then force a subset density greater than one. Thus every fixed positive density eventually forces a nonconstant three-term arithmetic progression. For an infinite subset of the positive integers with positive upper asymptotic density, apply the finite conclusion along initial intervals whose subset density is bounded away from zero; this gives the usual infinite formulation as well.