Brun’s theorem [1], the convergence of the sum of reciprocals of the twin primes, is the founding theorem of this paper’s worlds: I invoke it to prove that world L has positive density (Section 4), that world R’s zeta function converges at s = 1 (Section 4), and that world R’s Liouville sum converges (Section 6). In this paper, Brun 1919 is read as the first theorem ever proved about world R. The Hardy-Littlewood memoir [2] supplies the singular series constant Φ = 2 C 2 = 1.3203236 ⋯ , which appears here as the limiting increment that removal adds to the average prime gap (Section 9) and as the conditional density layer of the conjecture (Section 7.6). Dubner’s conjecture [3], catalogued with its 35 exceptional even numbers as OEIS entry A007534, asserts that every even number beyond 4208 is a sum of two twin primes; Section 8.4 decomposes it exactly into three channels modulo 6 and verifies that its exceptions are synchronized triples, one per channel. Beurling’s theory of generalized number systems [4], in the modern treatment of Diamond and Zhang, is the frame into which all the worlds of this paper fall: each is a Beurling system whose primes are the chosen set, and the Privileged World Theorem of Section 7 isolates the elementary structure that all of them share. Landau’s theorem [5] on integers composed of primes from a fixed congruence class calibrates the half-power point of the singularity spectrum in Section 7.5, where its constant is measured at 0.326 for the class 1 mod 4. Dirichlet’s theorem [6] and the Bertrand-Chebyshev theorem [7] supply the two provably infinite alphabets on which the general theorem is stress-tested; the second furnishes the counterexample world whose alphabet is infinite while its zeta converges far below s = 1 , forcing the refinement of Section 7.6. The primon-gas papers of Julia [8], Bakas-Bowick [9], and Spector [10], in which the integers are the states of a quantum gas, ζ its partition function, and the Liouville function the fermion parity, supply the statistical-mechanics vocabulary used, as labeled interpretation only, for world R’s grading, and the Bost-Connes system [11] anchors the reading of the pole at s = 1 as a phase transition, which world R provably never undergoes. Chen’s theorem [12] appears where the bilayer’s classes meet the deficit spectrum of the companion paper. Tao’s logarithmically averaged two-point Chowla theorem [13] is cited for contrast in Section 6: the parity problem it cracked in the ordinary integers does not exist in world R and is visible as a congruence in the pure lower world. The companion papers [14]-[16] supply the divisor-deficit framework, the deficit spectrum with its Gap Theorem, and the register discipline; results from them are cited, not reproved. The On-Line Encyclopedia of Integer Sequences [17] catalogs the atoms of this paper—A001359 (lower twins), A006512 (upper twins), A001097, A014574, A007534, and Section 10 records the new sequences this work contributes.

This section defines every term used in the paper, so that a student can read what follows without outside preparation.

Primes and twin pairs. A prime is an integer greater than 1 whose only divisors are 1 and itself. A twin prime pair is a pair of primes at distance 2, such as ( 11 , 13 ) . In the pair ( p , p + 2 ) , the prime p is called the lower twin and p + 2 the upper twin. The lower twins begin 3 , 5 , 11 , 17 , 29 , 41 , ⋯ (OEIS A001359); the upper twins begin 5 , 7 , 13 , 19 , 31 , 43 , ⋯ (OEIS A006512). The twin prime conjecture states that there are infinitely many such pairs. Note that the prime 5 appears in both lists: it is the upper twin of ( 3 , 5 ) and the lower twin of ( 5 , 7 ) . Lemma 1.1 below shows that 5 is the only prime with this double membership; I refer to it throughout as the double twin.

Indicators and the four position types. For any integer x , write 1 P ( x ) = 1 if x is prime and 0 otherwise. To each odd position x I attach the ordered pair ( 1 P ( x ) , 1 P ( x + 2 ) ) , which takes four possible values, written compactly as two-bit types:

In words: a position of type 11 starts a twin pair; type 10 is a prime whose upper neighbor x + 2 is composite (a lower-prime position); type 01 is a composite position whose upper neighbor is prime (an upper-prime position); type 00 has both entries composite (a composite position). Types 10 and 01 together are the singly-prime positions. These are names for the four values of the pair, nothing more.

Alphabets and worlds. An alphabet (or privileged set) is any nonempty set P of primes. The world generated by P, written W ( P ) , is the set of all positive integers whose prime factors all belong to P - that is, the number 1 (the empty product) together with every finite product of primes from P, repetition allowed. A world consists of a set of primes and the integers they generate, and nothing else. For example, W ( { 2 , 5 } ) = { 1 , 2 , 4 , 5 , 8 , 10 , 16 , 20 , 25 , ⋯ } . Unique factorization in the ordinary integers guarantees that factorization is also unique inside every world.

Membership, and integers outside both worlds. Members of a world are the integers it contains. An integer with prime factors drawn from two disjoint alphabets belongs to neither of the two worlds those alphabets generate: for the worlds L and R defined below, 15 = 3 ⋅ 5 has a factor (3) outside R’s alphabet and a factor (5) inside the removed set, so it is a member of neither L nor R. Such integers are called mixed. The two worlds are not a partition of the integers: L ∩ R = { 1 } and L ∪ R omits every mixed integer. This point matters for Theorem 8.3.

Removal. Given the set of all primes and a subset S, the removal of S means forming the complementary alphabet: all primes not in S. Removal is a precise operation on alphabets, not a metaphor: to remove the upper twins is to work with the alphabet consisting of every prime that is not an upper twin.

The four worlds of this paper. Let U denote the set of upper twin primes and T the set of lower twin primes. The paper studies:

i.e., generated by lower twins alone and upper twins alone. (Rs and R coincide; both symbols are kept because the pure pairLs, Rs is studied as a mirror pair.)

Analytic terms. The zeta function of a world is the Dirichlet series ζ W ( s ) = ∑ n ∈ W n − s , which by unique factorization equals the Euler product ∏ p ∈ P ( 1 − p − s ) − 1 wherever both converge. The reciprocal sum of an alphabet is ∑ p ∈ P 1 / p ; when it converges, so does ζ W ( 1 ) . The gap sequence of a world is the sequence of differences between consecutive members listed in increasing order; the gap sequence of an alphabet is defined the same way on the alphabet itself. The branch at m , for a member m , is the set of members divisible by m . The functions ω ( n ) and Ω ( n ) count the prime factors of n without and with multiplicity; the Liouville function is λ ( n ) = ( − 1 ) Ω ( n ) ; v 5 ( n ) is the exponent of 5 in n .

Equation numbering. Equations are numbered sequentially, with the equation set toward the center and the number at the margin, as in (1)-(4) above.

The following three lemmas collect the elementary modular arithmetic used throughout. They are stated once here, with one-line proofs, and cited wherever needed; none is a structural result of this paper.

Lemma 1.1 (Triples and the double twin). Among any three integers x , x + 2 , x + 4 , one is divisible by 3. Consequently: the only three primes in arithmetic progression with common difference 2 are 3, 5, 7; 2) the prime 5 is the only prime that is simultaneously a lower twin and an upper twin.

Proof. The three numbers cover all residues modulo 3, so one is divisible by 3; if all are prime, that one equals 3, forcing x = 3 . For (ii): if n is both a lower and an upper twin, then n − 2 , n , n + 2 are all prime, so n − 2 = 3 and n = 5 .

Lemma 1.2 (Twin residues). Let ( p , p + 2 ) be a twin pair with p > 3 . Then p ≡ 2 and p + 2 ≡ 1 ( mod 3 ) ; equivalently, for p > 3 the lower twin is ≡ 5 and the upper twin ≡ 1 ( mod 6 ) . The prime 3 is the unique twin divisible by 3, and the exceptional memberships of 3 and 5 are read off from the pairs ( 3 , 5 ) and ( 5 , 7 ) .

Proof. Neither member is divisible by 3 and they differ by 2, so they occupy the two nonzero classes; p ≡ 1 would force p + 2 ≡ 0 . Both are odd, which upgrades the mod-3 statement to mod 6.

Lemma 1.3 (Oddness and digit endings). Every twin prime is odd; hence every member of a world generated by twins alone (lower, upper, or any subset) is odd, and the sum of two such members is never such a member. An integer not divisible by 5 does not end in the digit 0 or 5.

Proof. The prime 2 is not a twin (its neighbors 0 and 4 are not prime), so twin alphabets consist of odd primes, whose products are odd; the sum of two odd numbers is even. The digit statement is divisibility by 5 read in base 10.

Section 2 measures the bilayer classification. Section 3 proves its structural theorems, citing the lemmas above for the elementary steps. Section 4 constructs worlds L and R and factors zeta, distinguishing throughout between real-axis boundary behavior (unconditional) and analytic continuation (not asserted). Section 5 develops the pure mirror worlds. Section 6 measures factor statistics and locates where the classical difficulties live. Section 7 proves the Privileged World Theorem and the singularity spectrum, with the two-layer refinement of the conjecture. Section 8 develops the twin generations and the Goldbach decomposition. Section 9 collects geometric interpretations, clearly labeled as such. Section 10 concludes.

The two singly-prime censuses above differ by exactly 1 at height 10⁶; more refined counts across the range never differ by more than 2. The following theorem says this is forced: the two singly-prime populations must always balance, because each sheet carries the same total number of primes up to boundary effects. Measurements show that the following theorem applies with the stated constant.

Theorem 3.1 (Balance). At every height X, the number of type-10 positions and the number of type-01 positions up to X differ by at most 2. Moreover, the running sum ∑ x ≤ X ( 1 P ( x ) − 1 P ( x + 2 ) ) over odd x is bounded by an absolute constant; up to 10⁶ it remains in { 1 , 2 } .

Proof. Each type-11 position contributes 1 − 1 = 0 to the running sum, each type-10 position +1, each type-01 position-1, each type-00 position 0; so the running sum equals (type-10 count) minus (type-01 count) up to X. On the other hand the sum telescopes: ∑ x ≤ X 1 P ( x ) − ∑ x ≤ X 1 P ( x + 2 ) counts the primes in the two windows [ 3 , X ] and [ 5 , X + 2 ] , which share all primes except those in the boundary intervals of length 2 at each end. Hence the difference is bounded by the number of primes in two intervals of length 2, which is at most 2.

Consecutive odd positions x and x + 2 share information: the second coordinate of the type at x (namely 1 P ( x + 2 ) ) is the first coordinate of the type at x + 2 . So not every type can follow every other; the sequence of types obeys transition rules. The measured transition table has zeros exactly where the following theorem predicts.

Theorem 3.2 (Grammar). In the sequence of types along the odd numbers, the first bit of each type equals the second bit of its predecessor. Consequently 11 can be followed only by 11 or 10; 10 only by 01 or 00; 01 only by 11 or 10; and 00 only by 01 or 00.

Proof. The type at x + 2 has first coordinate equal to the second coordinate of the type at x . If the type at x is 11 or 01, the next type has first coordinate 1, hence is 11 or 10. If the type at x is 10 or 00, the next type has first coordinate 0, hence is 01 or 00.

The grammar permits 11 to follow 11, yet the measurement found exactly one such occurrence in a million positions. Lemma 1.1 explains why: two adjacent twin positions would be three primes in arithmetic progression with common difference 2.

Corollary 3.3 (Unique11→11). The transition 11 → 11 occurs exactly once among the positive integers: at x = 3 , corresponding to the primes 3, 5, 7. Consequently, beyond this single site, consecutive twin positions differ by at least 6.

Proof. A transition 11 → 11 at position x means x , x + 2 , x + 4 are all prime, which by Lemma 1.1(i) forces x = 3 . For the spacing: twin positions at x and x + 2 are excluded beyond the triple, and twin positions at x and x + 4 would make x + 2 , x + 4 , x + 6 contain three primes spaced by 2 beginning at x unless excluded by the same lemma; hence the next twin position is at x + 6 or beyond.

The measurements show that a twin position is almost always immediately preceded by a type-01 position and immediately followed by a type-10 position (8167 and 8168 occurrences against 8169 twin positions, the deficits being boundary effects at the primordial triple). The following theorem shows this local pattern 01 ⋅ 11 ⋅ 10 is forced for every twin position beyond the triple.

Theorem 3.4(Forced neighborhood). For every twin position x > 3 : the type at x − 2 is 01, and the type at x + 2 is 10.

Proof. The type at x − 2 has second coordinate 1 P ( x ) = 1 , so it is 01 or 11; it cannot be 11, since that would put a twin position at x − 2 adjacent to the one at x , excluded by Corollary 3.3 for x > 3 . Hence it is 01. The type at x + 2 has first coordinate 1 P ( x + 2 ) = 1 , so it is 11 or 10; if it were 11 then x + 4 would be prime, and x , x + 2 , x + 4 would be three primes spaced by 2 with x > 3 , excluded by Lemma 1.1. Hence it is 10.

Corollary 3.5 (Uniqueness of the double twin). The prime 5 is the only prime that is simultaneously a lower twin and an upper twin.

Proof. Lemma 1.1(ii).

The bilayer thus compresses the twin prime conjecture, without loss, into a single census question—is the type-11 class infinite?—while its three sibling classes are settled: the singly-prime classes are infinite (if all sufficiently large primes were lower twins, chains of three primes spaced by 2 would form, contradicting Lemma 1.1; the mirror argument settles the other class), and type 00 is trivially infinite.

Splitting the primes into two alphabets splits every integer’s Euler factor into one product or the other. The following theorem records the consequence: the two worlds’ zeta functions multiply back to Riemann’s. The identity was verified numerically at s = 2 to six decimals ( 1.644934 = 1.644934 ) before being asserted.

Theorem 4.1 (Conservation). For Re   s > 1 :

Proof. The Euler product of ζ ( s ) runs over all primes. Partition the primes into U and its complement. The product of the factors over the complement is ζ L ( s ) by unique factorization in world L; the product over U is ζ R ( s ) likewise. Multiplying the two partial products restores the full product.

World L excludes only a sparse set of primes, so it should retain most integers. The next theorem quantifies this: L has positive density, given by a convergent product, and the convergence is exactly Brun’s theorem. The theorem is stated in two registers, kept separate on the referee’s principle that boundary convergence and meromorphic continuation are different things: what Brun’s theorem yields is the density, the convergence of the removal factor on the closed half-plane Re   s ≥ 1 , and a real-axis pole-type asymptotic at s = 1 ; it does not yield analytic continuation of either factor past that line, and no such continuation is asserted. Measurements show the theorem applies: the density at 10⁶ is 0.446, the partial product is 0.455, and the limiting value with the density-layer tail estimate is approximately 0.41.

Theorem 4.2 (WorldL). Write E ( s ) = ∏ q ∈ U ( 1 − q − s ) . Then:

(a) (Density.) The members of L have natural density δ L = E ( 1 ) = ∏ q ∈ U ( 1 − 1 / q ) , and this product converges to a positive number.

(b) (Boundary asymptotic.) The product E ( s ) converges absolutely and uniformly on Re   s ≥ 1 , defining a function analytic on Re   s > 1 and continuous on the closed half-plane. Consequently, as s → 1 within Re   s > 1 ,

This is a boundary statement along and near the real axis. Whether ζ L possesses a meromorphic continuation to a neighborhood of s = 1 , with (6) upgraded to a genuine simple pole, depends on continuing E past Re   s = 1 , which Brun’s theorem does not supply and which is not asserted here.

(c) (Zeros.) On Re   s > 1 , ζ L has no zeros (all Euler factors are nonvanishing there). Any comparison between the zero set of ζ L and the zeros of ζ requires an analytic continuation of E into the critical strip; under any hypothesis supplying such a continuation to a region Ω with E nonvanishing on Ω, the zero sets of ζ L and ζ would coincide on Ω. Unconditionally, no statement about zeros off Re   s > 1 is made.

(d) (Divisibility.) No member of L is divisible by 5 or by 7 (both are upper twins); hence, by Lemma 1.3, no member of L ends in the digit 0 or 5.

Proof. (a) By the sieve of the excluded primes: truncating U at height Q gives the density ∏ q ∈ U , q ≤ Q ( 1 − 1 / q ) for integers avoiding the truncated set, with the error controlled by the tail ∑ q > Q 1 / q , which tends to 0 because ∑ q ∈ U 1 / q converges by Brun’s theorem [1]. (b) For Re   s ≥ 1 , | q − s | ≤ 1 / q , so the product converges absolutely and uniformly by comparison with ∑ 1 / q ; analyticity in the open half-plane and continuity on the closed half-plane follow. The asymptotic (6) multiplies ( s − 1 ) ζ ( s ) → 1 by E ( s ) → E ( 1 ) = δ L . (c) Each factor 1 − q − s vanishes only where q − s = 1 , i.e., on Re   s = 0 , so no factor vanishes on Re   s ≥ 1 ; the conditional statement is immediate from ζ L = ζ E . (d) 5 and 7 are upper twins, hence not in L’s alphabet; the digit statement is Lemma 1.3.

World R is generated by a sparse alphabet, and Brun’s theorem again controls its analysis: the harmonic sum over the world converges, so there is no pole. What happens at s = 1 then depends entirely on whether the alphabet is infinite, and that is the twin prime conjecture. The measured partial product for ζ R ( 1 ) is 2.198.

Theorem 4.3 (WorldR). The series ζ R ( 1 ) = ∑ n ∈ N R 1 / n converges; world R’s zeta has no pole at s = 1 . Furthermore, exactly one of the following holds: (i) the twin primes are finite, in which case ζ R is a finite Euler product, a meromorphic function on the whole plane whose poles lie on the imaginary axis at the points s = 2 π i k / ln q , q ∈ U ; or (ii) the twin primes are infinite, in which case the Euler product of ζ R has infinitely many factors. The twin prime conjecture is equivalent to (ii).

Proof. Convergence at s = 1 : ζ R ( 1 ) = ∏ q ∈ U ( 1 − 1 / q ) − 1 = 1 / δ L , finite by Theorem 4.2, i.e., by Brun. The dichotomy is immediate: the number of Euler factors equals the number of upper twins, and each pair contributes exactly one upper twin. In case (i) the finite product is an elementary function; each factor ( 1 − q − s ) − 1 has poles exactly where q − s = 1 , i.e., at s = 2 π i k / ln q .

The further statement that in case (ii) the abscissa of convergence of ζ R equals 1 requires a lower bound on the density of the twins, which no theorem supplies; this is addressed honestly in Section 7.6.

Every upper twin except 5 leaves remainder 1 on division by 3 (Lemma 1.2). The next theorem converts this into a complete residue rule for all of world R. The mod-6 clause was corrected in revision: divisibility by 5 does not by itself place a member in class 5 mod 6, since 25 = 5 2 ≡ 1 ( mod 6 ) is a member of R; the correct criterion is an odd exponent of 5, and the corrected rule was verified without exception across the members to 10⁶, where the superseded criterion fails for exactly 1267 members, the first being 25 , 175 , 325 , 475 .

Theorem 4.4 (Grading). Every member n of world R satisfies

where v 5 ( n ) is the exponent of 5 in n . In particular no member is divisible by 3, and a member is congruent to 5 modulo 6 exactly when v 5 ( n ) is odd. Divisibility by 5 is necessary for the class 5 mod 6, but not sufficient: members with even positive v 5 lie in class 1 mod 6.

Proof. Write n = 5 a m with 5 ∤ m ; every prime factor of m is an upper twin other than 5, hence ≡ 1 mod 3 by Lemma 1.2, so m ≡ 1 mod 3 . Since 5 ≡ − 1 mod 3 , n ≡ ( − 1 ) a mod 3 . Both values ±1 are nonzero mod 3. For the mod-6 clause: members are odd (Lemma 1.3), and an odd number is ≡ 5 mod 6 exactly when it is ≡ 2 mod 3 , which by (7) happens exactly when v 5 ( n ) is odd. An odd v 5 entails 5 | n , giving necessity of divisibility; the members 25 , 175 , ⋯ show it is not sufficient.

Theorem 5.1 (Mirror census). At every height x , the number of lower twins up to x equals the number of upper twins up to x + 2 ; the two alphabets are equinumerous. Their intersection is exactly { 5 } .

Proof. The map p ↦ p + 2 is a bijection from lower twins to upper twins, increasing each element by 2; this gives the count statement. The intersection consists of primes that are both lower and upper twins, which is { 5 } by Lemma 1.1(ii).

Corollary 5.2 (Mirror residues). Every upper twin except 5 is ≡ 1 ( mod 3 ) ; every lower twin except 3 is ≡ 2 ( mod 3 ) ; and 3 is the unique twin divisible by 3.

Proof. Lemma 1.2.

Corollary 5.3 (Oddness). Every member of Ls and of Rs (and of R) is odd. The prime 2 belongs to neither pure alphabet.

Proof. Lemma 1.3.

Corollary 5.4 (Additive void). In each pure world, the sum of two members is never a member.

Proof. Lemma 1.3: all members are odd, sums of two members are even.

In the ordinary integers, controlling the average of the Liouville function λ ( n ) = ( − 1 ) Ω ( n ) is a problem of Prime-Number-Theorem strength, and the “parity problem”—the invisibility of λ to congruence-based sieves—is the recognized central obstruction to the twin prime conjecture [13]. Both pure worlds dispose of this problem, in opposite ways. Measurements first: in world R, the sum ∑ λ ( n ) / n over members up to 10⁶ came to 0.480 against the predicted product 0.489; in world Ls, the congruence law below was checked exhaustively on the distinct products of the small alphabet.

Theorem 5.5 (Parity by convergence, inR). The series ∑ n ∈ N R λ ( n )   n − 1 converges absolutely, with value ∏ q ∈ U q / ( q + 1 ) .

Proof. By unique factorization, the Euler factor of the series at a prime q is ∑ a ≥ 0 λ ( q a ) q − a = ∑ a ≥ 0 ( − 1 ) a q − a = ( 1 + 1 / q ) − 1 = q / ( q + 1 ) . The product over q ∈ U of ( 1 + 1 / q ) converges since ∑ 1 / q does (Brun), so the series converges absolutely to the stated product. ▫

Theorem 5.6 (Parity by visibility, inLs). Every member n of Ls not divisible by 3 satisfies

The Liouville function of the pure lower world is readable as a congruence.

Proof. Write n as a product of lower twins none equal to 3; by Lemma 1.2 each factor is ≡ − 1 ( mod 3 ) , so a product of Ω ( n ) of them is ≡ ( − 1 ) Ω ( n ) .

The two pure worlds have equal alphabets by count, yet the lower world is always at least as populous. The reason is that each lower letter is smaller than its mirror partner by exactly 2, and the advantage compounds through products.

Theorem 5.7 (Injection). For every x , the number of Ls-members up to x is at least the number of Rs-members up to x .

Proof. Extend the pair bijection multiplicatively: send ∏ p i a i ↦ ∏ ( p i + 2 ) a i . This maps Ls-members injectively to Rs-members (injectivity by unique factorization on both sides) and strictly increases every member exceeding 1. Hence every Ls-member ≤ x maps to a distinct Rs-member, and Rs-members ≤ x pull back only from Ls-members ≤ x . Counting gives the inequality.

Measurements in world R found that the members divisible by 5, when each is divided by 5, reproduce the entire member list exactly (6116 members each way at 10⁶); the same held for the divisors 7, 13, 19, and for the composite divisors 35 and 91. Measurements show that the following theorem applies; and that it applies to every world, not only to R.

Theorem 7.1 (Privileged World Theorem). Let P be any nonempty set of primes and W = W ( P ) . Then:

(a) (Division law.) For every member m ∈ W , the set of members divisible by m , with each divided by m , equals W: in symbols { n ∈ W : m | n } = m ⋅ W .

(b) (Gap covariance.) The gap sequence of the branch at m is m times the gap sequence of W.

(c) (Head law.) Let q 1 = min P . Every member of W below q 1 2 is 1 or a prime of P; consequently, for every member m , the branch at m below m q 1 2 is exactly m ⋅ ( { 1 } ∪ { p ∈ P : p < q 1 2 } ) , so the alphabet’s initial spacings appear, scaled by m , at the head of every branch.

(d) (Fixed point and tiling.) W is the least set satisfying X = { 1 } ∪ ∪ p ∈ P p X ; and the members exceeding 1 are partitioned by smallest prime factor: each lies in exactly one branch q ⋅ W ≥ q , where W ≥ q denotes the sub-world over letters ≥ q .

(e) (Recovery.) Define Λ W on W by ∑ d | n , d ∈ W Λ W ( d ) = ln n . Then Λ W is supported exactly on the powers of the primes of P, with Λ W ( p k ) = ln p ; equivalently,

the alphabet’s Dirichlet series satisfies P W ( s ) = ∑ k ≥ 1 μ ( k ) k log ζ W ( k s ) . The alphabet is therefore recoverable from the member list alone.

Proofs. (a) If n ∈ W and m | n , the multiset of prime factors of n / m is contained in that of n , hence lies in P, so n / m ∈ W ; conversely every element of m W is a member divisible by m . (b) The branch is m W by (a), an order-preserving rescaling of W by the factor m ; differences scale by m . (c) A composite member has at least two prime factors, each ≥ q 1 , hence is ≥ q 1 2 ; so below q 1 2 only 1 and primes of P remain. Apply (a) to transport this head into every branch. (d) W satisfies the set equation, since every member is 1 or p ⋅ (member) for its smallest factor p ; any set satisfying the equation contains 1 and is closed under multiplication by letters, hence contains all finite products, hence contains W; minimality follows. The tiling is the smallest-prime-factor decomposition, unique by unique factorization. (e) The defining recursion is the Möbius pair of the ordinary von Mangoldt identity restricted to W; by unique factorization in W, the standard computation gives Λ W ( n ) = ln p if n = p k with p ∈ P , and 0 otherwise; taking Dirichlet series and logarithms of the Euler product yields the displayed inversion.

[Measured] The theorem was verified exactly on: world R (divisors 5, 7, 13, 19, 35, 91; tiling 23,756 = 6116 + 3380 + 1596 + ∙∙∙ over 8169 branches; recovery of the first twenty upper twins from the list with no primality test); the arbitrary alphabet { 3 , 11 , 17 } ; the alphabet { 2 , 5 } ; the Dirichlet-infinite classes p ≡ 3 mod 4 and p ≡ 1 mod 4 ; and the Bertrand alphabet of Section 7.4.

Within world R, one branch of the self-similar structure is distinguished by the grading of Theorem 4.4.

Theorem 7.2 (Class exchange). In world R, multiplication by 5 exchanges the two residue classes modulo 3; multiplication by any other letter preserves them.

Proof. By (7), multiplying by 5 flips the sign ( − 1 ) v 5 ; multiplying by any letter ≡ 1 mod 3 leaves the residue unchanged.

[Measured] Sector sizes at 10⁶: 18,908 members in class 1 and 4849 in class 2; the exchange was verified exhaustively. In the statistical-mechanics reading of Section 9 this exchange plays the role of a two-valued charge generated by a single mode; here it is stated purely as a congruence.

It is tempting to argue that a world so perfectly self-reproducing cannot have a finite alphabet. The next proposition shows the temptation must be resisted: a world with a deliberately finite alphabet satisfies every law of Theorem 7.1.

Proposition 7.3 (Finite-alphabetcontrol). Let P = { 5 , 7 , 13 } . Then W ( P ) is an infinite set of integers satisfying the division law, gap covariance, head law, fixed-point equation, and recovery of Theorem 7.1, and the recovery yields exactly three primes at every height, forever.

Proof. Theorem 7.1 applies to every alphabet, finite or not; infinitude of W ( P ) follows from the powers of 5; the recovery statement follows from part (e), whose support is the prime powers of P, a set with three primes.

[Measured] All laws were verified on this world directly. The structural conclusion: self-similarity consumes the alphabet and cannot produce it. Furthermore, the classical escape from such an impasse is unavailable: Euclid’s proof of the infinitude of the primes leaves the multiplicative span by adding 1, and by Corollary 5.4 the pure twin worlds have no additive structure to exploit. The twin prime conjecture is, in this precise sense, an Euclid-type theorem missing from world R: a proof must construct an exit from a world shown here to have no additive door.

[Measured] Dirichlet’s theorem [6] guarantees the alphabet { p ≡ 1 mod 4 } is infinite; its world (9623 members below 10⁵) passed all laws exactly, and its counting function obeys the Landau shape N ( x ) ln x / x → 0.326 (measured at 10⁴ and 10⁵: 0.3259, 0.3265). The Bertrand alphabet—the smallest prime above each power of 2, infinite by [7]—passed all laws exactly as well, with reciprocal sum 0.7404 and ζ W ( 1 ) = 2.322 .

The real-axis boundary behavior of a world’s zeta at s = 1 measures how dense its alphabet is. Collecting the cases: the full alphabet gives the simple pole of ζ ; a positive-proportion alphabet such as p ≡ 1 mod 4 gives a half-power branch ( s − 1 ) − 1 / 2 with the Landau constant measured above [5]; the twin alphabet, at its conjectural Hardy-Littlewood density x / ln 2 x , gives the soft branch

( s − 1 ) log 1 s − 1 with amplitude Φ established in the companion paper [14]; and

the Bertrand alphabet gives no feature at s = 1 at all—its zeta converges even at s = 0.2 (measured value 9934.6). The exponent of the singularity is a meter of the alphabet’s density. All statements in this subsection concern behavior along the real axis approached from the right; none asserts continuation.

The Bertrand world forces a correction of a claim I would otherwise have overstated. Its alphabet is provably infinite, yet its zeta converges far below s = 1 : infinitude of an alphabet does not imply that the world’s zeta is singular at s = 1 . Since no theorem of mathematics bounds the density of the twins from below, the unconditional content of the twin prime conjecture is the existence layer: the Euler product of ζ R has infinitely many factors (Theorem 4.3). The identification of the boundary with s = 1 , and of the branch amplitude with Φ, is the density layer, conditional on the Hardy-Littlewood conjecture [2]. The two layers are logically independent refinements, and this paper keeps them separate.

Inside world R, ask the original question one level up: do two members ever sit at distance 2?

[Measured] Counts of such pairs: 16, 73, 359, 1874 at 10⁴, 10⁵, 10⁶, 10⁷; the normalized density P 2 ( x ) x / N ( x ) 2 stabilizes near 0.64. The first pairs: ( 5 , 7 ) ; ( 215 , 217 ) = ( 5 ⋅ 43 , 7 ⋅ 31 ) ; ( 245 , 247 ) = ( 5 ⋅ 7 2 , 13 ⋅ 19 ) ; ( 1415 , 1417 ) = ( 5 ⋅ 283 , 13 ⋅ 109 ) . In every observed pair the lower member is divisible by 5. Measurements show that the following theorem applies.

Theorem 8.1 (5-Anchor). If n and n + 2 are both members of world R, then 5 | n , and indeed v 5 ( n ) is odd.

Proof. By Theorem 4.4 no member is ≡ 0 mod 3 . The pair n , n + 2 occupies two distinct classes mod 3; avoiding 0 for both forces n ≡ 2 and n + 2 ≡ 1 ( mod 3 ) . By (7), n ≡ 2 means ( − 1 ) v 5 ( n ) = − 1 , i.e., v 5 ( n ) odd; in particular 5 | n .

The mirror world L s regenerates pairs more freely (1606 below 10⁶, normalized constant ≈1.48), with the prime 3 as a statistical anchor (92 percent of pairs have a member divisible by 3) but not an absolute one, since Ls possesses three admissible residue channels where R has one.

Iterate the construction: build a world from the upper members of the second-generation pairs, and ask for pairs again. The following theorem shows this pure continuation is impossible.

Theorem 8.2 (Extinction). Every upper member n + 2 of a second-generation pair satisfies n + 2 ≡ 1 ( mod 3 ) ; hence every product of such numbers is ≡ 1 ( mod 3 ) , and no two such products can differ by 2.

Proof. The congruence for n + 2 is part of the proof of Theorem 8.1. A product of numbers ≡ 1 mod 3 is ≡ 1 mod 3 . Two integers both ≡ 1 mod 3 differ by a multiple of 3; their difference is ≡ 0 mod 3 , never 2.

The only viable lineage carries both members of each pair forward, since the 5-anchored lower members supply the class-2 letters that a future pair requires. The mixed world holds 55,971 integers below 10⁹, and beyond the inherited pairs contains no new pair: the third generation is empty below a billion, with the smallest candidates excluded by S-unit finiteness (equations of the form 5 a 7 b ± 2 landing in the world have finitely many solutions per form). The generational census per comparable range: 8169, then 359, then 0.

The statement of this theorem was narrowed in revision: worlds L and R do not partition the integers (Section 1.2), so the residue argument excludes the class-0 position from world R but does not place it in world L—it may be occupied by a mixed integer. The measurements below now report the occupancy of that position exactly, and the exclusion is all the theorem asserts.

Theorem 8.3 (Template). Every second-generation pair ( n , n + 2 ) in world R occupies residues ( 2 , 1 ) modulo 3, and consequently the two odd positions below it occupy residues ( 0 , 1 ) at n − 2 and n − 4 respectively. The class-0 position n − 2 cannot belong to world R, since no member of R is divisible by 3 (Theorem 4.4). Nothing further about its membership follows from the residues: it may lie in world L or be a mixed integer belonging to neither world.

Proof. The residues of the pair are from Theorem 8.1’s proof. Stepping down by 2 subtracts 2 mod 3 at each step, giving n − 2 ≡ 0 and n − 4 ≡ 1 . Membership of a class-0 integer in R is excluded by Theorem 4.4. Membership in L is possible (3 belongs to L’s alphabet) but not forced: a class-0 integer with an upper-twin factor, such as 1803 = 3 ⋅ 601 with 601 an upper twin, is a mixed integer in neither world.

[Measured] Of the 359 pairs below 10⁶: the class-0 position n − 2 lies in world L in 223 cases and is a mixed integer in 136 cases (the first mixed occupants being 1803 , 2943 , 5373 , 6513 , 8043 ); it lies in world R in 0 cases, as the theorem requires. Restricting to pairs whose two lower positions both lie in L: 120 pairs extend downward to quadruples with pattern L , L , R , R , the first being ( 1 , 3 , 5 , 7 ) and ( 211 , 213 , 215 , 217 ) ; 152 extend upward to R , R , L , L , the first being ( 5 , 7 , 9 , 11 ) and ( 245 , 247 , 249 , 251 ) ; and 48 extend both ways into full hexads L , L , R , R , L , L , the first being ( 1 , 3 , 5 , 7 , 9 , 11 ) : the opening six odd numbers of arithmetic.

Dubner’s conjecture [3] states that every even number greater than 4208 is a sum of two twin primes (lower or upper). The residue laws of Corollary 5.2 split the statement into three independent channels.

Theorem 8.4 (Channel structure). Modulo 6: a sum of two upper twins (excluding the anomalies) lies in class 2; a sum of two lower twins lies in class 4; a sum of one upper and one lower twin lies in class 0. Hence, apart from representations using the anomalous letters 3 and 5, the even classes 2, 4, 0 mod 6 are served respectively by the channels upper + upper, lower + lower, and mixed.

Proof. By Corollary 5.2 and Lemma 1.2, non-anomalous upper twins are ≡ 1 and lower twins ≡ 5 mod 6 . Then 1 + 1 = 2 , 5 + 5 = 10 ≡ 4 , and 1 + 5 = 6 ≡ 0 mod 6 .

[Measured] To 10⁶: the upper + upper channel has 12 exceptional evens in its class (largest 4208), the lower + lower channel 11 (largest 4204), the mixed channel 12 (largest 4206); the union of the three exception lists reproduces Dubner’s list exactly, split by residue; and the classical exceptions verify as consecutive-even triples, one per channel and each exceptional site is a single scarcity of twins near n / 2 affecting all three channels at once.

Theorem 8.5 (Divisibility law for Goldbach in worldR). If an even number n ≡ 4 ( mod 6 ) is a sum of two members of world R, then 5 | n .

Proof. Members are ≡ 1 or 5 mod 6 (Theorem 4.4 with oddness). A sum ≡ 4 mod 6 requires both summands ≡ 5 mod 6 ; by the corrected clause of Theorem 4.4 each summand then has odd v 5 , in particular each is divisible by 5, hence so is the sum.

[Measured] 133,340 evens below 10⁶ are not sums of two R-members is equal to the density 4/15 of the class obstructed by Theorem 8.5, to four digits. Outside the obstructed class and sporadic small cases, coverage is essentially complete.