The Lumenary · First-year graduate

Reading the mathematical literature

Name: ______________________

Date: ______________________

You will learn to

Locate primary sources and surveys for a topic using arXiv, zbMATH Open or MathSciNet, and follow chains of citations backwards and forwards
Read a paper in passes, mapping its logical structure before reading proofs in detail
Reconstruct the omitted steps of a published proof, list the results it depends on, and identify where each hypothesis is used
Classify the claims in a paper as proved, cited, conditional, conjectured, heuristic or numerical, and judge what is genuinely open
Recognise that conventions differ between texts, and that preprints, peer review and computation each give a different kind of assurance

1.
A preprint posted yesterday on the arXiv claims a proof of the twin prime conjecture. You are writing a survey. What is the appropriate way to mention it?
- A.Cite it as a preprint that claims a proof, and state that the conjecture remains open until the claim has been checked.
- B.State that the twin prime conjecture has been proved, citing the preprint.
- C.Ignore it, because only peer-reviewed papers count.
- D.Wait until it appears in a prestigious journal and then treat it as certainly correct.
2.
You want reviews of the published papers on a topic, classified by subject, with their references and the later papers that cite them. Which resources are designed for this? (Choose all that apply.)
- A.zbMATH Open
- B.MathSciNet
- C.The arXiv
- D.The OEIS
3.
For $p = 29$ , Zagier's involution has the unique fixed point $(1, 1, z)$ . What is $z$ ?
4.
List the set $S = \{(x, y, z) \in \N^3 : x^2 + 4yz = 17\}$ (positive integers). How many elements does it have?
5.
Run Zagier's argument for $n = 9$ instead of a prime. Here $S = \{(1, 1, 2), (1, 2, 1)\}$ . Which step of the reconstructed proof breaks?
- A.Lemma 1: $(1, 2, 1)$ satisfies $x = y - z$ , a boundary case where Zagier's map is not defined, so the map is not an involution of $S$ .
- B. $S$ is infinite.
- C.The parity lemma fails.
- D.There is no candidate fixed point.
6.
Which of these sentences from the introduction of a (hypothetical) paper assert results that the paper itself proves? (Choose all that apply.)
- A."We prove that every sufficiently large integer of the form … is the sum of a prime and a square-free number (Theorem 1.1)."
- B."By a theorem of Siegel [12], the class number grows like …"
- C."Assuming the generalised Riemann hypothesis, we show that the bound holds for all $n \ge 2$ (Theorem 1.3)."
- D."Numerical computations up to $10^{12}$ suggest that the constant $2$ is optimal."
- E."It is well known that the exceptional set is finite."
7.
Which Lean statement says exactly what Zagier's note proves?
- A.∀ p : ℕ, p.Prime → p % 4 = 1 → ∃ a b : ℕ, a ^ 2 + b ^ 2 = p
- B.∀ p : ℕ, p % 4 = 1 → ∃ a b : ℕ, a ^ 2 + b ^ 2 = p
- C.∀ p : ℕ, p.Prime → ∃ a b : ℕ, a ^ 2 + b ^ 2 = p
- D.∃ a b : ℕ, ∀ p : ℕ, p.Prime → p % 4 = 1 → a ^ 2 + b ^ 2 = p
8.
A paper states: " $|M(n)| \lt \sqrt n$ has been verified for all $1 \lt n \le 10^{16}$ ", where $M$ is the Mertens function. What may a careful reader conclude?
- A.That there is no counterexample to the Mertens conjecture up to $10^{16}$ — and nothing beyond. In fact the conjecture is false (Odlyzko and te Riele, 1985).
- B.That the Mertens conjecture is true, since $10^{16}$ cases cannot all be coincidences.
- C.That any counterexample must be larger than $10^{16}$ , so the conjecture is "true in practice".
- D.Nothing, because computations are never reliable.
9.
Turn Zagier's proof into an algorithm. Write two_squares(p) that, for a prime $p \equiv 1 \pmod 4$ , builds $S = \{(x, y, z) : x^2 + 4yz = p\}$ (positive integers), finds a fixed point $(x, y, y)$ of the swap $(x, y, z) \mapsto (x, z, y)$ , and returns the pair (x, 2*y).
Computational — Practice

Answer key — Reading the mathematical literature

1.
(a). Describe the preprint accurately: it claims a proof, and the claim has not yet been verified. Then the status of the conjecture is still "open". Over the following months, look for responses from experts, revised versions, or publication — and for the community's verdict.
2.
zbMATH Open and MathSciNet. Both index the published literature with reviews, classify it by the Mathematics Subject Classification, and link references and citations. The arXiv hosts preprints; the OEIS is a sequence encyclopedia.
3.
$1^2 + 4 \cdot 1 \cdot z = 29$ gives $z = 7$ . Check that it is fixed: it is in the middle case ( $y - z = -6 \lt 1 \lt 2 = 2y$ ) and $(2y - x, y, x - y + z) = (1, 1, 7)$ .
4.
$x$ is odd since $p$ is odd. $x = 1$ : $yz = 4$ gives $(1,1,4), (1,2,2), (1,4,1)$ . $x = 3$ : $yz = 2$ gives $(3,1,2), (3,2,1)$ . So $|S| = 5$ — odd, as Zagier's argument predicts. The swap $(x,y,z) \mapsto (x,z,y)$ fixes $(1, 2, 2)$ , giving $17 = 1^2 + 4^2$ ; Zagier's map fixes $(1, 1, 4)$ .
5.
(a). For $(1,2,1)$ we have $x = 1 = y - z$ , so none of the three strict cases applies: Zagier's sentence silently assumes this never happens, and Lemma 1 proves it for primes ( $x = y - z$ would make $p = (x + 2z)^2$ a square). Reading a proof well means finding such hidden uses of the hypotheses.
6.
(a) and (c). (a) is an unconditional result of the paper; (c) is a conditional result — the paper proves an implication. (b) is cited, (d) is numerical evidence for a conjecture, and (e) is folklore. Annotating claims like this is the fastest way to see what a paper actually establishes.
7.
(a). (b) fails at $21$ , (c) fails at $3$ , and (d) puts $\exists$ before $\forall$ . Note that Mathlib's Nat.Prime.sq_add_sq uses the hypothesis p % 4 ≠ 3 instead, which also covers $p = 2 = 1^2 + 1^2$ : a slightly stronger, non-equivalent statement of "the same" theorem.
8.
(a). Verification up to $10^{16}$ proves the inequality for those $n$ and says nothing about larger $n$ . Odlyzko and te Riele proved in 1985 that $\limsup M(n)/\sqrt n \gt 1$ , so the conjecture fails for some (unknown, enormous) $n$ .
9.
```
def two_squares(p):
    x = 1
    while x * x < p:
        m = (p - x * x) // 4
        y = 1
        while y * y <= m:
            if y * y == m:
                return (x, 2 * y)
            y += 1
        x += 2
    return None
```
A fixed point $(x, y, y)$ of the swap means $p = x^2 + 4y^2$ , so it suffices to find odd $x$ with $(p - x^2)/4$ a perfect square. Zagier's parity argument guarantees that the search succeeds for every prime $p \equiv 1 \pmod 4$ — it is an existence proof, and this brute-force search takes about $\sqrt p$ steps per $x$ . Much faster algorithms exist, such as Cornacchia's algorithm (1908).