# 03 · Prerequisites for each of the 20 steps, rewound to high school *AI-drafted (Claude), November 20, 2026. Not reviewed by a mathematician. Each step's "In the note" line paraphrases remark.pdf, page 4. The "plain meaning" lines are our reading [inferred]: the note's entries are terse, and the author warns some may not make sense. Hours are rough estimates of study beyond high school needed to follow the step's idea, not to check it.* **How to read each entry** - **In the note:** a paraphrase of the timeline line - **Plain meaning:** one or two sentences - **Prerequisite chain:** from the research tool at the top down to the undergraduate course it rests on - **Rewind to high school:** the earliest idea in the high-school curriculum it grows out of, with a number you can check - **Hours:** to follow the idea, beyond high school The twelve high-school roots, which the infographic uses as color chips: | Code | High-school root | Where it shows up | |---|---|---| | **P** | Polynomials: degree caps the roots; double roots | 1, 2, 4, 10 | | **M** | Clock arithmetic (mod p), Fermat's little theorem | 1, 2, 3, 11, 17, 18 | | **D** | Derivatives (calculus) | 1, 2, 7 | | **C** | Complex numbers, the unit circle, roots of unity | 3, 8, 11 | | **G** | Geometric series | 3, 5 | | **L** | Logarithms and growth rates | 5, 6, 13, 14 | | **S** | A sum of squares is ≥ 0; dot products; Cauchy–Schwarz | 3, 4, 8, 16 | | **A** | Area of a parallelogram = determinant | 1, 4, 13, 18 | | **F** | Factorials, prime factorization, gcd | 2, 9, 10, 15 | | **T** | Summing a table by rows or by columns | 3, 12, 16 | | **H** | Pigeonhole | 11, 13 | | **Q** | Arithmetic progressions, number bases | 19 | --- ## Act I · Find the trick in the toy world ### Step 1 · Stepanov determinant · Oct 6 - **In the note:** "Stepanov determinant." - **Plain meaning:** Stepanov's 1969 proof of the finite-field RH builds a polynomial that vanishes to high order at every solution of the curve's equation. The degree caps how many solutions there can be. This step recasts the "such a polynomial exists" part as a statement about a determinant [inferred from the name]. - **Prerequisite chain:** - Stepanov–Bombieri method (Iwaniec–Kowalski §11) - ← Hasse derivatives; polynomials over the finite field F_q; Frobenius x ↦ x^q - ← abstract algebra (fields, F_p, F_q) - ← linear algebra (dimension counting; a homogeneous system with more unknowns than equations has a nonzero solution; determinants) - **Rewind to high school:** - **(P)** A degree-d polynomial has at most d roots. Counting with multiplicity, (x−1)²(x−2) has 3 roots. If a polynomial vanishes to order m at N points, then m·N ≤ d. With d = 100 and m = 10, that's at most 10 points. - **(D)** A double root is where the graph *touches* the axis: p(a) = 0 and p′(a) = 0. - **(M)** In mod 5, x⁵ − x is zero at all 5 points (Fermat). Its derivative is 5x⁴ − 1 ≡ −1, so x⁵ behaves like a constant when you differentiate. That's Stepanov's lever. - **(A)** A 2×2 system has a unique solution exactly when ad − bc ≠ 0. - **Hours:** about 300 (linear algebra, abstract algebra, finite fields, plus reading Stepanov). ### Step 2 · A derivative that dies mod p at CM points · Oct 9 - **In the note:** - At a CM curve, for a prime p that stays prime ("inert") in the CM field K, a natural derivation D satisfies D^p ≡ 0. - The a-th derivative of a theta value, Dᵃ·θ(0), divided by the a-th power of a period Ω, is divisible by p^{a/p + a/p³ + ⋯ − O(log a)}. (Read from the rendered PDF; plain-text extraction garbles this line.) - **Plain meaning:** Over finite fields, differentiating p times kills everything, and that's what makes multiplicity cheap. This step finds a *number-field* echo of that fact. At special "CM" points, a natural derivative of modular forms acts like it dies mod p. Its high derivatives at those points, scaled by a period, are divisible by a large, steadily growing power of p: about a/p factors once a is large. That's the same flavor of count as the factors of p in a!. - **Prerequisite chain:** - Modular forms; the Maass–Shimura / Serre derivative - ← CM elliptic curves (C/Z[ω], C/Z[i]) and supersingular reduction at inert primes; the Chowla–Selberg period Ω - ← algebraic number theory (how primes split in Z[i] and Z[ω]) - ← complex analysis (power series); p-adic valuation - **Rewind to high school:** - **(D)** d/dx xⁿ = n·xⁿ⁻¹, so the p-th derivative of anything carries a factor p!. - **(F)** Legendre's formula counts the powers of p in a!: ⌊a/p⌋ + ⌊a/p²⌋ + ⋯. For 100!, that's 50 + 25 + 12 + 6 + 3 + 1 = 97 twos. The note's exponent a/p + a/p³ + ⋯ has the same shape as this sum, a/p plus smaller terms, though it skips p² (it's not literally the power of p in a!). - **(M, C)** Primes ≡ 3 mod 4, such as 3, 7 and 11, are never a sum of two squares. They stay prime in Z[i]. That's what "inert" means. - **Hours:** about 450 on top of step 1 (complex analysis, algebraic number theory, modular forms basics, elliptic curves). ### Step 3 · qN = Mᵏ + R: a sign-only proof in the toy world · Oct 19 onward - **In the note:** - Use the "path quadric" x₁x₂ + x₂x₃ + ⋯ + x_{k−1}x_k. - Footnote 3 sums t(x₁)⋯t(x_k)·e_q(h·quadric) over h and the x's. - Summing over h first gives the left side. It has a definite sign from middle cohomology (via Bertini). - The right side is a main term Mᵏ (the "zero mode") plus a remainder R that is square-root small, by iterating Parseval k times. - The two signs are opposite, so there's no cancellation of the inverse roots, which finishes the proof. - Then: look for the same "automorphy" over number fields, through a theta function on a sixfold cover of GL₅ and a quadric on a CM curve. - **Plain meaning:** - Count the solutions of one cleverly chosen equation in two ways. - One way gives a number whose **sign** geometry tells you in advance. - The other way gives a big term Mᵏ plus a small remainder. - If any hidden root of the curve were bigger than √q, its k-th power would eventually outgrow everything and force the wrong sign. - That's a new, short route to the theorems of Weil and Deligne. The author calls it "new and shocking." It's the toy model for the whole project. - **Prerequisite chain:** - Trace functions and ℓ-adic sheaves; Deligne's Weil II; Bertini; Katz's methods - ← additive characters e_q(x) = e^{2πix/q}, Parseval on Z/q, quadratic forms over F_q - ← Fourier analysis on finite groups - ← linear algebra, complex numbers - For the second half: metaplectic covers and Kubota theta functions, Weil's interpretation of quadrics, the Poincaré bundle - **Rewind to high school:** - **(G, C)** 1 + r + r² + ⋯ + r^{q−1} = (r^q − 1)/(r − 1). Take r a q-th root of unity other than 1, and the sum is 0. Summing e(hm/7) over h = 0…6 gives 7 if m ≡ 0 and 0 otherwise; we checked. Summing over h first therefore *picks out* the solutions of quadric = 0. - **(S)** |z|² = z·z̄. Parseval is Pythagoras in many dimensions. - **(L)** Exponential growth, the "tensor-power trick": if r ≤ C^{1/k} for every k, then r ≤ 1. With q = 101, a root of size 101^{0.55} beats √q by a factor of 1.26 at k = 1, and by 10,201 at k = 40. - **(T)** "Sum over h first" just means choosing the order of summation. - **Hours:** about 900 on top of step 2 (Fourier analysis, algebraic geometry, an étale cohomology survey). Following the GL₅ half adds automorphic forms (see step 8). ### Step 4 · W(1_S) ≥ 0 via a determinant: "Stepanov with multiplicity one," then bootstrap · Oct 6 onward - **In the note:** Positivity of W on indicator-type test functions, proved by a determinant. Read as "Stepanov with multiplicity one again; bootstrap!!" - **Plain meaning:** - Weil (1952) showed RH is equivalent to a functional W, built from the explicit formula, being ≥ 0 on every "square" test function. - Proving W ≥ 0 for some test functions, by writing W as a determinant that can't be negative, has the same shape as Stepanov's argument with multiplicity 1. - So if multiplicity 1 gives something, multiplicity m should give more. That's the bootstrap. - **Prerequisite chain:** - Weil's explicit formula and positivity criterion - ← Fourier and Mellin transforms - ← Gram matrices and positive semi-definite matrices - ← linear algebra; real analysis - **Rewind to high school:** - **(S)** (a − b)² ≥ 0, so a² + b² ≥ 2ab. - **(S, A)** For vectors u and v, |u|²|v|² − (u·v)² ≥ 0. That's a 2×2 "Gram determinant," and it equals the *squared area* of the parallelogram the vectors span. A determinant of dot products can't be negative. - **Hours:** about 250 on top of step 1. ## Act II · Aim at the primes ### Step 5 · Switch to the smoothed prime sum · Oct 20 - **In the note:** - Work with Σ_{n≤X} Λ(n)·(X/n − 1), minus X log X, plus (1 + γ)X. - Guinand–Weil supplies the sign. A "variance" test-function trick makes things look like a trace formula. - Ideas from Mehta-type integrals, averaging over twists or at infinity. - **Plain meaning:** - Fix the exact target. Weight each prime power n ≤ X by ln p·(X/n − 1). - The total is X ln X − (1 + γ)X, plus a part controlled entirely by the zeros. - RH is about how small that leftover part is. - **Prerequisite chain:** - Explicit formula (Guinand–Weil), Mehta/Selberg integrals (random matrices), trace-formula shape - ← analytic number theory (Chebyshev ψ, the Prime Number Theorem, Mellin inversion) - ← complex analysis (contour integrals, residues) - ← calculus - **Rewind to high school:** - **(L)** The area under 1/x from 1 to n is ln n. - 1 + 1/2 + ⋯ + 1/n ≈ ln n + 0.5772. That constant is Euler's γ. - Primes and prime powers: Λ(8) = ln 2 and Λ(6) = 0. - We computed the leftover at X = 10⁶. It's 11.04, against main terms of size 10⁷. - **Hours:** about 500 (complex analysis, analytic number theory). ### Step 6 · One test function, a strong conclusion · Oct 8 - **In the note:** W applied to (the indicator of (0, 1]) + (about X^η)·(the indicator of (0, 1/X]) is ≥ 0, and that gives "> 1 − η." - **Plain meaning:** - Proving W ≥ 0 for one simple two-step test function would already give a strong statement labeled "> 1 − η," for small η. - The note doesn't say which quantity exceeds 1 − η. Two natural readings [inferred]: - the share of zeros on the critical line (about 41% until August 2026, when an Anthropic report raised the bound to 67.2%) - a zero-free region near Re s = 1 (no "quasi-RH" half-plane is known) - Either would have been a major result on its own. - **Prerequisite chain:** - Weil functional; Mellin transform of indicator functions (∫₀¹ x^{s−1} dx = 1/s) - ← zero-density and proportion results (Selberg, Levinson, Conrey) - ← complex analysis - **Rewind to high school:** - **(L)** Step functions. Small exponents: X^{0.01} grows, but slowly. - Inequality bookkeeping: if A + B ≥ 0 and you know B, then A ≥ −B. - **Hours:** about 100 on top of steps 4–5. ### Step 7 · Determinants of GGP periods: interlacing gives 0 or 1 · Oct 14 - **In the note:** Determinants for W-values containing the desired periods, namely GGP periods. Values 0 or 1 mean interlacing weights, or values up to m for non-interlacing ones. - **Plain meaning:** - A **period** is an integral of an automorphic form over a smaller symmetry group. - Gan–Gross–Prasad describe exactly when such a period can be nonzero. For unitary groups U(n) ⊂ U(n+1), a piece shows up once (multiplicity 1) exactly when two lists of weights **interlace**, and otherwise not at all. - The step builds determinants from these periods, so the 0/1 pattern acts like multiplicity, with "up to m" allowed in the non-interlacing case. - **Prerequisite chain:** - GGP conjecture, Ichino–Ikeda formula, automorphic periods - ← representation theory of compact Lie groups (highest weights, branching rules, Gelfand–Tsetlin patterns) - ← group theory, linear algebra (eigenvalues of symmetric matrices) - **Rewind to high school:** - **(D, P)** Rolle's theorem: between two roots of p there's a root of p′, so their roots **interlace**. For p with roots 1, 3, 4, 7, we computed that p′ has roots 1.71, 3.51, 6.02. - Cauchy's interlacing for eigenvalues is the same picture. - **Hours:** about 600 (representation theory, an introduction to automorphic forms). ## Act III · Build multiplicity from chains ### Step 8 · Iterated theta lifts: multiplicity m, and m ≈ √X · Nov 15 - **In the note:** Iterated theta lifts U(X₀) → U(X₁) → ⋯ → U(X_m) between totally definite unitary groups, with π ↦ π ⊞ μ_j[t]. This "wins multiplicity m." Take m about √X. - **Plain meaning:** - A **theta lift** transports an automorphic form from one symmetry group to a partner group, through a theta function built from a quadratic (here Hermitian) form. - Doing it m times in a chain adds one new piece, a character μ_j, to the form's "parameter" at each step. - Each added piece plays the role of one more order of vanishing in Stepanov's argument. - This is the central idea Claude proposed: chain length √X, mirroring Stepanov's multiplicity √q [primary]. - **Prerequisite chain:** - Theta correspondence (Howe duality), the Weil representation, Arthur parameters and isobaric sums ⊞, Hecke characters - ← automorphic forms on unitary groups over CM fields - ← representation theory, algebraic number theory, modular forms - **Rewind to high school:** - **(S)** Count the ways to write n as a sum of two squares. 5 = 1² + 2², which gives 8 ways with signs and order. Put these counts in one series: θ(q)² = 1 + 4q + 4q² + 4q⁴ + 8q⁵ + ⋯. That's the ancestor of every theta function. - **(C)** Rotations keep x² + y² fixed. For complex z, z·z̄ = |z|² is a "Hermitian form," and the symmetries that preserve it are the unitary group. - **(L)** Why √X: it's the balance point. See step 14. - **Hours:** about 1,200 on top of steps 2 and 7 (automorphic forms, theta correspondence). This is the hardest single prerequisite. ### Step 9 · Norms, level and cancellation, with divisibility from Chowla–Selberg, Weyl and Siegel · Nov 15 - **In the note:** Norms of the lifts, the level, and cancellation of order-m terms, with divisibility proved via Chowla–Selberg, Weyl and the Siegel mass formula. - **Plain meaning:** - Measure each lift's size (its norm; Rallis's formula turns this into an L-value) and its "level." - Show the large order-m terms cancel, leaving quantities provably divisible by high powers of primes. - The divisibility comes from three classical exact formulas: Chowla–Selberg (CM periods as Gamma values), Weyl's dimension formula, and Siegel's mass formula. - **Prerequisite chain:** - Rallis inner product formula, Siegel–Weil formula, L-values - ← Chowla–Selberg, the Weyl dimension formula, the Siegel mass formula - ← the Gamma function, representation theory, quadratic forms - **Rewind to high school:** - **(F)** Γ(n) = (n − 1)! fills in the factorials. - Counting up to symmetry: the mass formula counts each shape weighted by 1/(number of its symmetries), like counting necklaces. - Prime factorization tells you exactly what divides what. - **Hours:** about 500 on top of step 8. ### Step 10 · A lower bound on divisibility of the m-fold lift · Nov 15 - **In the note:** Sufficient lower bound on divisibility of the m-fold iterated lift. - **Plain meaning:** Prove that the final quantity is divisible by at least a certain large power. That's the number-field stand-in for "vanishes to order m." - **Prerequisite chain:** p-adic valuations, integral structures on spaces of forms ← step 9. - **Rewind to high school:** - **(F, P)** 96 = 2⁵·3, so 2 divides 96 five times. - A nonzero number divisible by p^m has size at least p^m. Being highly divisible means being big, or else being zero. That's the integer version of "a polynomial vanishing to high order is either big in degree or zero." - **Hours:** about 100 on top of step 9. ### Step 11 · An Iwasawa tower as the analog of F_{q^n}/F_q · Oct 6 - **In the note:** An Iwasawa tower, by analogy with a pigeonhole argument on F_{q^n} over F_q. - **Plain meaning:** - Over a finite field you can count again in the bigger fields F_{q^n}, for every n. Pigeonhole or power-sum arguments over n then squeeze the hidden roots. - Over the rationals, the closest analog is a **tower** of fields, such as Q(ζ_p) ⊂ Q(ζ_{p²}) ⊂ ⋯. - Iwasawa built his theory in 1959 on exactly this analogy. - **Prerequisite chain:** - Iwasawa theory (Z_p-extensions) - ← cyclotomic fields, Galois theory - ← algebraic number theory; Turán's power-sum method - **Rewind to high school:** - **(H)** Pigeonhole. - **(C, M)** Roots of unity: the solutions of z^{p^n} = 1 form a growing tower as n grows. - Powers of a point on the unit circle eventually come back close to 1, by pigeonhole on angles. - **Hours:** about 500. ### Step 12 · Seesaw: the period becomes a Bessel period · Nov 15 - **In the note:** By a seesaw, the period of the iterated lift is a Bessel period that doesn't depend on the Hecke characters (Rallis). - **Plain meaning:** - A **seesaw identity** (Kudla) computes one integral in two ways, by swapping which pair of groups you integrate over first. - Here it turns the hard period into a more standard one (a "Bessel period") that no longer depends on the characters chosen in the chain. - **Prerequisite chain:** seesaw dual pairs, Bessel and Fourier–Jacobi periods ← theta correspondence (step 8) ← Fubini's theorem. - **Rewind to high school:** **(T)** Sum a table by rows or by columns and you get the same total. In our 3×4 table, both give 13. The seesaw is that idea for integrals over groups. - **Hours:** about 200 on top of step 8. ## Act IV · Pay the bills ### Step 13 · Bost slope inequality · Oct 6 - **In the note:** "Bost slope inequality." - **Plain meaning:** - Bost's inequality from Arakelov geometry compares the "arithmetic size" (slopes) of lattices across an injective map. - It's the modern, vector-space form of "a nonzero integer is at least 1," and it's the standard engine for auxiliary-polynomial arguments over number fields. - **Prerequisite chain:** - Arakelov geometry (Hermitian vector bundles over Spec Z, heights) - ← geometry of numbers (Minkowski), Siegel's lemma - ← algebraic number theory, linear algebra - **Rewind to high school:** - **(A)** The area of the parallelogram spanned by (a, b) and (c, d) is |ad − bc|. - Lattice points. - A nonzero integer n has |n| ≥ 1. - **(H)** Siegel's lemma is pigeonhole applied to integer vectors. - **(L)** Slopes are log-volumes per dimension. - **Hours:** about 400. ### Step 14 · The cost of each lift: the ⅔ wall · Nov 15 - **In the note:** - A precise computation of the "cost" of each lift showed that a priori the method stops at ⅔. - The attempted fix was to get a main term and either transfer the constant to a function field or kill it by twisting, which needed "division." - **Plain meaning:** - Each lift in the chain costs a little. - Added up, the naive method only proves the error is below about X^{2/3}, not X^{1/2}. - Closing the gap from ⅔ to ½ needed extra ideas (steps 15–19). - **Prerequisite chain:** asymptotic bookkeeping ← steps 8–10. - **Rewind to high school:** - **(L)** Minimize A/m + B·m. The best m is √(A/B) and the minimum is 2√(AB), from AM–GM. - A cartoon of the wall [toy, inferred]: if the saving is X/m and each link costs X^{1/3}, the best you get is X^{2/3}. If links were free (cost 1), you'd get X^{1/2} with m = √X. That's why the chain length is √X. - **Hours:** about 50 on top of steps 8–10. ### Step 15 · Glue the integral structures across Hecke characters for a "gcd" · Nov 15 - **In the note:** Glue together an integral structure across Hecke characters for a "gcd," meaning extra averaging. - **Plain meaning:** Don't rely on one chain of characters. Combine a whole family so the guaranteed divisibility is the gcd across the family. Averaging over many choices buys the extra saving. - **Prerequisite chain:** Hecke characters, modules over rings of integers ← algebraic number theory. - **Rewind to high school:** - **(F)** gcd. Bézout: gcd(12, 18) = 6 = 12·(−1) + 18·1. - Averaging reduces noise: the average of n coin flips wobbles by about 1/√n. - **Hours:** about 150. ### Step 16 · Relative trace formula, Cauchy–Schwarz, near-orthogonality · Oct 14 - **In the note:** - A relative trace formula, seen in a different half-plane argument. - It uses the automorphy of μ from the sixfold cover of GL₅. - Cauchy–Schwarz reduces it to bounding the norm of a projected Poincaré series. - The columns of the integral structure are approximately orthogonal. - **Plain meaning:** - A **trace formula** equates a sum over automorphic forms (the "spectral side") with a sum over geometric data. - Using one, plus Cauchy–Schwarz, the needed bound reduces to the size of one specific function. - Approximate orthogonality keeps the cross-terms small. - **Prerequisite chain:** - Relative trace formula (Jacquet), Poincaré series, metaplectic forms - ← the Selberg trace formula, Hilbert spaces - ← functional analysis, automorphic forms - **Rewind to high school:** - **(S)** Cauchy–Schwarz: |u·v| ≤ |u||v|. - Perpendicular vectors: |u + v|² = |u|² + |v|² (Pythagoras). For nearly perpendicular vectors it nearly holds. - **(T)** A Poincaré series is "periodization": add up f(x + n) over all integers n to get a periodic function. - **Hours:** about 700. ### Step 17 · A weakened Vandiver conjecture · Nov 15 - **In the note:** For the classes in H¹(O_E[1/ℓ], Q_ℓ/Z_ℓ(−e)) that couldn't be killed by twisting, fight for a significant weakening of Vandiver. A footnote says this already gives vanishing of a number of K-groups of Z. - **Plain meaning:** - Some arithmetic obstructions (Galois-cohomology classes) survived all the twisting. - To rule them out, the proof establishes a weak form of Vandiver's conjecture, a roughly 175-year-old open question about class numbers of cyclotomic fields. - The author notes that this is "still pretty strong" in its own right [primary]. - **Prerequisite chain:** - Galois cohomology, Kummer theory, K-theory of Z, Lichtenbaum's conjectures - ← class groups and Iwasawa theory (step 11) - ← algebraic number theory - **Rewind to high school:** - **(F)** 6 = 2·3 = (1 + √−5)(1 − √−5). In some number systems, factorization isn't unique. The class number measures how badly. - Bernoulli numbers come from 1^k + 2^k + ⋯ + n^k. - Kummer's Fermat story: 37 is the first prime dividing a Bernoulli numerator. That makes it the first "irregular" prime. - **Hours:** about 800. ### Step 18 · Covolumes, theta-series index, congruences between lifts and forms · Nov 15 - **In the note:** Computation of covolumes, the theta series' index, and congruences between lifts and forms. - **Plain meaning:** Exact volume computations (how big a fundamental domain is), lattice indices, and congruences between different automorphic forms, all needed to make the divisibility bookkeeping exact. - **Prerequisite chain:** covolumes of arithmetic groups (Siegel, Prasad), congruences of modular forms ← modular forms, lattices. - **Rewind to high school:** - **(A)** The area of a lattice's fundamental parallelogram. - **(M)** Congruences: Ramanujan's τ(n) ≡ σ₁₁(n) mod 691. We checked n ≤ 8; for example, τ(6) = −6,048 ≡ 171 ≡ σ₁₁(6) mod 691. - **Hours:** about 300. ### Step 19 · Choose the tower using Behrend · Oct 19 - **In the note:** Precise choice of the Iwasawa tower, specifically using Behrend. The author says intuitively it will be unnecessary [primary]. - **Plain meaning:** To pick the parameters of the tower, they needed a large set of choices with **no three in arithmetic progression**. Behrend's 1946 "integer points on a sphere" construction supplies one. - **Prerequisite chain:** additive combinatorics (Salem–Spencer, Behrend, Roth) ← elementary number theory. - **Rewind to high school:** - **(Q)** Arithmetic progressions (3, 7, 11). - Write numbers in base d. - Points on a sphere x² + y² + z² = r: the midpoint of two of them lies strictly inside, so no three are in a line, so there's no progression. - Honest numbers: below 729, the simple base-3 set has 64 elements. Behrend's set below 6,561 has only 30. Behrend only wins for astronomically large ranges. - **Hours:** about 40. ## Act V ### Step 20 · GRH · Nov 16 - **Plain meaning:** The pieces assemble, and the Lean formalization is reported complete [primary]. Unverified by outside experts as of November 20 [secondary]. - **Prerequisites:** everything above. --- ## The course map in hours (rough, beyond high school) | Milestone | Adds | Cumulative hours | |---|---|---| | **M1 · Understand the statement of RH** | Calculus 150, complex numbers and intro complex analysis 100, elementary number theory 60 | **~310** | | **M2 · Understand the toy world** (finite-field RH, Stepanov; steps 1, 4) | Linear algebra 120, abstract algebra 150, finite fields 40, reading Stepanov 40 | **~660** | | **M3 · Follow Acts I–II** (steps 2, 3, 5–7) | Full complex analysis 120, analytic number theory 200, Fourier analysis 100, algebraic number theory 150, modular forms 200, representation theory 150 | **~1,580** | | **M4 · Follow Acts III–IV at the idea level** (steps 8–19) | Automorphic forms 400, theta correspondence 300, Iwasawa theory 250, Arakelov basics 150, étale cohomology survey 200, trace formulas 200, K-theory overview 80, additive combinatorics 30 | **~3,190** | | **M5 · Check the proof** | Mastery of all of the above, plus reading the Lean | **6,000+**, and the author says he himself lacks mastery of higher-rank automorphic forms [primary] | For comparison, the Ramanujan track in [the Ramanujan–Langlands folder](../ramanujan-langlands/03-coursework.md) is about 2,350 hours to read the corollary.