Which Millennium Problem to Attack First: The Decision Analysis

On April 5, 2026, the Profiled autonomous discovery system completed a systematic decision analysis comparing BSD (Birch and Swinnerton-Dyer conjecture) against P vs NP as the primary focus for concentrated Millennium Prize attack. The recommendation: BSD Conjecture. The decision process involved ten independent criteria, deep mathematical DNA analysis, honest assessment of existing results, and an expected value calculation that includes "plutonium" — valuable publishable research that emerges even if the main proof fails.

This article documents the complete decision analysis, exactly as produced, without smoothing the rough edges or hiding the fact that the P vs NP "proof" in the system's portfolio has UNVERIFIED_CLAIMS status with circularities found.

The Honest Truth: Starting from Reality

Before the decision analysis, the system audited the actual status of its existing work on both problems:

The contrast is stark: one problem has a "proof" with circular reasoning and status UNVERIFIED_CLAIMS; the other has a verified 66.9% score and a clear research direction. The decision analysis begins from this asymmetry.

Mathematical DNA: Structural Comparison

The deep analysis produced a "mathematical DNA" comparison — a characterisation of the true structural nature of each problem and what it implies for attack strategies:

The "natural tools" entry is the most informative. For BSD, the p-adic methods are a genuine unifying framework: p-adic L-functions on one side, p-adic Galois representations on the other, and the Iwasawa main conjecture connecting them. There is a research program with known techniques, known sub-goals, and known partial results (Wiles, Kolyvagin, Perrin-Riou, Kato, Skinner-Urban).

For P vs NP, the "Natural tools" entry is literally a question mark. After 40 years of effort by the world's best complexity theorists, three distinct barriers have been identified that block all existing approaches:

• Relativization barrier (Baker-Gill-Solovay 1975): P vs NP cannot be resolved by techniques that relativize — i.e., that work the same way whether or not an oracle is present. Most known diagonalization arguments relativize.
• Natural proofs barrier (Razborov-Rudich 1994): Any circuit lower bound proof with certain "natural" properties (constructivity and largeness) can be turned into an algorithm for breaking pseudorandom generators — which would imply P = NP, creating a contradiction. Most known lower bound techniques are natural.
• Algebrization barrier (Aaronson-Wigderson 2009): Algebraic extensions of relativization (using low-degree polynomials) cannot resolve P vs NP. This blocks the arithmetic geometry approaches that might naively seem to transfer from BSD.

Finding a P vs NP proof requires inventing a technique that avoids all three barriers simultaneously. No such technique is known. No research program credibly claims to have a direction around them. This is the honest state of affairs, and it drives the decision.

The Ten-Criterion Comparison: Complete Table

Result: BSD wins 10-0-1. The single tie is expected value — the P vs NP prize is $1M like BSD, but the higher uncertainty means higher maximum expected value if a proof is found (unlikely), creating rough equivalence in pure prize calculation. Every other dimension favours BSD.

BSD: The Mathematical Path Forward

The Birch and Swinnerton-Dyer conjecture states that for an elliptic curve E/ℚ, the rank of E(ℚ) (the number of independent rational points of infinite order) equals the order of vanishing of the L-function L(E, s) at s = 1. This is a profound prediction connecting algebra (the rational points group) with analysis (the L-function's behaviour at a specific complex number).

Phase 1 Attack Plan (Months 1–3)

The Phase 1 plan has three parallel research streams:

BSD PHASE 1: FOUNDATIONS (Months 1-3)
════════════════════════════════════════════════════════════

STREAM 1: MOTIVIC COHOMOLOGY
  ├─ Voevodsky's motivic cohomology for elliptic curves
  ├─ H¹_mot(E, ℚ(1)) and its relation to E(ℚ) ⊗ ℚ
  ├─ Bloch-Kato formulation of BSD:
  │    Tam(E) · |Ш(E)| · Ω · R
  │    ─────────────────────────── = L^(r)(E,1) / r!
  │    |E(ℚ)_tors|² · [E(ℚ):ℤ-generators]
  └─ Goal: understand the Euler system machinery

STREAM 2: p-ADIC HODGE THEORY
  ├─ Fontaine's period rings: B_dR, B_crys, B_HT
  ├─ p-adic periods: comparison isomorphism
  │    H¹_dR(E/ℚ_p) ⊗ B_dR ≅ H¹_ét(E, ℚ_p) ⊗ B_dR
  ├─ The p-adic L-function L_p(E, s) via interpolation
  └─ Goal: understand Perrin-Riou's p-adic regulator

STREAM 3: PERFECTOID METHODS
  ├─ Scholze's perfectoid spaces
  ├─ Tilting correspondence: E/K ↔ E/K♭
  ├─ Almost mathematics: the almost étale theory
  └─ Goal: understand how perfectoid relates to BSD
       via the prismatic cohomology connection

MILESTONE: Complete literature map of Bloch-Kato for GL₂
════════════════════════════════════════════════════════════

The Bloch-Kato Formulation: Why It Is the Right Framework

The Bloch-Kato conjecture is the generalisation of BSD to arbitrary motives. For an elliptic curve E over ℚ, BSD is the Bloch-Kato conjecture for the motive H¹(E)(1). The Bloch-Kato formulation:

ord_{s=1} L(E, s) = dim_ℚ H¹_f(ℚ, V_p(E))

where V_p(E) = T_p(E) ⊗ ℚ_p is the p-adic Tate module
and H¹_f(ℚ, V_p(E)) is the Selmer group (Bloch-Kato definition)

The p-adic methods "unify both sides" precisely because the p-adic L-function, the p-adic Selmer group, and the p-adic period all live in the same p-adic universe. The Iwasawa main conjecture — proved by Kato (2004) and extended by Skinner-Urban (2014) — is a deep result that connects the p-adic L-function with the Selmer group in p-adic families. It is the single most important partial result toward BSD and it sits squarely in the p-adic framework.

The Plutonium Concept: Valuable Results That Do Not Require Full BSD

The term "plutonium" refers to valuable publishable results that are generated as byproducts of the BSD attack, even if the main proof fails. This is a crucial part of the expected value calculation:

The expected value with plutonium: $2M–$6M. This range accounts for the prize ($1M for BSD proof, probability 40–50%) plus the research value of publishable results generated even on a failed proof attempt (probability 80%). The lower bound $2M assumes no prize but several strong publications. The upper bound $6M assumes the prize plus multiple high-impact papers from subsidiary results.

The Full ASI Arsenal Deployed for BSD

The April 5, 2026 decision document specifies deploying the complete autonomous discovery arsenal. Twelve primary engines:

Plus seven additional engines: MetaLearningOrchestrator (meta-learning over the entire system), RecursiveSelfImprovement (system-level improvement), ConsciousnessEmergenceEngine (emergent cross-domain insight generation), and four more supporting engines for context management, memory, evaluation, and reporting.

Why Not P vs NP? The Full Case

For completeness, and because the system's own P vs NP work had UNVERIFIED_CLAIMS status, the case against P vs NP as the primary focus deserves full articulation:

1. Three simultaneous barriers: No technique is known that avoids relativization, natural proofs, and algebrization simultaneously. Finding such a technique is itself a major open problem. Attacking P vs NP requires first solving this meta-problem.
2. No continuous gradient: Unlike BSD (where 66.9% is a real, informative score) or Riemann (where 97.2% reflects genuine numerical evidence), P vs NP has no natural notion of "partial progress" that can guide an evolutionary search. Either a lower bound circuit proof avoids all three barriers or it does not. There is no intermediate score.
3. Circular reasoning in existing attempt: The system found circularities in its own P vs NP work. This is not a recoverable situation — a circular proof must be restructured from scratch, not patched.
4. No computational validation loop: BSD can be tested against millions of curves. P vs NP has no analogous dataset. The hypothesis "P ≠ NP" cannot be tested numerically in any informative way.
5. Lower synthesis probability: 20% vs 50% for BSD. The synthesis proof strategy that worked for Navier-Stokes (assume blowup, derive contradiction) does not have a natural analogue for P vs NP.

The conclusion is not that P vs NP should be abandoned — it is a critically important problem, and a proof or disproof would be one of the most significant events in the history of mathematics. The conclusion is that given the current state of knowledge, tools, and verified progress, BSD is the better target for a concentrated attack program in 2026.

The Decision in One Sentence

BSD has a clear path (Bloch-Kato via p-adic methods), real existing progress (66.9% honest score, 90 hypotheses), known powerful tools (Iwasawa theory, perfectoid spaces, Euler systems), computational validation available (millions of curves in the Cremona database), and high plutonium value (6 categories of publishable results). P vs NP has none of these properties. BSD wins 10-0-1. The full ASI arsenal is deployed for BSD.