PRH | Inter | 5.9 • Blurred Spectral Reconstruction and the Hecke Principle

Blurred Spectral Reconstruction and the Hecke Principle Galois Theory as Harmonic Analysis, Anabelian Data as Spectral Probes Aleksandar Perišić December 2025 Abstract Thesis. The abelian side of Galois theory is harmonic analysis in disguise: characters are frequencies, class field theory is Pontryagin duality, and Hecke/Dirichlet L-functions are Mellin–Fourier coefficients that separate arithmetic. The anabelian side admits no single “DFT,” so we work spectrally by probes (representations, traces, kernels) that extract spectra while averaging (“blurring”) other structure. The central question becomes: how much can we still recover? Where this lands. We (i) formalize the abelian/harmonic picture; (ii) set up a sharp→blurry ladder for anabelian probes; (iii) state a Hecke Principle (blur-as-method) with a concrete protocol; (iv) give guardrails for “exists” steps; and (v) end with a practical credo for using blur to read spectra. 1 Abelian Galois = harmonic analysis (the discrete picture) Cyclotomic/DFT intuition For N≥1,Gal(Q(ζN)/Q)≃(Z/NZ)×is finite abelian. Its continuous characters χ: (Z/NZ)×→ C×are the 1-D irreps (the “frequencies”), with the usual orthogonality relations. Dirichlet characters and their L-functions L(s, χ)diagonalize congruence classes the way a discrete Fourier transform diagonalizes translations. Class field theory as Pontryagin duality Global class field theory identifies Gal(Kab/K)∼ =\ CK/H, the (profinite) dual of a quotient of the idele class group CK; continuous characters of CK parametrize abelian extensions, and Hecke L-functions arise from these characters via Tate’s Poisson summation on the adeles. In short: the spectrum (characters) is the abelian Galois group. Finite fields and Frobenius Over Fq, additive characters x7→ e2πi TrFq/Fp(x)/p give a literal Fourier transform on Fn q, while the Galois group is generated by Frobenius. Exponential sums are spectral data of Frobenius acting on cohomology. 2 What changes in the anabelian world There is no single transform that diagonalizes a general profinite π1or G= Gal(L/K). Instead we use a ladder of probes—each extracts spectra but blurs something. 1 A ladder from sharp to blurry 1. Full Tannakian data (no blur). The neutral Tannakian category of ℓ-adic local systems (with ⊗) reconstructs π1; all tensor relations are kept. 2. All matrix coefficients (noncommutative Fourier). Peter–Weyl on compact groups: L2(G)decomposes into irreps with their matrix coefficients; keep them all ⇒no loss. 3. Characters/pseudocharacters (conjugacy blur). Keeping only traces collapses each conjugacy class to one value. Nevertheless: Fact (Brauer–Nesbitt + Chebotarev).Semisimple Galois representations with equal Frobenius traces on a density-1set of primes are isomorphic (so traces still determine the semisimplification). 4. Artin/automorphic L-functions. Further packaging into Euler products loses the ability to vary test functions, but strong multiplicity-one phenomena often still pin down objects from almost-all local eigenvalues. 5. One zeta only (heavy blur). Dedekind ζKforgets a lot: distinct (nonisomorphic) fields can be arithmetically equivalent with the same ζKand splitting data. Where the analysis lives Two master tools implement and control the blur: •Trace formula (global): a nonabelian Poisson summation. Test functions act as windows: sharper spectral localization induces more geometric averaging (and vice versa). •Heat/Hecke kernels (local/global):et∆or Hecke semigroups damp high eigenvalues by e−tλ;tis a literal blur scale. As t↓0the blur disappears. 3 Blur as a method: what we gain, what we lose Fix a probe (representation, trace, test kernel); call the resulting averaging the blur. Two rules of thumb match practice across number theory, dynamics, and probability: •Positivity + approximate identity simplify convergence/identification; the price is loss of fine configuration data (phases, extensions). •Recoverability depends on the probe. Traces give semisimple data; full matrix coefficients or ⊗-structure let you reconstruct the group. Remark (A precise arithmetic instance).On the explicit-formula channel, even tests fproduce asmoothed identity X ρb f(ρ) = X p,k≥1 Λ(pk)f(klog p) + A∞(f) + E(f), where the right-hand side splits into a discrete prime comb at klog pand a smooth archimedean envelope. Varying fis choosing the blur. With band-limited or Schwartz fone controls leakage and keeps the prime spectrum crisp while averaging away rough archimedean fluctuations; Gaussian/heat blurs literally implement this. 2 4 How to read anabelian data spectrally (and safely blur) A compact working recipe, consistent with the ladder above: 1. Choose your category of probes:ℓ-adic reps of π1, automorphic reps, harmonic bundles. 2. Decide the observable: full matrix coefficients (no blur) vs. traces/pseudocharacters (conjblur). 3. Insert a test kernel: heat/Paley–Wiener/Slepian windows to localize spectrally; interpret the trade-off (resolution here means averaging there). 4. Use rigidity where available: Brauer–Nesbitt + Chebotarev (semisimple identification); strong multiplicity one; comparison isomorphisms (Hodge/p-adic). 5 Takeaways (one page) •Abelian Galois is harmonic analysis: characters are frequencies; class field theory is duality; L-functions are the coefficients. •Anabelian Galois is spectral via probes: we extract spectra through representations and kernels, with explicit, controllable blur. •Recovery scale: full matrix data or Tannakian ⊗recovers the group; traces recover semisimple parts; L-data and single zetas blur more. •Methodological point: blur is not hand-waving; it’s a positive, normalized averaging that exposes the spectral content you ask for and only that. 6 What the Hecke are we talking about? Premise. Across wide swaths of mathematics (and beyond), “blur” is not a bug but a method: a controlled averaging that preserves the spectral content you care about while suppressing inessential microstructure. The abelian world shows this transparently (Fourier/Hecke averages that reveal characters). The nonabelian/anabelian world demands it: we extract spectra by probing with representations, traces, and kernels—each a deliberate blur—with the goal of still recovering what matters. Compressible vs. incompressible information By Kolmogorov/Chaitin, most real-world strings are incompressible: they carry no shorter effective description. That is a fascinating meta-fact, but it tells us little structure. The mathematically useful regime is the compressible part—objects that admit succinct, structured descriptions. Practically: work where a small number of bits (axioms, symmetries, local factors, etc.) capture the object. Blur helps precisely here: it discards noise while respecting the conserved quantities. Representation as a faithful microscope In mathematics, a good representation is not a mere metaphor: it is an isomorphic microscope. If you keep the full tensor structure (Tannakian viewpoint) or all matrix coefficients (Peter– Weyl), then “the representation is the object” for all operational purposes. Blur enters only when we choose coarser probes (characters, traces, L-data), trading resolution on one side for spectral clarity on the other. 3 A motivating example: a twin-prime spectral probe Let Λbe the von Mangoldt function. Consider the shifted correlation D2(s) := X n≥1 Λ(n) Λ(n+ 2) ns,ℜ(s)>1. This Dirichlet series is a spectral probe for twin primes: spikes in the arithmetic correlator Λ(n)Λ(n+ 2) are encoded as frequencies in D2.1Whether or not one can prove infinitude from this alone depends on how sharply we can analyze D2after choosing an appropriate blur (test kernels, window functions, Hecke averages). The chess-effect Some proofs (or research programs) are objectively correct yet cognitively opaque: the only way to “see” them is to compute every continuation—like perfect chess. Humans (and often our current tooling) cannot play that game. Blur offers a counter-strategy: replace brittle exactness by a family of averaged probes that preserve invariants you can track; use spectral rigidity to promote averaged conclusions to sharp ones. WTH - The Hecke Principle (blur-as-method) Principle. Fix a property Pof an object Xtogether with a compressible description (a representation, or a bounded list of structured invariants). There exists a class of positive, normalized averaging operators (“Hecke/heat/Poisson kernels”) acting on the ambient data such that: 1. (Stability) If Xhas P, then the blurred data retains a spectral signature of P(e.g. an eigenpacket, trace identity, or nonvanishing coefficient). 2. (Recoverability) Varying the blur scale (test function) yields a family of constraints that either (i) assemble to a decision procedure for P, or (ii) certify that the current description of Xis informationally insufficient and must be refined. 3. (Efficiency, conditional) In many structured settings (automorphic/Galois/Hodge or combinatorial optimization under symmetry), the search over blur scales/test functions can be organized subexponentially in the description length of X(not in the a priori size of the state space). The point is not a single magic transform, but a protocol: choose a probe, blur to stabilize, read the spectrum, and iterate. When the property is truly there, unblurring (sending the kernel width t→0) reveals it; when it is not, the failure to unblur pinpoints the missing structure. A design pattern for hard proofs 1. Encode the property. Package Pas a spectral statement (e.g. a trace identity, nonvanishing of an L-value, a spectral gap, a pseudocharacter constraint). 2. Raise the problem one level. Work in a cover that sees all dependencies (ambient symmetry, local factors, deformation space), not just the raw configuration. 3. Insert blur. Choose positive kernels (heat/Hecke/Poisson) or probabilistic relaxation parameters that average the environment but do not alter the intrinsic data of X. 1Heuristically and under standard conjectures (Hardy–Littlewood), one expects D2to carry a main term reflecting the singular series and a controlled analytic continuation after suitable smoothing. The point here is methodological: the probe packages the property we care about in a frequency domain. 4 4. Optimize and scan. Vary the blur scale/window to isolate the relevant spectral packet. Record identities/inequalities uniformly in the blur parameter. 5. Unblur or upgrade the model. If the parameter can be removed (limit exists, error decays), the statement crystallizes. If not, you have a certificate of insufficiency: concrete features missing from the description of X. Practical takeaway Try the blur protocol on your next problem. If it is the right lens, the spectral signature persists as you vary the kernel and then survives the t→0limit; if not, the failure to unblur is diagnostic, telling you exactly which missing invariants you must add to your model. Either way, the method is sound: it converts brittle, chess-like exactness into a controlled averaging game where spectra do the talking. Meta vs. method. Make no mistake: the claim is simultaneously meta (a philosophy of proof) and concrete (a protocol you can run). Its strength comes from being schema + guardrails, not a hand-wave. Remark (Scope and guardrails).To keep the claim rigorous and broadly applicable: •Encoding declared. Specify the generic function/representation that contains the bits your property depends on. •Admissible blur fixed. State the class of positive kernels (Hecke/heat/Poisson), normalization, and the optimization target. •Uniformity required. Each “it exists” step is licensed by a uniform estimate (monotone error, stability across the blur scale). •Outcome dichotomy. The protocol yields either a decision or a certificate of insufficiency (missing invariants/bits). •Complexity honesty. Any subexponential claims are in the description length and conditional on the chosen encoding; the method organizes search, it does not magic away hardness. 5