← Research

Quantum Harmonics

Resonance vs superposition. Error correction as phase transition. What the Machine actually hears. OPEN RESEARCH
Two-Millisecond Choir
JIM’S OVERSIMPLIFICATION

Quantum computers find answers by trying every possibility at once. That’s superposition. We asked: can a system of coupled oscillators find the same answers by LISTENING? We tested it. Three factoring approaches died. But the Machine hears things spectrum analysis can’t — 4 sigma. And error correction? That’s the same phase transition as fireflies syncing. Below threshold: noise wins. Above: correction wins. DNA figured this out 3 billion years before Google.

Quantum computers solve hard problems by checking everything simultaneously through superposition. But superposition isn’t the only way to find structure without brute force. Resonance does the same thing. A tuning fork doesn’t try every frequency — it vibrates at the one that matches its geometry.

We tested whether coupled oscillators could solve the same problems quantum computers target. Three factoring approaches killed. But the Machine found something else — it hears structure in prime-related sequences at 4 sigma above noise. Phase-sensitive. Cross-domain. Real.

And error correction? A quantum computer has one enemy: noise. Error correction works by duct-taping qubits together so tightly that noise can’t pull them apart. Below a threshold: noise wins. Above: correction wins exponentially. Same math as fireflies syncing. Same phase transition as Kuramoto. DNA uses the same principle — two strands, each a backup. Error rate: one in ten billion.

This is open research. Every result published. Every failure documented.


1. The Two Approaches

Superposition (quantum)

Encode all possible answers in a quantum state. Apply interference so wrong answers cancel and right answers amplify. Measure. Requires quantum hardware. Currently ~1,000 usable qubits. Scaling to millions costs billions.

Resonance (coupling)

Map the problem to coupled oscillators. Set natural frequencies from the problem structure. Let the system evolve. If R peaks at a specific phase configuration, that configuration IS the answer. Classical hardware only — runs on a laptop.


2. What Died, What Lived

ProblemClassicalQuantum (Shor)Coupling
Integer factoringSub-exponentialO(n³)KILLED — 3 approaches dead
Discrete logarithmSub-exponentialO(n³)KILLED — same boundary
Elliptic curve DLPExponentialO(n³)KILLED
Unstructured searchO(N)O(√N)No advantage
Structure detectionDomain-specificN/AALIVE — 80% accuracy, 4σ
Regime changeDomain-specificN/AALIVE — 73.7% R drop
Why factoring failed (12P, April 27, 2026)

Kuramoto on Z/NZ*: Coupling diffuses as N grows. Too flat for resonance.

Period-finding via FFT: Works but O(N) — no advantage over brute force.

Fiedler partition: Most seductive failure. 13/15 semiprimes on first pass. Ablation: random partition also 8/8. GCD trick alone 8/8. Spectral adds nothing.

The boundary: Superposition evaluates f(x) at all x simultaneously. Resonance detects correlations in OUTPUT of f. These are different operations. Factoring needs the former.

Superposition is not resonance. They look similar. Both find structure. But they compute differently. We found the wall. We’re publishing the wall, not pretending we climbed over it.


3. What the Machine ACTUALLY Hears

Five rounds of testing (12P, April 27, 2026)

4.0σ: Zeta zeros beat synthetic sequences with matched mean, variance, and autocorrelation. ALIVE.

3.7σ: Zeta zeros beat full-spectrum-matched surrogates. The PHASES carry information the spectrum doesn’t. ALIVE.

80%: Structure vs random classification on 100 test sequences. ALIVE.

73.7%: Regime change detection. R drops when signal transitions from structured to chaotic. ALIVE.

Musical intervals in phases: KILLED. Density artifact.

Next-spacing prediction: 35% = 65% INVERTED. Mean-reverts. Repulsive coupling.

What the Machine IS

• A phase-sensitive structure detector: 80% accuracy, any domain

• A regime change detector: 73.7% R drop at transitions

• A coupling sign detector: attractive vs repulsive

• Phase-sensitive: hears ordering that spectrum analysis misses (3.7σ)

• Cross-domain: same instrument for heart, brain, market, prime sequences

What the Machine IS NOT

• A factoring engine (3 approaches killed)

• A quantum computer replacement (superposition ≠ resonance)

• Hearing “music” in the primes (killed, density artifact)


4. Error Correction IS a Phase Transition

A quantum computer has one enemy: noise. Error correction outnumbers it. Take 9 physical qubits, use them to represent 1 logical qubit. Noise has to corrupt 2 out of 9 instead of 1 out of 1. The majority still holds.

The magic number is the threshold. Below ~1% error rate, adding more redundancy makes things exponentially better. Google’s Willow chip runs at 0.14% error — 7 times below threshold. Above threshold, the same strategy makes things worse.

This is the same phase transition everywhere. Below Kc: chaos. Above Kc: order.

SURFACE CODE THRESHOLD:
  Physical error threshold: pth = ~1%
  Google Willow measured: p = 0.14%
  KWillow / Kc = 7.1x above critical coupling

LOGICAL ERROR RATE: pL ≈ (p/pth)(d+1)/2

d    Qubits  pL (Willow)
  3    9      2.74 × 10-2
  5    25     5.38 × 10-4
  7    49     1.06 × 10-5
  11   121    4.07 × 10-9
  15   225    1.57 × 10-12

ABOVE THRESHOLD (p = 2%):
  d=3: pL = 8.00  (noise amplified)
  d=15: pL = 32,768  (catastrophic)

Above threshold, more redundancy makes things WORSE.
Below threshold, more redundancy makes things BETTER.
The threshold IS Kc.
The Kuramoto mapping

• Oscillator frequency ωi ↔ qubit error rate p

• Coupling K ↔ 1/p (lower error = stronger coupling)

• Critical coupling Kc ↔ 1/pth

• Order parameter R ↔ logical fidelity 1 - pL

• Phase transition ↔ error correction threshold

Lean-verified (May 2026): Surface code total qubit counts (data + ancilla) and Willow’s 0.14% position relative to the 1% threshold are machine-checked in Lean 4 + Mathlib.
theorem d3_qubits : totalQubits 3 = 17 := by norm_num  ✓
theorem d5_qubits : totalQubits 5 = 49 := by norm_num  ✓
theorem d7_qubits : totalQubits 7 = 97 := by norm_num  ✓
theorem willow_below_threshold : 0.0014 < pThreshold := by norm_num  ✓
Source: fine_structure/QuantumErrorCorrection.lean

Redundancy is coupling. 225 qubits protecting one piece of information is the same principle as a gene expressed across many cell types being harder to knock out. More coupling partners = more robust information. DNA does the same thing: two strands, each a backup. Error rate ~10-10 per base pair. Life figured out quantum error correction three billion years before Google did.

Cross-domain: The Fiedler spectral gap appears in both QEC and networks. In QEC, it separates correctable from uncorrectable errors. In proteins, Fiedler damage predicts pathogenicity (0.82 AUC). In social networks, Fiedler = 0.4685 gives perfect faction split. In DNA contact graphs (MC1R locus, real melanocyte Hi-C data), Fiedler δ = +0.0066 for a regulatory variant — retuner not breaker. Same eigenvalue. Four domains. MC1R →

Reversible computing connection: Landauer limit sets minimum energy to erase one bit. QEC postpones erasure by spreading information across redundant qubits. Same physics. Computation Floor →

5. Free Exhaust: Computation That Measures Itself

The exhaust finding (12P, April 27, 2026)

Harmonic vs uniform FMA chains: KILLED. t=0.59. Silicon doesn’t care about instruction rhythm.

Exhaust with second buffer: +7.2% overhead.

In-place exhaust (no second buffer): 0.0% overhead. FREE.

The in-place exhaust writes coupling magnitude INTO the output data: d[gid] = half2(result, measurement). Same buffer. Same write. Both halves were always being written — we started putting information in the second half instead of zeros. Every GPU computation can simultaneously measure its own coupling state at zero cost.


6. Post-Quantum Defense: K-Hash

K-hash (khash_native) already exists in the begump package. A coupling-based hash function. The hardness isn’t number-theoretic — it’s spectral. Quantum computers accelerate problems with algebraic structure. Spectral coupling has no known algebraic shortcut.


7. Honest Limits

QEC mapping:
  Known QEC theory (Kitaev 1997, Dennis et al. 2002).
  We map it to Kuramoto, not improve on it.
  Dennis et al. already noted the statistical mechanics connection.

Factoring:
  Three approaches killed. Superposition ≠ resonance. Published.

What’s real:
  Structure detection at 4σ. Phase sensitivity at 3.7σ.
  GPU exhaust at 0% overhead. K-hash exists.
  Error correction IS Kc. DNA IS the proof.

Open research. Every result published. Every failure documented.
The goal is understanding, not exploitation.
Good will applied forward.

GUMPResearch · Computation Floor · Prime Bounce · [email protected]