IonQ and Quantinuum built an Orbit Processing Unit without knowing it

Plain-language explainer for doi:10.5281/zenodo.21360907 (#604)

The one-sentence version

The Mølmer-Sørensen (MS) gate — the native 2-qubit gate of every trapped-ion quantum computer — is exactly the BIND opcode of the Origami ISA, and BIND depth (not qubit count) is the correct metric for quantum advantage.

The Mølmer-Sørensen gate is a BIND

In 1999, Mølmer and Sørensen showed that two trapped ions, illuminated simultaneously by a bichromatic laser field, undergo a joint evolution:

\[U_{\rm MS}(\theta) = \exp\!\left(-\frac{i\theta}{2}\, \sigma_x \otimes \sigma_x\right)\]

The shared phonon modes of the ion crystal mediate the coupling — they act as a carrier that is traced out after the gate, leaving entanglement in the qubit degrees of freedom. At θ = π/4 this produces a maximally entangled Bell state from a product input.

In ISA language, this is exactly BIND: the H² holonomy that couples two ORBIT paths through a shared intermediate state (the phonon bus). The phonon is not stored — it is the carrier, just as the associator in an octonion product is the algebraic carrier of the G₂ holonomy.

The identification is not an analogy. The MS gate is generated by σ_x ⊗ σ_x, which is the two-body generator that crosses Schubert cell boundaries in Gr(k, 2N) — the Grassmannian describing the state space of N two-level ions. Single-qubit rotations (ORBIT) move within a Schubert cell; the MS gate (BIND) is the opcode that crosses between cells and generates entanglement that cannot be factored.

Every trapped-ion circuit is an ISA programme

Any N-qubit algorithm on a trapped-ion processor decompiles to:

INIT:     ORBIT[|00...0⟩]        -- Doppler + sideband cooling
SINGLE:   ORBIT[R(θ,φ) on ion i] -- Bloch sphere rotation; ~1–10 μs, >99.99%
PHASE:    TWIST[R_z(φ) on ion i] -- virtual frame update; zero physical time
ENTANGLE: BIND[U_MS(π/4), i,j]   -- phonon-bus holonomy; ~100–500 μs, ~99.9%
MEASURE:  LABEL[σ_z on ion i]    -- state-selective fluorescence

The key observation: ORBIT and TWIST gates are essentially free in both time and error (>99.99% fidelity, implemented in microseconds). The BIND gate (MS gate) is ~100× slower and contributes ~10× more error per operation. Circuit quality is determined by BIND count, not total gate count.

This is the trapped-ion analogue of T-count in fault-tolerant quantum computing: the BIND count measures the H² resource consumed by the computation.

BIND count for common gates: the KAK theorem

The KAK decomposition (Cartan decomposition of SU(4)) proves that any 2-qubit unitary can be implemented with at most 3 MS gates plus single-qubit rotations. The minimum number of MS gates needed depends on where the gate sits in the Weyl chamber:

Gate BIND count Weyl chamber position
Identity (no-op) 0 Corner (trivial)
CNOT 1 Face
CZ 1 Face (same Weyl class as CNOT)
iSWAP 2 Edge
SWAP 3 Far corner
Any SU(4) ≤ 3 Interior

The Weyl chamber coordinate c₂ — which Paper 596 uses to predict DFT failure in chemistry — is the same coordinate that determines BIND count here. The same geometry describes both the correlation structure of molecules and the entanglement structure of quantum gates. This is not a coincidence: both are Grassmannian problems, and the Weyl chamber is the quotient of the Grassmannian by the symmetric group.

All-to-all connectivity = zero wasted BINDs

In a superconducting processor (IBM, Google), qubits couple only to their nearest neighbours on a 2D grid. Moving entanglement across the chip requires SWAP gates — and each SWAP costs 3 BINDs. A long-range entangling operation between distant qubits might require O(N) SWAPs = O(N) wasted BINDs.

In a trapped-ion crystal, every ion couples to every other through the shared motional modes. Any two ions can perform a direct MS gate with no routing overhead. The Schubert cell decomposition of Gr(k, 2N) is fully connected: every cell is reachable from every other in O(N) MS gates, with no SWAP cost.

Quantitative consequence for GHZ state preparation (n qubits):

Platform BIND count BIND depth Circuit fidelity at n=10
IonQ / Quantinuum (all-to-all) n−1 1 F ≈ (0.999)⁹ ≈ 0.991
IBM (nearest-neighbour) n−1 n−1 F ≈ (0.995)⁹ ≈ 0.956

Both platforms execute exactly n−1 BINDs. The fidelity gap comes entirely from BIND depth — the number of sequential MS-gate layers. On all-to-all hardware, all n−1 BINDs run in parallel (depth 1); on a chain, they must run sequentially (depth n−1). Fidelity decays with depth, not count.

The 99.9% fidelity threshold = β* snap

Quantinuum H2-1 achieved 2-qubit MS gate fidelity ≥ 99.9% (Evered et al., Nature 2023). The surface code fault-tolerance threshold is approximately 99.4% for circuit-level noise.

ISA interpretation: crossing from below to above the fault-tolerance threshold is a β* snap — the same phase transition that separates single-reference from multi-reference quantum chemistry in Paper 596, and H⁰ from H² behaviour throughout the ISA.

  • Below threshold (F < 99.4%): each additional BIND layer adds more physical error than QEC removes. The logical error rate grows with code size.
  • At threshold (F = β*): the error correction gain exactly cancels the physical error. A fixed point.
  • Above threshold (F > 99.4%): logical error decreases exponentially with code size. QEC works.

Quantinuum H2-1 has crossed β*. IBM Heron (best ~99.5% for some qubit pairs) is close but not consistently above across all pairs and gate types. This makes trapped ions the first physical H² platform in the fault-tolerant Schubert regime.

BIND depth is the correct quantum advantage metric

Current quantum advantage claims are stated in terms of qubit count or total circuit depth. The ISA says this is the wrong measure.

Why BIND depth is what matters: Classical simulation of quantum circuits via tensor networks requires a bond dimension χ that grows as 2^{BIND_depth}. Single-qubit gates (ORBIT, TWIST) do not increase entanglement entropy and do not grow χ. Only BIND gates generate entanglement and make classical simulation hard.

A circuit with 1000 ORBIT gates and 5 BIND gates is easy to simulate classically (χ = 2⁵ = 32). A circuit with 5 ORBIT gates and 50 BIND gates may be classically intractable (χ = 2⁵⁰). Total gate count is irrelevant; BIND depth is the resource.

The reframing for IonQ and Quantinuum: their hardware advantage is not “N qubits” — it is “99.9% BIND fidelity with all-to-all connectivity.” This means:

  1. Every BIND in the programme can be executed directly (no SWAP overhead)
  2. BIND depth equals the circuit’s critical-path BIND count (no sequential routing penalty)
  3. At 99.9% per BIND, a 50-BIND-depth circuit achieves F ≈ 0.999⁵⁰ ≈ 0.951 — still useful

The ISA makes this comparison hardware-independent and quantitative.

The dictionary

Trapped-ion concept ISA / Grassmannian concept
MS gate U_MS(π/4) BIND opcode on Gr(k, 2N)
Single-qubit rotation R(θ,φ) ORBIT (geodesic on S²)
Virtual Z-rotation R_z(φ) TWIST (Berry phase, zero time)
Phonon bus H² holonomy carrier (traced out)
KAK decomposition Minimal BIND count for any SU(4) gate
All-to-all connectivity Complete Schubert cell accessibility
BIND count T-count analogue for trapped ions
BIND depth Quantum advantage metric (classical χ ∝ 2^{depth})
99.9% fidelity threshold β* snap: phase transition to fault-tolerant regime
Quantinuum H2-1 First physical OPU operating above β*

See also:

For the full technical treatment, see doi:10.5281/zenodo.21360907