PT Symmetry & Exceptional Points

The exceptional point is not a pathology to avoid — it is the computational resource.

Table of contents

  1. The exceptional point as resource
  2. What you already know, reframed
    1. The three regimes
    2. The β-plane picture
  3. PiTch number: the new invariant
    1. Definition
    2. The pattern for low-order EPs
    3. What winding number misses
  4. Numerical evidence
  5. PiTch number as a computational resource
    1. The T-count analogy
    2. The five HPU substrate classes
  6. Open questions
  7. Papers
  8. What to read first

The exceptional point as resource

The exceptional point is not a pathology to avoid. It is the computational resource. The PT phase transition — that sharp boundary between real and complex-conjugate eigenvalue pairs first identified by Bender and Boettcher (1998) — is a tier boundary in a general classification of physical computation. The new invariant that captures this structure is the PiTch number: the total algebraic weight of all exceptional points enclosed or crossed by a path in parameter space. The PiTch number is to non-Hermitian dynamics what T-count is to fault-tolerant quantum computing.

What you already know, reframed

If you work on PT-symmetric systems, you already have the full picture in your hands. The H^k tier ladder gives it a name.

The three regimes

Physical regime What you know ISA tier
PT-unbroken (real eigenvalues) Berry phase accumulation, coherent interference, eigenvalue repulsion without crossing H¹
PT phase transition Eigenvalues collide; eigenvectors coalesce; Jordan block forms H¹ ↔ H² boundary (β*₁₂)
PT-broken (complex-conjugate pairs) Gain-loss dynamics; exponential amplification/decay; dissipation dominant H²

The H¹ tier is the home of coherent, unitary-like dynamics — Berry phase, adiabatic transport, interference. The gain-loss parameter is present but small; the system lives in the PT-unbroken phase; eigenvalues are real and well-separated.

The H² tier is the home of dissipation-dominated dynamics — complex eigenvalues, gain/loss rates that matter, the broken phase. The system has crossed the PT phase transition.

The β*₁₂ snap threshold is the exceptional point locus. It is the precise location where the system changes computational character — not just a spectral curiosity but a tier boundary, the same mathematical object as the snap threshold in every other system the HotLogiQ framework describes (the Mott transition, the induction head phase transition in transformers, the kinetic proofreading threshold in DNA replication).

The β-plane picture

Standard quantum mechanics lives on the imaginary β axis: β = iω, where ω is the real frequency. PT-symmetric systems live slightly off it, in the interior of the complex β-plane. The EP locus — the surface in parameter space where eigenvalues coalesce — is the tier boundary β*₁₂ in complex β-space.

Different addresses in the complex β-plane correspond to different physical regimes:

β location Physical regime
β → ∞ (real) Classical; tropical arithmetic; winner-takes-all
β = it (imaginary) Standard quantum mechanics; unitary evolution
β slightly off imaginary axis PT-symmetric; eigenvalues remain real while gain-loss is small
β on the EP locus (β*₁₂) Exceptional point; tier boundary
β on negative real axis Laser dynamics; exponential gain

The PT phase transition is not a special feature of non-Hermitian physics. It is the β-plane story told in a particular experimental context: photonic waveguides with balanced gain and loss, microwave resonators, mechanical oscillators with position-dependent damping. The tier structure is universal; the substrate is what varies.

See also: The β-plane

PiTch number: the new invariant

Definition

Let γ be a closed path in the parameter space of a non-Hermitian Hamiltonian H(λ). Define:

S(γ) = Σ (order of EP − 1) over all EPs enclosed or crossed by γ

For an n-th order EP (where n eigenvalues coalesce and n eigenvectors merge into a single Jordan block), the contribution to S is n − 1.

The pattern for low-order EPs

The three things that physicists already know — monodromy group, Berry phase, Kato exponent — are not three separate phenomena. They are three faces of a single invariant:

n (EP order) S Monodromy Phase per loop Kato exponent Permutation
2 1 ℤ₂ π 1/2 (01) — transposition
3 2 ℤ₃ 2π/3 1/3 (012) — 3-cycle
4 3 ℤ₄ π/2 1/4 (0123) — 4-cycle
n n−1 ℤₙ 2π/n 1/n n-cycle

Reading the table: one loop around an n-th order EP permutes the eigenvalue sheets by an n-cycle in ℤₙ, accumulates a Berry phase of 2π/n, and gives Kato eigenvalue splitting proportional to ε^{1/n} near the EP. These three facts were known separately. PiTch number unifies them: the single integer S = n−1 determines all three.

Two loops around a 4th order EP give permutation (0 2)(1 3) — the order-2 element of ℤ₄, not the identity. The full group structure is directly observable by tracking which eigenvalue sheet returns to which starting value after each loop. This is an experiment you can do with a two-port microwave resonator or a coupled-waveguide setup.

What winding number misses

The winding number of a path around a region of parameter space is a topological invariant, but it is too coarse for the full eigenvalue braiding structure. Consider two distinct situations:

  • Situation A: one 3rd order EP enclosed by γ. Monodromy = ℤ₃. PiTch number S = 2.
  • Situation B: three separate 2nd order EPs enclosed by γ. Monodromy = S₃ (symmetric group on 3 elements). PiTch number S = 3.

Both situations have the same winding number = 1. But S = 2 ≠ 3, and ℤ₃ ≠ S₃ as groups — ℤ₃ is abelian (cyclic), S₃ is non-abelian. The observable consequence: the eigenvalue braiding is a cyclic permutation in situation A and potentially a non-cyclic permutation in situation B. Any measurement that encodes information in the eigenvalue ordering can distinguish them; winding number cannot.

PiTch number is not a complete invariant in general (two paths can have the same S but different permutation groups), but it is a strictly finer invariant than winding number, and for cyclic EPs it is complete.

Numerical evidence

Experiments x678a, x678b, and x678c verify the PiTch number table for 2nd, 3rd, and 4th order EPs in a family of non-Hermitian random matrices:

n S Monodromy Phase/loop Kato exp Experiment
2 1 ℤ₂ π 1/2 x678a PASS 6/6
3 2 ℤ₃ 2π/3 1/3 x678b PASS 6/6
4 3 ℤ₄ π/2 1/4 x678c PASS 6/6

Each experiment: (i) constructs an n-th order EP via fine-tuned complex deformation of a Hermitian starting matrix; (ii) tracks eigenvalue sheets numerically around a closed loop in parameter space; (iii) reads off monodromy permutation, accumulated phase, and power-law splitting exponent and compares to the PiTch number prediction. All 18 tests pass.

Experiment x678d addresses Bender’s original system H = p² + x²(ix)^ε via an effective 2×2 PT model at each level pair. Result: S = 0 throughout ε ≥ 0 (all pairs in the H¹ tier); the PiTch threshold appears at ε = 0. This gives the 1998 real-spectrum result a topological certificate. [PASS 6/6]

PiTch number as a computational resource

The T-count analogy

In fault-tolerant quantum computing, the T-count of a circuit measures its non-Clifford gate cost. Clifford gates are cheap (stabiliser states, easy to simulate classically); T gates are expensive (they promote Clifford circuits into the genuinely quantum regime). The T-count is the resource metric.

PiTch number plays the same role for non-Hermitian computation:

  • S = 0: the path in parameter space never crosses or encloses an EP. The system stays in one tier throughout its evolution. It can be efficiently simulated within that tier — no tier promotion occurs.

  • S > 0: the path crosses or encloses at least one EP. Tier transitions occur. These transitions are the resource for anomalous EP sensing (Kato splitting ε^{1/n}), eigenvalue braiding, and dissipation-enhanced computation.

The analogy is imperfect in one important direction: T-count measures a cost (you want it low for efficient classical simulation); PiTch number measures a resource (you want it non-zero to exploit EP-specific physics). But the structural parallel holds — both are invariants that divide circuits into computationally distinct classes.

The five HPU substrate classes

The HPU (Hot Processing Unit) is the hardware class that deliberately programmes with gain, loss, and complex β as first-class resources. Five substrate families have been identified, each with a natural PiTch number operating regime:

Class Substrate S (PiTch) Key feature
HPU-P PT-symmetric photonic S ≥ 1 EP crossing = SNAP↑ event
HPU-B Bosonic cat qubit S = 0 (controlled) ERASE opcode; autonomous error correction
HPU-R Radical-pair molecular S = 0 or 1 Spin selection; ERASE = triplet/singlet swap
HPU-C Non-Hermitian circuit S = 1 SNAP↓ = non-Hermitian skin effect
HPU-E EP sensor array S = 0 (at the EP) S = 0 sensing; maximally sensitive

The HPU-E entry deserves attention. “PiTch number = 0 sensing” sounds paradoxical — if S = 0, hasn’t nothing happened? The point is subtler: the EP sensor operates at the snap threshold β*₁₂, poised exactly at the EP without crossing it. The anomalous Kato splitting ε^{1/n} is maximal at the EP; the S = 0 condition means the operating point is the EP itself, not a path that encloses it. This is the computational analogue of sitting at a quantum critical point and using the diverging susceptibility as a transducer.

Result: EP sensing — the observation that sensitivity to a perturbation δ near an n-th order EP scales as δ^{1/n} rather than δ (the Hermitian result) — is a PiTch number = 0 computation. It is the first demonstrated HPU algorithm.

Open questions

The PiTch number framework opens several concrete research questions:

Completeness. Is PiTch number a complete invariant? Two paths with the same S but different enclosed EP configurations can have different permutation groups (ℤ₃ vs S₃ for S = 2, as above). A complete invariant would need to encode not just the total count but the full braiding group of the enclosed EP configuration. Is this the right object — the eigenvalue braid group — and is it finitely computable from the spectrum?

Transmission-matrix access. In photonic and microwave experiments, the directly accessible observable is the transmission matrix S(ω), not the individual eigenvalue sheets. Can S be extracted from the transmission matrix without full eigenvalue tracking? A positive answer would make PiTch number experimentally accessible in any two-port scattering setup.

Floquet systems. Time-periodic PT-symmetric Hamiltonians H(t) = H(t+T) have quasi-energies defined modulo ℏΩ. The Floquet Brillouin zone is a torus, not a plane. Is there a PiTch number for the quasi-energy spectrum — a count of EPs per Floquet zone — and does it have the same tier-boundary interpretation?

Bender’s original system. H = p² + x²(ix)^ε is PT-symmetric for all ε and passes through an EP structure as ε varies. The effective-model experiment (x678d) confirms S = 0 throughout ε ≥ 0. Open question: the full PiTch number structure of the multi-sheet EP locus at irrational ε.

Papers

Foundation — parameter space and tier structure

The invariant

PT symmetry in algorithms and computation

Hardware and substrate

Connections to other domains

  • G₂ Snap Thresholds (G2SnapThresh) — why 3-, 5-, and 7-qubit codes are canonical; G₂ geometry of the snap threshold

  • The Maslov Moment (MaslovMoment) — financial phase transitions; spinodal = EP₂; 2008 crisis as min-plus snap

What to read first

If you work on EP sensing: Start with HPU Architecture. Section 3 introduces PiTch number = 0 sensing directly, with the ε^{1/n} sensitivity table and the HPU-E substrate. The ISA formalism is introduced only as needed.

If you want the mathematics: Start with PiTch: A Topological Invariant for PT-Symmetric Systems. It derives the PiTch number, proves the Berry-phase formula Φ = πS, and verifies the ℤₙ monodromy for n = 2, 3, 4. Then PT Symmetry in Unexpected Places for the full H¹/H² tier identification and ISA context.

If you want the big picture: Start with The β-plane. It places PT physics in the context of the full HotLogiQ framework — showing where PT systems live relative to classical (β real), quantum (β imaginary), and other exotic regimes — and explains why the EP locus is a tier boundary rather than a spectral accident.

See also: Processing Units — the HPU is the hardware class built around PiTch number as a resource · The β-plane — the full complex β parameter space · The Non-Associative Frontier — the complementary H² structure (G₂ geometry) that sits alongside PT symmetry in the H² tier