# Techniques: Quantum Hamiltonian Certification

## What the paper contributes (one paragraph)
The paper formalises the Hamiltonian certification problem: given black-box access to the time evolution $e^{-iHt}$ of an unknown Hamiltonian $H$, decide whether $H$ is close to or far from a known target $H_0$, without first learning $H$. It gives a direct certification framework that reaches the optimal total evolution time $\Theta((\varepsilon_2-\varepsilon_1)^{-1})$ under the normalised Frobenius norm with no structural assumptions, and extends optimally to all Pauli norms and normalised Schatten $p$-norms for $1\le p\le 2$ in the one-sided-error setting, implying a quadratic advantage over Hamiltonian learning in Pauli 1-norm. Matching lower bounds confirm optimality, and certification under the operator norm is shown to be coQMA-hard. An ancilla-free variant based on dual stabiliser sampling keeps inverse-precision scaling using only product-state inputs, making the protocol suitable for near-term devices.

## Techniques offered
- **Hamiltonian certification (property test from real-time evolution)** - decides whether an unknown Hamiltonian is within $\varepsilon_1$ of, or at least $\varepsilon_2$ from, a target Hamiltonian using only queries to its time evolution, with no structural assumptions.
  - guarantee: statistical property test with optimal total evolution time $\Theta((\varepsilon_2-\varepsilon_1)^{-1})$ under the normalised Frobenius norm, plus matching lower bounds across Pauli and normalised Schatten $p$-norms ($1\le p\le 2$).
  - quote: "This work introduces a direct and efficient framework for Hamiltonian certification, which distinguishes whether an unknown Hamiltonian matches a target specification within given precision bounds."
- **Ancilla-free (dual-stabiliser) Hamiltonian certification** - a product-state protocol that certifies a Hamiltonian without auxiliary qubits or controlled evolution, using dual stabiliser-state syndrome sampling in place of Bell states and amplitude estimation.
  - guarantee: solves certification to precision $(\varepsilon,4\varepsilon)$ under the normalised Frobenius norm in total evolution time $\widetilde O(m^{3/2}\varepsilon^{-1})$, retaining inverse-precision scaling on near-term hardware.
  - quote: "we develop an ancilla-free certification method that maintains the inverse precision scaling while eliminating the need for auxiliary qubits, making our approach immediately accessible for near-term quantum devices with limited resources."
- **Hamiltonian amplitude encoding** - encodes the Pauli-coefficient information of the (Trotterised) residual time-evolution operator into a measurable state amplitude that can be read out by square-root amplitude estimation.
  - guarantee: statistical estimate; the encoded amplitude $p_1(t)$ is provably upper- and lower-bounded by the residual Hamiltonian's Pauli coefficients for small $t$, estimable to additive error in $O(1/\varepsilon)$ queries.
  - quote: "we introduce the \\emph{Hamiltonian amplitude encoding} method, which encodes the Pauli information of the evolution operator as the measurable amplitude of a state."

## Where it could apply
- Target primitive(s): Trotterised simulation, QPE / quantum metrology and sensing pipelines, any analogue or digital time-evolution primitive whose generating Hamiltonian must be verified.
- Target application group(s): platforms_tooling, error_correction (device verification and benchmarking), chemistry_materials (verifying simulated many-body and chemistry Hamiltonians).
- Code family / hardware assumptions: hardware-agnostic; the ancilla-free variant targets near-term devices with limited qubits and noisy multi-qubit gates (no surface-code or specific substrate assumption).

## Caveats
The certification guarantees hold for traceless Hamiltonians with constant-bounded Pauli coefficients and assume coherent access to the time-evolution operator (the ancilla-based optimal protocol needs $n+2$ ancillas and controlled evolution; the ancilla-free variant trades this for a wider $(\varepsilon,4\varepsilon)$ gap and $m$-dependent evolution time). Certification under the operator (normalised Schatten $\infty$) norm is coQMA-hard and so is not efficiently suppliable. This is a theory paper with no released implementation.