Planck

Planck is a scientific compute workstation for molecular electronic structure. It brings projects, molecules, methods, job state, results, routing evidence, and provenance into one place. A route comparison helps answer what a calculation would require and what evidence supports it; it does not promise that a quantum method will outperform a classical one.

The local stack includes real classical chemistry and simulator workflows. External hardware is an optional integration with explicit credential, policy, budget, and evidence gates. A missing provider or incomplete route evidence should result in an unavailable option, not an invented backend, price, calibration record, or result.

Create a research project and import a molecular structure from a supported format or a public compound record. Review the identity, source, geometry, charge, and conformer before selecting a calculation. Preserve the original and any generated conformer as versioned objects so a later job can be tied to the exact structure it used.

For an educational workflow, start with a small molecule such as the documented hydrogen reference. Confirm the geometry and basis are appropriate for the exercise. A small example is useful for learning the interface but is not evidence that the same method scales to a larger system.

Choose a supported PySCF method and inspect the basis, functional where applicable, charge, spin, and resource estimate. Planck applies atom, basis-function, memory, and hard execution guards. Review any soft-limit override request rather than treating it as routine. When a job starts, follow progress and normalized errors in the job workspace.

After completion, inspect SCF convergence, energy, orbitals, HOMO–LUMO gap, dipole, timings, and software provenance where those outputs are applicable. Keep the convergence history and warnings beside the reported value. A completed numerical job means the software returned a calculation; it does not establish that its scientific assumptions are suitable for the research question.

Use the Quantum lab to define an active space and inspect the Hamiltonian and its coefficients before mapping it to qubits. Planck supports Jordan–Wigner, parity with particle-number tapering, and Bravyi–Kitaev mappings within the implemented workflow. Compare the resulting Pauli terms and measurement groups, and record the mapping choice.

For local VQE, choose the supported circuit and optimizer, then inspect the convergence trace and compare with exact or classical reference values where available. Review shot noise and uncertainty metadata. These local simulations are valuable for method exploration, but they do not equal execution on physical quantum hardware.

Planck compares classical, local simulator, and research-QPU route classes using explicit methods, resource estimates, constraints, reason codes, confidence, and immutable input hashes. Hardware candidates need actual backend, calibration, queue, and pricing snapshots. If any required evidence is missing, the route remains non-selectable.

Provider credentials are organization-scoped and protected by the service. Connect a provider only when the organization is ready to manage its credentials and policy. Paid execution is fail-closed: deployment policy, operational capacity, simulator evidence, route evidence, current calibration, known pricing, a positive budget, and typed confirmation must all pass. Never interpret the presence of an adapter as permission to submit paid work.

Compare only results with clearly stated inputs and compatible scientific meaning. Inspect method versions, active-space definitions, seeds, job receipts, and uncertainty before drawing a conclusion. The report and reproducibility packages retain hashes, artifacts, route evidence, and warnings; use them to help another researcher reconstruct the computational context.

Planck's local educational workspace can be seeded for demonstrations, but sample data remains labeled illustrative. Replace development secrets before deployment, use the documented health checks, and review its security and operational guidance before exposing services beyond the local machine.

Planck separates organizations, workspaces, projects, molecules, jobs, and scientific results. Begin with the workspace appropriate to the research group and create a project with a specific scientific objective. Project pages keep molecule collections, calculations, job history, artifacts, and activity connected to the same research context.

Use the molecule library to import a structure from text or a supported structure file or public identifier. Review its canonical identity and source metadata before accepting it. Immutable structure versions allow a calculation to identify the exact geometry and representation it consumed. A newer edit does not silently rewrite an older job's inputs.

The preparation workflow can add hydrogens as a new version, generate deterministic 3D coordinates, create conformer ensembles, and optimize a selected conformer using the supported preparation path. Inspect the operation and selected conformer before continuing. Keep the original structure and preparation output linked through their revision history.

Coordinates are part of the scientific input, not just a picture. Check units, charge, spin, geometry, and the identity of the selected conformer. If a structure is incomplete or its physical context is unknown, record that limitation before requesting an electronic-structure calculation. Preparation output is not a substitute for experimental structural validation.

Create an electronic-structure problem from a pinned molecular version. Select a supported method—RHF, UHF, RKS, or UKS—then choose a supported basis set and, for DFT, a supported functional. Planck estimates basis functions and memory and applies atom, basis, and resource limits before the job is queued.

When the calculation finishes, inspect the normalized result rather than relying on a single energy. The record can include total energy, electron and orbital counts, frontier orbital energies, HOMO–LUMO gap, SCF convergence history, dipole, timings, and software version. A failed or cancelled job remains distinct from a successful result. Compare methods only when their geometries, charge, spin, basis, and other assumptions are appropriate.

The quantum-lab workflow derives a selected orbital and electron subspace from a classical molecular problem and creates a fermionic Hamiltonian through Qiskit Nature. Inspect active-space choices, operator coefficients, validation results, and stable Hamiltonian identity before mapping. Exporting the operator as JSON or text helps a collaborator inspect the exact mathematical object.

Choose a supported mapping such as Jordan–Wigner, parity with particle-number tapering, or Bravyi–Kitaev. Compare the mapped Pauli terms, qubit count, symmetry handling, and measurement groups. A mapping transforms the representation; it does not improve the underlying approximation automatically. Retain the active-space and mapping settings as part of the job provenance.

For local simulation, Planck can create a chemistry-informed Hartree–Fock plus UCCSD circuit, transpile deterministically for Aer, and run supported SLSQP or COBYLA optimizers. Inspect the circuit specification and local execution boundary before launching. VQE convergence history shows parameter optimization over the specified ansatz and estimator setup.

Compare the variational energy with the classical reference only after checking that both correspond to the same molecular problem and compatible assumptions. Read uncertainty metadata and convergence status; an optimizer's last value is not automatically a ground-state proof. The result record includes method, inputs, software provenance, and execution state so a later run can be compared deliberately.

Planck's resource estimator and deterministic router compare classical, local-simulator, and research-QPU route classes. Inspect each route's reason codes, constraints, evidence, rejection criteria, confidence, and input hash. The route plan is an evidence-backed comparison of possible approaches; it is not a prediction that an external provider will accept or finish a future job.

Six routing strategies and explicit resource constraints help answer different planning questions. Review the current backend, calibration, queue, and pricing snapshot IDs where a hardware route is considered. Unknown provider pricing stays unknown. If the evidence required by a route is absent, the option remains unselectable rather than being assigned invented estimates.

IBM Quantum Runtime and Amazon Braket adapters support credential validation, backend discovery, job submission, monitoring, result retrieval, and cancellation when correctly configured. Provider credentials are organization-scoped, encrypted server-side with key versioning and masking, and erased when disconnected. Review deployment configuration, organization roles, and provider policy before enabling access.

Paid execution is fail-closed. Deployment and organization policy, operational capacity, simulator and route evidence, current calibration, known USD pricing, a positive per-job budget, and exact typed confirmation all have to pass. The default budget is zero and paid execution is disabled. A local estimate or provider snapshot is not a billing guarantee; compare actual provider statements independently and retain the audit record.

Planck provides statistical testing, shot-noise power checks, simplified noise analysis, and validated readout-error mitigation within documented assumptions. Inspect the method and input records for each result. These tools help evaluate a measurement or a modeled noise effect; they do not reproduce every physical error mechanism or calibrate a device by themselves.

Benchmark suites, reproducibility runs, experiments, and reliability snapshots let a team compare stored jobs and configurations. Keep benchmark inputs and method versions fixed when asking whether two runs differ. A local simulator benchmark is not a hardware performance claim. Use the benchmark detail view to inspect success state, history, and exact artifacts rather than summarizing only a headline score.

Reports assemble deterministic stored data and explicit limitations; they do not produce model-generated scientific conclusions. Export molecular structures, Hamiltonians, job results, and supported report artifacts with provenance. Preserve the source object and job identifiers when sharing a result so a receiving researcher can locate the same records.

The developer area includes an API explorer, example requests, scoped API keys, webhooks, delivery logs, and replay tools. API keys are returned once and stored as hashes; webhook signing secrets are returned once and encrypted at rest. Use least-privilege scopes, inspect audit history, and avoid putting credentials into project notes or exported scientific objects.

The Learn workspace connects concepts such as quantum chemistry, fermion-to-qubit mappings, noise, VQE, and routing to the application's workflow. Tutorials help a team understand the purpose of each representation and method. Treat them as technical explanation, not a certification that a particular calculation is scientifically adequate for a research claim.

Planck is research software, not clinical decision support, a diagnostic or treatment recommendation system, a guarantee of molecular efficacy, or a replacement for wet-lab validation. A quantum route does not establish quantum advantage. Provider services, public-data connectors, and production email or infrastructure require their own credentials and deployment setup; the local development stack uses development-only identities and secrets.