Fault-Tolerant Algorithm Supply Chain — Operating Quantum Computers

Fault-tolerant quantum computing will need a software supply chain for algorithms. The unit of work is not a notebook or a circuit; it is an algorithm package that carries problem assumptions, precision targets, resource estimates, logical-operation requirements, magic-state demand, compiler assumptions, and verification tests.

Qualtran and Azure Quantum's resource estimator illustrate the emerging need to express algorithms in forms that support analysis before hardware exists at scale. [R100] [R104]

57.1 Supply-chain stages

Diagram loads as you read

57.1 Supply-chain stages · Figure 1

View diagram source

flowchart LR
    Math[Algorithm theory] --> Spec[Executable specification]
    Spec --> Resource[Resource estimate]
    Resource --> Logical[Logical circuit package]
    Logical --> Compile[FT compiler]
    Compile --> Factory[Magic-state and code factory plan]
    Factory --> Runtime[Logical runtime contract]
    Runtime --> Evidence[Evidence package]

Each stage must preserve assumptions. Losing assumptions is how optimistic papers become impossible deployments.

57.2 Algorithm package manifest

Illustrative listing · yaml

ft_algorithm_package:
  package_id: hamiltonian-simulation-qsp-v2
  problem_family: chemistry
  algorithm_family: qsp | qpe | amplitude_estimation | grover | custom
  precision_target:
    epsilon: 1.0e-3
    failure_probability: 1.0e-2
  assumptions:
    oracle_model: documented
    input_loading: included | excluded
    arithmetic_model: fixed_point
    error_model: logical_stochastic
  resource_estimates:
    logical_qubits: recorded
    t_count: recorded
    t_depth: recorded
    clifford_count: recorded
    magic_state_rate: recorded
  verification:
    toy_instances: included
    simulator_tests: included
    algebraic_invariants: included

57.3 Resource estimate lineage

Diagram loads as you read

57.3 Resource estimate lineage · Figure 2

View diagram source

flowchart TB
    Algorithm[Algorithm package] --> Count[Logical resource counts]
    Count --> Code[QEC code assumptions]
    Code --> Hardware[Physical hardware assumptions]
    Hardware --> Factory[Magic-state factory assumptions]
    Factory --> Estimate[Physical qubits + runtime]
    Estimate --> Sensitivity[Sensitivity analysis]

A single physical-qubit estimate without sensitivity analysis is not an estimate. It is a point guess.

57.4 Toolchain boundaries

Diagram loads as you read

57.4 Toolchain boundaries · Figure 3

View diagram source

flowchart LR
    Frontend[Algorithm frontend] --> IR[Logical IR]
    IR --> Optimize[Logical optimizer]
    Optimize --> Synthesis[Gate synthesis]
    Synthesis --> Layout[Code/layout planner]
    Layout --> Factory[Factory scheduler]
    Factory --> Runtime[Logical runtime]

QIR profiles, OpenQASM control features, and other IR layers can help separate language semantics from target capabilities, but only if the boundaries are explicit. [R94] [R96]

57.5 Magic-state accounting

Many fault-tolerant algorithms are constrained less by logical qubit count than by non-Clifford operation demand.

Diagram loads as you read

57.5 Magic-state accounting · Figure 4

View diagram source

flowchart TD
    Algorithm[Algorithm] --> Clifford[Clifford operations]
    Algorithm --> NonClifford[Non-Clifford operations]
    NonClifford --> TCount[T count]
    NonClifford --> TDepth[T depth]
    TCount --> FactoryDemand[Magic-state factory demand]
    TDepth --> RuntimeDemand[Runtime bottleneck]
    FactoryDemand --> PhysicalCost[Physical cost]
    RuntimeDemand --> PhysicalCost

Operators should track magic-state throughput as a capacity metric, not an implementation detail.

57.6 Verification ladder

Diagram loads as you read

57.6 Verification ladder · Figure 5

View diagram source

stateDiagram-v2
    [*] --> AlgebraicTests
    AlgebraicTests --> ToySimulation
    ToySimulation --> LogicalSimulation
    LogicalSimulation --> ResourceEstimate
    ResourceEstimate --> HardwareFeasibility
    HardwareFeasibility --> RuntimePilot
    RuntimePilot --> ProductionCandidate

Stage	Purpose
algebraic tests	verify identities and invariants
toy simulation	test small instances exactly
logical simulation	test encoded behavior where possible
resource estimate	expose cost and bottlenecks
hardware feasibility	compare against roadmap assumptions
runtime pilot	test operational path

57.7 Dependency risks

Diagram loads as you read

57.7 Dependency risks · Figure 6

View diagram source

flowchart TB
    Risk[Algorithm supply-chain risk] --> Oracle[Oracle or data-loading excluded]
    Risk --> Precision[Precision target unclear]
    Risk --> Compiler[Compiler optimization assumed]
    Risk --> Factory[Magic-state factory ignored]
    Risk --> Verification[No independent tests]
    Risk --> Hardware[Hardware assumptions stale]

Fault-tolerant planning must not hide data-loading and oracle costs in footnotes.

57.8 Release gates

Illustrative listing · yaml

ft_release_gate:
  candidate: algorithm-package-v2.1
  gates:
    - all assumptions machine-readable
    - resource estimates generated by at least two configurations
    - sensitivity analysis reviewed
    - toy instances pass exact simulation
    - logical operation set mapped to target roadmap
    - magic-state demand reported
    - deprecation policy defined

57.9 From roadmap to service

Google's dynamic surface-code work and other QEC advances show progress toward operational logical-qubit services, but product planning still needs an algorithm supply chain that can survive changes in code distance, decoder latency, physical error rates, and hardware topology. [R90]

Diagram loads as you read

57.9 From roadmap to service · Figure 7

View diagram source

flowchart LR
    Roadmap[Hardware roadmap] --> Assumptions[Assumption registry]
    Algorithm[Algorithm package] --> Assumptions
    Assumptions --> Estimate[Resource estimates]
    Estimate --> Decision[Investment decision]
    Decision --> Update[Update assumptions]
    Update --> Estimate

The goal is not perfect prediction. The goal is disciplined adaptation.