QCloud Quantum Correction Layer
Selective refinement layer for applying bounded, higher-resolution force corrections to prioritized regions during a running simulation.
Interpretation note. In this documentation, QCloud denotes a selective refinement pathway used to inject higher-cost reference corrections on prioritized regions. The term should be read as an implementation label for the current workflow rather than as a blanket claim of full-system ab initio dynamics.
▶ COMPONENTS — click any component to expand implementation details
Region Selector
Priority-based molecular subgraph selection
At each scheduled refinement step, the region selector identifies which parts of the system should receive higher-resolution reference corrections. Selection criteria include:
- Structural importance (binding interfaces, catalytic sites)
- Prior correction magnitudes (particles that received large corrections previously)
- Event analyzer priority scores (regions where structural events are occurring)
- Computational budget (total QCloud compute is bounded per step)
Regions are molecular subgraphs — connected sets of CG beads that form a physically meaningful unit for quantum evaluation.
- Structural importance (binding interfaces, catalytic sites)
- Prior correction magnitudes (particles that received large corrections previously)
- Event analyzer priority scores (regions where structural events are occurring)
- Computational budget (total QCloud compute is bounded per step)
Regions are molecular subgraphs — connected sets of CG beads that form a physically meaningful unit for quantum evaluation.
Reference Correction Model
Higher-resolution force deltas on selected regions
For each selected region, the refinement layer evaluates a higher-resolution reference model and computes the difference from the base classical force estimate. The output is a set of per-particle force deltas:
Each force delta is bounded to protect numerical stability:
Energy deltas are also bounded so the refinement layer does not dominate the base CG dynamics.
ΔFQCloud(i) = Freference(i) − Fclassical(i)
Each force delta is bounded to protect numerical stability:
|ΔFcomponent| ≤ 5.0 kJ/(mol·nm)
Energy deltas are also bounded so the refinement layer does not dominate the base CG dynamics.
Event Analyzer
Structural event detection from correction patterns
Maintains per-particle statistics of correction magnitudes using exponential moving averages. When a correction spike exceeds baseline + nσ, it classifies the event:
Bond forming — large correction on particles moving closer together
Bond breaking — large correction on particles moving apart
Conformational shift — sustained elevated correction above baseline
Interface rearrangement — correlated correction spikes across a binding interface
Detection threshold: correction > baseline + spike_threshold_sigma × std
Bond forming — large correction on particles moving closer together
Bond breaking — large correction on particles moving apart
Conformational shift — sustained elevated correction above baseline
Interface rearrangement — correlated correction spikes across a binding interface
Detection threshold: correction > baseline + spike_threshold_sigma × std
Adaptive Feedback Loop
Corrections inform next region selection
Event detection generates per-particle priority scores that feed directly back into the next region selection cycle:
This creates an closed-loop prioritization mechanism in which higher-cost refinement is preferentially directed toward regions showing the strongest evidence of change or physical relevance. As the system evolves, the selected regions can shift accordingly.
priority(i) = base_score(i) + min(0.5, |ΔF(i)| / baseline(i))
This creates an closed-loop prioritization mechanism in which higher-cost refinement is preferentially directed toward regions showing the strongest evidence of change or physical relevance. As the system evolves, the selected regions can shift accordingly.
force coupling
propagation chain
▶ HOW CORRECTIONS REACH ATOMIC COORDINATES — the propagation path from selective refinement to downstream coordinate analysis
QCloud → Forces → Velocities → Positions → Trajectory → Back-map → AA
Complete propagation path
The QCloud force correction ΔF on particle i at step t affects the position at step t+1 through the integrator:
This position change is stored in the trajectory. When back-mapped to AA, the reconstructed coordinates inherit the propagated state produced by the refinement step. Downstream contact, hydrogen-bond, and interface analyses are therefore conditioned by the same simulation history.
Δx(QCloud) = ½ Δt² · ΔFQCloud / m
This position change is stored in the trajectory. When back-mapped to AA, the reconstructed coordinates inherit the propagated state produced by the refinement step. Downstream contact, hydrogen-bond, and interface analyses are therefore conditioned by the same simulation history.