Controllability¶
Core Idea¶
Controllability is the structural property that determines whether an agent's available inputs can steer a system's state into any desired region such that: (1) a system is controllable when, for any initial state \(x_0\) and any target state \(x_1\), there exists an admissible input trajectory \(u(t)\) that drives the system from \(x_0\) to \(x_1\) in finite time — formally, for a linear time-invariant system \(\dot x = Ax + Bu\), controllability reduces to the rank condition on the controllability matrix \(\mathcal{C} = [B, AB, A^2 B, \ldots, A^{n-1} B]\) (full rank \(\Leftrightarrow\) controllable); for nonlinear systems, the analogous condition uses Lie bracket algebra (Chow's theorem, Sussmann's controllability criterion); practical operational definitions in management contexts test whether "interventions of specified type-and-magnitude can move this system's key variables through the desired range"[1]; (2) controllability is the information-theoretic dual of observability (see #390) — observability asks "do outputs reveal state?"; controllability asks "can inputs steer state?"; Kalman's 1960 seminal work established this duality via \((A, B) \text{ controllable} \Leftrightarrow (A^T, B^T) \text{ observable}\), making the two reciprocal structural properties of the same state-space model; (3) controllability delivers the structural precondition for intervention, healing, policy, and goal-directed action — without controllability, desired states are structurally unreachable no matter how clever the strategy; software systems without deployable fixes cannot be healed; patients without effective treatments cannot recover; policies without available levers cannot reshape outcomes; organizations without change-authority cannot execute transformations; recognizing uncontrollability prevents wasted effort on impossible goals and redirects attention to structural redesign; (4) the concept generalizes across domains — control engineering (Kalman controllability, pole-placement via state feedback, linear-quadratic regulator design, reachability analysis for safety-critical systems)[1], software engineering and infrastructure (deployability, rollbackability, feature-flag control surfaces, deployment authority structure; controllability-of-production as a core SRE concern; circuit-breakers, rate-limiters, and admin APIs as controllability levers)[2], medicine and public health (medication effects, surgical interventions, behavior-change levers — each a controllability claim; uncontrollable conditions trigger different strategies — palliation rather than cure), policy and governance (which economic variables can tax-and-subsidy policy actually move? Which social variables resist all known interventions?), game theory and economics (controllability of outcomes in principal-agent games, information-asymmetric markets), biology (controllability of cellular and organismal states via genetic, pharmacological, or developmental interventions; network controllability in systems biology — Barabási-Liu-Slotine 2011), climate and earth systems (controllability of climate via anthropogenic intervention is limited; geoengineering debates are partly controllability questions) — all deploy the "can intervention move the state to the target?" structural question.
How would you explain it like I'm…
Can You Steer It?
Whether You Can Steer It
Controllability
Structural Signature¶
A triple \((X, U, f)\) where \(X\) is the state space, \(U\) is the admissible-input space, and \(f: X \times U \to TX\) (vector field, continuous dynamics) or \(f: X \times U \to X\) (discrete dynamics) defines how inputs move state[3]. Controllability asks whether, for every pair \((x_0, x_1)\), some admissible input trajectory steers \(x_0\) to \(x_1\) in finite time (or, in bounded-time controllability, within a specified horizon). For linear systems, the controllability matrix rank condition is necessary and sufficient; for nonlinear systems, the controllability-Lie-algebra rank condition generalizes. Variants include: structural controllability (generic controllability based on system-matrix sparsity pattern, independent of parameter values); small-time local controllability (controllability within a neighborhood, possibly within a short horizon); robust controllability (preserved under model uncertainty); constrained controllability (controllability when inputs are restricted to a subset — e.g., bounded thrust); controllability gramian (quantifies how controllable each state direction is, supporting model reduction and actuator placement)[4]; network controllability (controllability of large-scale interconnected dynamical systems via input placement at selected nodes — a rich theoretical and applied literature since Liu-Slotine-Barabási 2011).
What It Is Not¶
- Not authority or command-chain control — controllability in the engineering sense is a structural property of the state-input dynamics; organizational-authority language uses "control" more broadly. A manager has organizational authority but may face uncontrollable outcomes (market forces, regulatory environment, employee motivation) — authority and controllability are independent concepts that are often conflated.
- Not observability — controllability moves state; observability sees state. The two are Kalman-dual structural properties; both are required for full state-feedback control. A controllable-but-unobservable system can be pushed in any direction but you don't know where you're pushing from; an observable-but-uncontrollable system is diagnostically rich but operationally helpless.
- Not stability — a system may be stable (disturbances damp out) but uncontrollable (inputs can't move it to desired states), or controllable but unstable (inputs can move it but left alone it diverges); the concepts are orthogonal. Feedback control combines both: use controllability to drive state while compensating instability.
- Not manipulability in a manipulation-or-exploitation sense — controllability is a neutral structural property; whether using it is ethical depends on the action and context. Controllability of a medical condition enables treatment (good); controllability of a population's beliefs enables manipulation (ethically fraught). The ethical questions are separate from the structural property.
- Not a guarantee of optimality — controllability says some input trajectory can reach the target; it says nothing about cost, efficiency, or robustness. Optimal control theory (LQR, model-predictive control, Pontryagin's maximum principle) addresses cost-optimal controllability. A controllable-but-expensive-to-control system may be practically unreachable given budget constraints.
Broad Use¶
- Control engineering (core domain): Kalman's controllability (1960) and the reachability-canonical-form decomposition; state-feedback pole placement; linear-quadratic-regulator design; reachability analysis for safety-critical systems (verifying that unsafe states are not reachable); network controllability (Liu-Slotine-Barabási 2011 — identifying minimum sets of driver nodes that control complex networks, applied to biology, infrastructure, and social networks).
- Software and infrastructure engineering: Deployability, rollback-ability, feature-flag infrastructure, config-management systems as controllability levers; "controllability of production" as core SRE concern; circuit breakers and rate limiters as controllability mechanisms; canary deployment as low-risk controllability exercise; chaos engineering (Netflix Chaos Monkey) as controllability-under-failure verification.
- Medicine and public health: Pharmacotherapy (drug-effect landscape as controllability map), surgery (mechanical controllability of anatomy), behavior change (limited controllability of lifestyle via clinical intervention); uncontrollable conditions trigger distinct strategies (palliative care, disease-modifying therapies rather than cure); network medicine studies controllability of biological networks for therapeutic targeting.
- Policy and governance: Monetary policy (interest rates as controllability lever for inflation and employment), fiscal policy (taxation and spending), regulatory policy (rules as controllability of firm behavior); often partial controllability — policy can move some variables and not others, and indirect effects (substitution, adaptation) degrade controllability; political-economy controllability (how responsive are voters to policy?).
- Biology and systems biology: Controllability of cell state via genetic circuits (synthetic biology); pharmacological controllability of disease (targeted therapies); gene-regulatory-network controllability; Barabási-Liu-Slotine network-controllability analysis applied to protein interaction networks identifying drug targets.
- Game theory and economics: Controllability of outcomes in principal-agent games (incentive alignment); market-power as controllability of prices; regulator controllability of firm behavior; AI-alignment as controllability of advanced AI system behavior (an active research frontier).
- Climate and earth systems: Controllability of atmospheric CO₂ via emissions reductions; geoengineering debates as proposals to expand controllability (solar radiation management, carbon dioxide removal); limits of climate controllability given feedback loops and slow ocean-response times.
- Robotics and autonomous systems: Motion-planning controllability (can the robot reach the goal from the current state given its actuator dynamics?); controllability of underactuated systems (acrobot, cart-pole, hovercraft with limited actuators — often controllable but only via non-trivial control strategies).
Clarity¶
Names the structural precondition for intervention and goal-directed action. Without the controllability frame, analysts may persist in trying to move uncontrollable variables (wasting effort), may overlook controllable leverage points (missing opportunity), or may conflate organizational authority with actual controllability (authority without controllability is powerless; controllability without authority is unexpectedly potent). With the frame, the analyst asks: which states are reachable from the current state via available inputs? Which states are structurally unreachable? What new inputs would expand the reachable set? What is the cost of reaching a given target? This structural clarity supports principled intervention design, honest diagnosis of "things we cannot fix from here," and systematic investment in controllability-expanding infrastructure (actuators, deployment pipelines, policy instruments).
Manages Complexity¶
Compresses intervention planning into a reachability question. Instead of enumerating every possible action sequence, controllability analysis identifies what set of states is reachable from the current state and separates feasible targets from infeasible ones. This enables principled actuator placement (which actuators maximize reachable-set volume?), principled priority-setting (which target states are both valuable and reachable?), and principled impossibility arguments (uncontrollable states are not reachable by any strategy). Network controllability analysis compresses controllability of million-node systems into small driver-node sets; pole-placement compresses stability-and-performance objectives into linear algebra on controllability-canonical-form matrices. The frame supports decomposition: reducing a complex controllability question to a sequence of simpler ones by waypoint planning or hierarchical control.
Abstract Reasoning¶
The controllability abstraction asks: what is the state space? What are the available inputs and their constraints? Is the desired state reachable from the current state? If not, can we expand the input set to make it reachable? What is the cost of each reachable target? What is the reachable-set geometry (shape, volume, time-to-reach)? This transfers across physical control systems, software infrastructure, medical intervention, policy design, biology, economics, and autonomous systems. A mature analysis identifies uncontrollable states explicitly (accepting limits on intervention), quantifies input authority and reachable-set properties, and balances investment in expanding controllability against acceptance-based responses for the truly uncontrollable. Immature analysis assumes controllability without verification, conflates authority with actual reach, or abandons intervention entirely when partial controllability would still deliver value.
Knowledge Transfer¶
| Domain | State \(x\) | Inputs \(u\) | Controllability question |
|---|---|---|---|
| Spacecraft | Position, velocity, attitude | Thrusters, reaction wheels | Can we reach target orbit? |
| Software production | System configuration, running version | Deployments, feature flags | Can we roll back / push fixes? |
| Patient physiology | Disease state | Medications, surgery, behavior | Can we achieve remission? |
| Economy | GDP, inflation, employment | Interest rates, fiscal policy | Can policy hit targets? |
| Gene network | Expression state | Drugs, genetic interventions | Can we control cell fate? |
| Robot | Joint positions, velocity | Motor torques | Can we reach the goal? |
| Climate | CO₂ concentration, temperature | Emissions, geoengineering | Can we stabilize warming? |
| Organization | Strategy, culture, capabilities | Hiring, structure, incentives | Can we execute transformation? |
| Financial market | Prices, volumes | Orders, market-making | Can we move price against order flow? |
| AI system | Model behavior, outputs | Training, prompts, RLHF | Can we align model with intent? |
Across rows, the "reachable-via-inputs" pattern transfers with full structural fidelity. Cross-domain transfer is strong: the control engineer's reachability-set geometry informs software-deployment safety analysis; the systems-biologist's network-controllability methods inform drug-target prioritization; the political economist's policy-lever analysis informs organizational-transformation planning. Controllability is one of the most-transferable engineering abstractions for goal-directed action.
Examples¶
Formal/abstract¶
Spacecraft thruster controllability[5]. Consider a rigid-body spacecraft in low Earth orbit with three thrusters mounted for attitude control. The dynamics are \(\dot\omega = J^{-1}(\tau - \omega \times J\omega)\) where \(\omega\) is angular velocity, \(J\) is the inertia tensor, and \(\tau\) is the control torque produced by thrusters. With three thrusters along orthogonal body axes, the system is fully controllable: any desired angular-velocity vector can be produced. If one thruster fails (say, the z-axis thruster), the system becomes underactuated in the attitude-rate space — only torques in the \(xy\)-plane can be directly commanded. Controllability analysis via Lie brackets reveals whether angular velocity about the z-axis is still reachable through combinations of x-y thrusting and the nonlinear gyroscopic coupling \(\omega \times J\omega\): for a generic (non-symmetric) inertia tensor, the Lie bracket \([f_x, f_y]\) spans the z-axis direction, so the system remains controllable but only via non-trivial maneuvers (spin up x-axis and y-axis to couple into z-axis through the cross product). This is the classical "satellite with two thrusters" problem. Practical consequence: the mission can still recover orientation but requires more complex (longer-time, higher-fuel) maneuvers. If two thrusters fail (only one remaining along a single axis), controllability analysis shows the system loses full controllability; certain attitudes become unreachable without external influence, and the spacecraft is declared lost or placed in safe mode. This example illustrates the reach of controllability theory: it quantifies exactly what remains achievable under actuator degradation, informs fault-tolerant-control design (pre-position redundant actuators, certify fault-response protocols), and supports explicit impossibility claims when they hold. Aerospace missions (from Apollo-era fault-tolerance to Mars-rover stuck-wheel recovery to recent SpaceX booster-recovery) have all used controllability reasoning to decide whether operations can continue.
Mapped back: Instantiates the structural signature directly — controllability triple (X, U, f), rank condition on the controllability matrix, Lie-bracket analysis for nonlinear extension, structural controllability under parameter-independent rank conditions, and gramian-based directional analysis supporting actuator placement. Lie-bracket reasoning quantifies what remains achievable under actuator degradation, supporting fault-tolerant design and impossibility claims.
Applied/industry¶
A platform engineering team at a global e-commerce company builds its "production controllability" initiative around explicit controllability reasoning[6]. The business problem: production incidents are increasingly expensive (lost revenue, reputational damage, regulatory attention), and response time depends critically on how much of the production system can be rapidly changed — undeployable systems have near-zero controllability and long resolution times; highly controllable systems resolve incidents in minutes. The team's work includes: (a) controllability inventory — cataloging every production variable (configuration parameters, feature flags, traffic-routing weights, resource allocations) and the "input channels" that can change each (deployments, flag flips, admin API calls, manual console operations) with associated latency, blast radius, and authorization requirements; (b) deployment pipelines as controllability infrastructure — the team invests in CI/CD sophistication (fast deploys, automated tests, canary stages, automatic rollback) because each pipeline improvement expands the reachable-set volume of production changes within a given time budget; © feature flags as real-time controllability — feature flags provide near-instant control over application behavior without deployment; the platform emphasizes flag hygiene (flag ownership, lifecycle, naming) because flag proliferation without cleanup yields a chaotic controllability surface; (d) admin APIs and runbooks — every production subsystem has documented runbooks that name the controllability levers (what can be changed, by whom, via what interface, with what blast radius); incident response references runbooks directly; (e) controllability-as-a-metric — the team measures "time-to-change" for common categories of production change (deploy a code fix, flip a feature flag, reallocate resources, redirect traffic) and tracks improvement over quarters; these metrics feed performance reviews; (f) controllability gaps as engineering backlog — post-mortems frequently identify controllability gaps (a variable that matters but cannot be easily changed during an incident); gaps become instrumented engineering projects; (g) blast-radius analysis for each controllability lever — because high controllability can be dangerous (a wrong flip can break production), the team carefully analyzes blast radius for each lever, restricts high-risk levers behind authorization gates, and invests in safe-deploy mechanisms; (h) chaos-engineering as controllability verification — the team regularly injects failures to verify that claimed recovery controllability actually works; claimed controllability that hasn't been tested is treated as unverified. The team's director of platform engineering describes the program as "treating production like a controllable system — we expand what can be changed, reduce the time and risk of changing it, and measure our control surface explicitly." Customer-facing reliability improves markedly as a result: the site's high-priority-incident MTTR drops from hours to minutes over two years. The practice is a direct, industrial-scale transfer of control-engineering controllability principles into production software operations.
Mapped back: Shows the same structural signature instantiated in industrial-scale software operations — controllability inventory as enumeration of (X, U) pairs, deployment pipelines + feature flags as input channels with quantified latency and blast radius, time-to-change metrics as effective-controllability measurements, and chaos-engineering as run-time controllability verification. The hours→minutes MTTR collapse is the operational signature of expanding production controllability through deliberate platform engineering.
Structural Tensions¶
T1 — Controllability cost versus value — actuator proliferation and safety risk[7]. Expanding controllability (adding actuators, deploying feature flags, increasing admin API surface) costs engineering effort and increases risk (each new control lever can be misused or can fail in unanticipated ways). The tension between "what we might need to change" and "what we can safely expose as changeable" drives platform engineering: flag proliferation causes chaos; flag scarcity causes inflexibility; finding the balance is ongoing work.
T2 — Controllability versus stability / unintended consequences[8]. High controllability permits rapid change, which can destabilize the system. Feedback effects, time delays, and nonlinearities mean that naive use of controllability levers can push the system into undesirable states. The tension between using controllability aggressively (fast response) and cautiously (avoiding unintended consequences) is mediated by testing, canary deployment, feature-flag staged rollout, and human review; no universal right balance exists.
T3 — Structural controllability versus effective controllability under cost constraints[9]. The system may be controllable in principle (formal rank condition satisfied) but require enormous resources to reach certain states (high input energy, long horizons, many actuators operating in coordination). The distinction between in-principle reachable and practically reachable within budget is essential in real-world problems: climate mitigation is formally controllable via emissions reductions, but politically- and economically-constrained effective controllability is much tighter.
T4 — Controllability versus observability imbalance[10]. Controllability and observability are Kalman-dual, and balanced investment yields the strongest posture. Systems heavy in one and light in the other exhibit distinctive pathologies: highly controllable-but-unobservable (you change things without knowing their current state — courts disaster); highly observable-but-uncontrollable (you see everything happening but cannot intervene — chronic frustration). The reciprocal tight-pair structure makes joint investment prudent; tracking both together prevents imbalance.
T5 — Controllability constraints and the design of reachable regions[11]. Real systems have bounded inputs (thrusters with limited thrust, software deployments that take time, policy instruments with political limits); these constraints define what's reachable within a budget. The tension is between designing for nominal operation (use modest control inputs) and designing for recovery from extreme states (need large inputs for emergency maneuvers). Underestimating input constraints leads to unfeasible control plans; overestimating them wastes potential.
T6 — Controllability and unforced evolution of the system[12]. Even fully controllable systems have drift: without continuous intervention, they evolve toward equilibria or attractors. The tension is between active control (continuously apply inputs to maintain desired state) and letting favorable natural dynamics work (designing the system so it naturally evolves toward desired states, reducing control effort). Mature control design often combines both: steering with natural dynamics rather than fighting them.
Structural–Framed Character¶
Controllability sits at the structural end of the structural–framed spectrum: it is a pure relational pattern, the same in any domain where it appears, and nothing about its meaning depends on a particular field's vocabulary or assumptions. It names whether an agent's available inputs can steer a system from any initial state to any target state in finite time—reducing, for a linear time-invariant system, to a rank condition on the controllability matrix.
The definition is given by the formal triple of a state space, an admissible-input space, and a dynamics map, and it transfers without change whether the system is an aircraft's attitude, a chemical reactor's temperature, or an economic model's state variables. It carries no evaluative weight: controllability is a structural fact about reachability, not a verdict. Its origin is mathematical rather than institutional, it can be stated without reference to human practices, and applying it feels like reading a property the system's dynamics already possess. On every diagnostic, it reads structural.
Substrate Independence¶
Controllability is a highly substrate-independent prime — composite 4 / 5 on the substrate-independence scale. Its signature — an agent able to steer a system to any desired state via admissible inputs — has a purely structural formal definition and reaches across engineering design, systems theory, mathematics, and organizational systems. The abstraction itself is tier-1 clean. What holds it a notch below the ceiling is that nearly all the examples come from spacecraft dynamics and platform engineering, both computational-engineering domains, leaving transfer to biological or social systems comparatively unexplored.
- Composite substrate independence — 4 / 5
- Domain breadth — 4 / 5
- Structural abstraction — 5 / 5
- Transfer evidence — 3 / 5
Neighborhood in Abstraction Space¶
Controllability sits in a sparse region of abstraction space (76th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely rather than landing on a neighbor.
Family — Computational Process & Control (12 primes)
Nearest neighbors
- Continuity — 0.80
- Observability — 0.78
- Homeostasis — 0.78
- Algorithm — 0.75
- State and State Transition — 0.75
Computed from structural-signature embeddings · 2026-05-29
Not to Be Confused With¶
Controllability must be distinguished from Observability, its Kalman-dual tight-pair partner, because the two are reciprocal but structurally distinct properties. Controllability asks "can inputs move the system to desired states?" — it is about actuation authority and state reachability. Observability asks "do outputs reveal the system's internal state?" — it is about sensing and state inference. The two are Kalman-dual in the formal sense: a system is controllable if and only if its dual (transpose of the system matrices) is observable. Both are required for full state-feedback control: without controllability, the system cannot be steered to the desired state no matter how clever the feedback law; without observability, the current state cannot be inferred from measurements, so the feedback law has no basis. A controllable-but-unobservable system can be pushed in any direction but the operator doesn't know where it is being pushed from (flying blind). An observable-but-uncontrollable system is diagnostically rich (everything can be measured) but operationally helpless (nothing can be moved). Mature control design balances investment in both; systems heavy in one and light in the other exhibit characteristic pathologies.
Controllability is also distinct from Stability, though the two are frequently coupled in practice. Stability describes whether a system's dynamics naturally damp out disturbances (a ball in a bowl naturally returns to the bottom if nudged; an inverted pendulum tips over). Controllability describes whether external inputs can move the system from one state to another. A stable system may be uncontrollable (the dynamics are good, but no inputs reach the desired state); an unstable system may be controllable (inputs can steer it, but the system will diverge if left alone without constant control). Feedback control combines both: use controllability to drive the state toward desired conditions while compensating for instability via the feedback law. The distinction prevents the confusion that "if the system is stable, control is easy" (not necessarily — stability helps but does not guarantee controllability) or "if we can control it, we're done" (not necessarily — controllability without stability requires active continuous compensation).
Controllability differs from Authority or Command-Chain Control in organizational language. Authority in the management sense is the organizational right or permission to make decisions and direct resources. A manager may have organizational authority (the right to delegate tasks, approve spending) but face uncontrollable outcomes (market forces, regulatory environment, employee motivation). Authority is a social/institutional property; controllability is a structural property of the system dynamics. A manager with authority can direct action but has no guarantee of outcome if the system is not structurally controllable. Conversely, a practitioner without formal authority may discover controllable leverage points (ways to move the system through indirect influence) that more formally powerful people missed. The distinction prevents conflating organizational rank with actual system reach.
Controllability is not Optimality or Efficiency. Controllability says some input trajectory can reach the target state; it says nothing about cost, energy consumption, or speed. A system may be controllable but require enormous inputs (saturating all available actuators) or prohibitively long time horizons to reach the desired state. Optimal control theory (LQR, Pontryagin's maximum principle, model-predictive control) addresses cost-optimal controllability — balancing the goal of reaching a state against the cost of reaching it. A controllable-but-expensive-to-control system may be practically unreachable within budget constraints even though reachable in principle. The distinction emphasizes that controllability is a necessary condition for goal-directed action but not sufficient — cost and feasibility require additional analysis.
Solution Archetypes¶
Solution archetypes in the catalog that build on this prime — directly (this prime is a source ingredient) or as a related prime.
Built directly on this prime (5)
- Control Delegation
- Control Surface Creation
- Controlled Phase Transition
- Observability Instrumentation
- Phase-Space Mapping
Also a related prime in 65 archetypes
- Activation Energy Cost-Benefit Analysis
- Adaptive Reconfiguration
- Adaptive Scheduling
- Adaptive Threshold Recalibration
- Approach–Avoidance Decomposition
- Autonomous Action Zone Protection
- Balancing Loop Stabilization
- Checkpoint and Rollback
- Constraint Envelope Adjustment
- Constraint Propagation and Decoupling
Notes¶
Engineering-origin with mathematical formalization — Kalman established the rank-based characterization of controllability for linear systems (1960), giving the concept formal foundations[1]. Predecessors include Cauchy's work on ODE reachability and classical mechanics' notion of controllability via external forces. Systems-thinking framings extend the concept into organizational and policy domains. Nonlinear controllability (Chow's theorem 1939, Sussmann 1983) and network controllability (Liu-Slotine-Barabási 2011) are major theoretical extensions. Companion to #390 observability (Kalman dual — reciprocal tight pair; the classic structural duality in state-space theory), #388 homeostasis (controllability is the actuation prerequisite for homeostatic regulation), #387 requisite_variety (controllability must have sufficient variety to handle disturbance variety), #394 leverage_points (leverage-point analysis is partial-controllability analysis — finding the most controllable intervention points in a system), and #71 feedback_loop (closed-loop control requires both observability and controllability)[13]. Strong transfer targets: aerospace and autonomous-vehicle safety certification, production software controllability platforms, pharmacological-target-selection in network medicine, policy-lever design in regulatory economics, AI-alignment research, and any engineering or management context where "can we actually move this variable?" is a binding question. Review flag: tight_pair_with_observability — the two concepts are paradigmatic Kalman duals, structurally reciprocal, and best analyzed jointly; their separation produces distinctive pathologies.
References¶
[1] Kalman, R. E. (1960). "On the general theory of control systems." Proceedings of the First IFAC Congress, 1, 481–492. ↩
[2] Beyer, B., Murphy, N. R., Rensin, D. K., Kawahara, K., & Thorne, S. (Eds.). (2018). The Site Reliability Workbook: Practical Ways to Implement SRE. O'Reilly Media. ↩
[3] Kailath, T. (1980). Linear Systems. Prentice Hall. ↩
[4] Moore, B. C. (1981). "Principal component analysis in linear systems: Controllability, observability, and model reduction." IEEE Transactions on Automatic Control, 26(1), 17–32. ↩
[5] Sussmann, H. J. (1983). "Lie brackets, real analyticity, and geometric control." In Differential Geometric Control Theory, 1–116. Birkhäuser. ↩
[6] Humble, J., & Farley, D. (2010). Continuous Delivery: Reliable Software Releases through Build, Test, and Deployment Automation. Addison-Wesley. ↩
[7] Perrow, C. (1984). Normal Accidents: Living with High-Risk Technologies. Basic Books. ↩
[8] Åström, K. J., & Murray, R. M. (2008). Feedback Systems: An Introduction for Scientists and Engineers. Princeton University Press. ↩
[9] Bryson, A. E., & Ho, Y.-C. (1975). Applied Optimal Control: Optimization, Estimation, and Control. Hemisphere Publishing. ↩
[10] Skogestad, S., & Postlethwaite, I. (2007). Multivariable Feedback Control: Analysis and Design (2nd ed.). Wiley. ↩
[11] Wonham, W. M. (1985). Linear Multivariable Control: A Geometric Approach (3rd ed.). Springer-Verlag. ↩
[12] Ashby, W. R. (1956). An Introduction to Cybernetics. Chapman & Hall. ↩
[13] Liu, Y.-Y., Slotine, J.-J., & Barabási, A.-L. (2011). "Controllability of complex networks." Nature, 473(7346), 167–173. ↩
[14] Chow, W. L. (1939). "Über Systeme von linearen partiellen Differentialgleichungen erster Ordnung." Mathematische Annalen, 117, 98–105.
[15] Liu, Y.-Y., Slotine, J.-J., & Barabási, A.-L. (2011). "Controllability of complex networks." Nature, 473(7346), 167–173.