When designing optimization algorithms for complex engineering systems—such as electric vehicle routing, edge computing networks, or smart grids—we inevitably face a fundamental trade-off between mathematical tractability and physical accuracy. To guarantee that an algorithm can find a solution quickly, researchers frequently rely on simplified cost and energy models.
While these abstractions make optimization problems solvable, they often omit non-linearities, dynamic constraints, and environmental dependencies. A growing body of research focuses on identifying where these simplifications fail, quantifying their impact, and developing frameworks that maintain algorithmic tractability without sacrificing real-world viability.
The Tractability-Accuracy Dilemma
Why do we simplify models in the first place? Consider an optimization problem designed to schedule tasks or allocate power. To use powerful mathematical frameworks like linear programming, convex optimization, or real-time Model Predictive Control (MPC), the underlying cost functions must be well-behaved (e.g., differentiable, convex, or time-invariant).
However, physical systems are rarely well-behaved. Real batteries degrade non-linearly, wireless signals attenuate unpredictably, and economic costs scale with complex market dynamics. When we force these systems into simplified models, we create a gap between theoretical optimality and practical performance.
┌───────────────────────────┐ ┌───────────────────────────┐
│ Simplified Model │ │ Real-World System │
│ • Linear cost functions │ vs │ • Non-linear degradation │
│ • Constant parameters │ │ • Thermal dependencies │
│ • Ideal efficiencies │ │ • Dynamic overheads │
└───────────────────────────┘ └───────────────────────────┘
│ │
▼ ▼
Theoretical Optima Unexpected Failures /
(Fast, Provable, Convex) Accelerated Degradation
Where Simplifications Break Down
The tension between simplified models and real-world complexity manifests across three primary domains: electrochemical storage, hardware power profiles, and infrastructure economics.
1. Energy Storage and Electrochemical Realities
In energy-managed systems, batteries are frequently modeled as ideal reservoirs with constant efficiencies and fixed internal resistances. For instance, Chau developed competitive online algorithms for data center load balancing by assuming ideal charging and discharging efficiencies in Competitive online algorithms for geographical load balancing in data centers with energy storage. Similarly, Luo et al. relied on established, static battery models to optimize distributed real-time embedded systems in Battery-Aware Static Scheduling for Distributed Real-Time Embedded Systems.
When these algorithms are deployed, however, the physical limitations of the batteries degrade performance:
- Variable Internal Resistance: In vehicle-to-grid (V2G) systems, control strategies often emulate batteries using simplified DC power supplies. As Xia et al. note in An MPC Scheme with Enhanced Active Voltage Vector Region for V2G Inverter, neglecting changes in internal resistance during extended discharge cycles can skew accuracy and performance predictions.
- Thermal and Environmental Sensitivities: Vehicle dynamics models often assume short trip durations and constant temperatures to keep car-following equations tractable, a limitation highlighted by Lee et al. in Energy Efficient Nonlinear Microscopic Dynamical Model for Autonomous and Electric Vehicles. In reality, operating under low temperatures introduces severe non-linearities, such as solid electrolyte interphase (SEI) growth and lithium plating, which Xie et al. demonstrate in The Dilemma of C-Rate and Cycle Life for Lithium-Ion Batteries under Low Temperature Fast Charging.
2. Hardware and Network Power Profiles
In computing and communication networks, energy consumption is often abstracted as a direct, linear function of workload or message size.
- Communication Overheads: In mobile cloud computing, the energy cost of location updates is frequently modeled as directly proportional to message size. As Wang et al. point out in Cloud service-aware location update in mobile cloud computing, this abstraction overlooks hardware-specific factors like modulation, coding schemes, and tail energy states.
- Static Circuit Consumption: Network optimization models often treat circuit power consumption as an ideal or negligible constant. Pejoski et al. show in Wireless Powered Communication Networks With Non-Ideal Circuit Power Consumption that ignoring non-ideal circuit power at energy-harvesting nodes leads to suboptimal throughput and unfairness.
- Architectural Complexities: Even on modern workstations, tools designed to help developers estimate mobile app energy usage must scale down complex CPU behaviors. As Mittal et al. explain in Empowering developers to estimate app energy consumption, architectural factors and performance counter scaling challenges make perfectly accurate, lightweight power modeling incredibly difficult to achieve.
3. Socio-Economic and Infrastructure Costs
When optimization moves from physical hardware to economic deployment, cost models are often simplified into static, linear, or unit-based functions to maintain algorithmic tractability.
- Omission of Operational Dynamics: In cloud computing, economic models evaluating serverless versus serverful architectures often assume that operational costs are identical, ignoring the nuances of electrical power consumption under high core utilization (Lin et al. in Serverless Boom or Bust? An Analysis of Economic Incentives).
- Hardware-Centric Cost Functions: Network planners deploying traffic monitors often optimize strictly for hardware expenses, omitting real-world complexities like labor, real estate, and ongoing maintenance (Zang et al. in Traffic monitor deployment in IP networks).
- Subsystem Isolation: Infrastructure configuration methods for solar-thermal power stations often simplify investment costs by focusing exclusively on primary subsystems like the solar field and thermal storage, potentially overlooking auxiliary costs associated with full-scale plant deployment (Ma et al. in Optimal Configuration Method for the Installed Capacity of the Solar-Thermal Power Stations).
Bridging the Gap: Techniques for Realistic Optimization
To overcome the limitations of simplified models without making optimization mathematically impossible, researchers utilize several key strategies:
Dual Closed-Loop and Hybrid Modeling
Instead of relying on a purely physical or purely empirical model, researchers combine them. A reduced-order physical model captures the core dynamics, while an empirical model dynamically updates parameters in real time based on observed data.
For example, Xu et al. implemented this in Dual Closed-Loops Capacity Evolution Prediction for Energy Storage Batteries Integrated with Coupled Electrochemical Model. By using a dual closed-loop framework, they dynamically update parameters to account for complex operating conditions, keeping capacity prediction errors within 2% even when initial profiles are simplified.
Expanding the Control Region
Another approach is to mathematically expand the boundaries of the control signal to compensate for unmodeled hardware variations. In An MPC Scheme with Enhanced Active Voltage Vector Region for V2G Inverter, Xia et al. expanded the "Active Voltage Vector Region" within a Model Predictive Control scheme. This expansion allows the controller to generate a more continuous control signal, smoothing out the current harmonics caused by unmodeled battery dynamics.
Approximate Dynamic Programming and Robust Heuristics
When non-linear factors (like road slope or air conditioning usage in EV routing) cannot be ignored, traditional exact optimization fails. Researchers turn to Approximate Dynamic Programming (ADP) or robust heuristics to handle these time-dependent, stochastic factors without suffering from the curse of dimensionality (Kan et al. in Simulation-Based Electric Vehicle Sustainable Routing with Time-Dependent Stochastic Information).
Conclusion
Simplified cost and energy models are a necessity for system optimization, but they are not a final destination. The core mechanism driving this research field is iterative refinement: starting with a tractable, simplified model to establish baseline guarantees, identifying the specific real-world variables that degrade that baseline, and introducing targeted mathematical corrections to bridge the gap. Understanding why an optimization works requires recognizing not just the elegance of the mathematics, but the physical friction it abstracts away.
Key Foundational Work
Optimal Energy Aware Clustering in Sensor Networks (2002) — contributed 1 concept elements to this direction.