Most algorithmic frameworks treat a battery like a bucket of water: pouring energy in or drawing it out occurs at a constant, linear rate, and the bucket's capacity degrades predictably based on how much water it has held. As noted by Mischa Döhler et al. in Novel Techniques for Analysis and Design of Cross-Layer Optimized Wireless Sensor Networks, this linear simplification is computationally convenient but physically flawed.
Chemical batteries are highly non-linear systems. Treating them as linear reservoirs leads to catastrophic scheduling errors, premature hardware failure, and highly inaccurate lifetime predictions. To build efficient energy systems—whether for electric vehicles, drones, or smart grids—we must model the underlying non-linear electrochemical dynamics.
The Recovery Effect: Why Resting Restores Capacity
One of the most counterintuitive non-linear phenomena is the chemical recovery effect (or capacity regeneration). When a battery undergoes a period of high demand, its available capacity appears to drop rapidly. However, if the battery is allowed to rest, it temporarily recovers some of this lost capacity.
Daniele Magazzeni et al. demonstrate in Automatic Construction of Efficient Multiple Battery Usage Policies that scheduling policies that exploit this effect by intelligently switching between multiple resting batteries can extend device lifetimes by up to 20%.
To understand why this happens, we have to look at the diffusion of active materials inside the cell. During intense discharge, the concentration of reactants near the electrodes drops faster than they can be replenished from the bulk electrolyte, causing a premature voltage drop that makes the battery appear empty. Resting allows these concentrations to equalize.
Heng Wang et al. capture this in Modeling, metrics, and optimal design for solar energy-powered base station system using the Kinetic Battery Model (KiBaM), which mathematically splits the battery's charge into an "available" charge well and a "bound" charge well.
Mengda Cao in A Method for Interval Prediction of Satellite Battery State of Health Based on Sample Entropy and Changsong Ma et al. in Prediction of power battery health state based on data-driven Gaussian process regression algorithm emphasize that failing to track these non-linear capacity jumps causes traditional models to misinterpret recovery as random noise, severely reducing predictive sensitivity.
The Tail-End Slowdown: Non-Linear Charging Curves
Another critical failure of linear modeling occurs during charging. A common assumption in vehicle routing problems is that a vehicle charges at a constant rate. In reality, as a battery's State of Charge (SOC) increases, its internal chemical resistance rises, and the charging rate slows down dramatically—particularly during the final third of the cycle.
Lihui Yi explores this in Optimal EV Charging Decisions Considering Charging Rate Characteristics and Congestion Effects, showing how this non-linear curve fundamentally alters optimal routing and station-congestion dynamics.
Similarly, Yuzhen Zhou in The electric vehicle routing problem with partial recharge and vehicle recycling and Rito Brata Nath et al. in On the Impact of Co-Optimizing Station Locations, Trip Assignment, and Charging Schedules for Electric Buses highlight that ignoring this tail-end slowdown results in fleet schedules that are physically impossible to execute, causing unexpected operational delays in real-world environments.
Path-Dependent Wear: Non-Linear Degradation
Linear models assume that battery degradation is a fixed, constant cost per unit of energy cycled. However, physical battery wear is highly non-linear and path-dependent. It depends heavily on temperature fluctuations, depth of discharge (DoD), and current spikes.
Michael David Taylor shows in Joule Counting Correction for Electric Vehicles Using Artificial Neural Networks that a battery's state is a non-linear function of its operational history, driven by factors like internal resistive heating. When batteries participate in high-frequency operations like grid arbitrage, these non-linearities compound.
Yuanyuan Shi et al. in Leveraging energy storage to optimize data center electricity cost in emerging power markets note that incorporating sophisticated aging models is vital to prevent systems from "trading themselves to death" during market participation.
Marek Petrik et al. in Optimal Threshold Control for Energy Arbitrage with Degradable Battery Storage and Enrico H. Gerding et al. in Online Mechanism Design for Vehicle-to-Grid Car Parks warn that while linear assumptions make mathematical proofs of optimality or game-theoretic truthfulness easier, they create a dangerous gap between theoretical models and physical reality.
Rate Capacity and Voltage Sag
Finally, high-current demands trigger non-linear drops in effective capacity, known as voltage sag. Davide Scaramuzza et al. observe this in Range, Endurance, and Optimal Speed Estimates for Multicopters, noting that aggressive drone maneuvers draw high currents that cause sudden voltage drops. This means the battery runs out of usable energy much faster than a linear model would predict, making accurate flight-time estimation impossible without accounting for dynamic current loads.
Key Foundational Work
An energy framework for the network simulator 3 (NS-3) (2011) — contributed 1 concept elements to this direction.
Novel Techniques for Analysis and Design of Cross-Layer Optimized Wireless Sensor Networks (2007) — contributed 1 concept elements to this direction.
Automatic Construction of Efficient Multiple Battery Usage Policies (2011) — contributed 1 concept elements to this direction.