Mathematical optimization - Practical Applications

Applications of Mathematical Optimization Introduction Mathematical optimization is not merely a theoretical subject—it's one of the most practically important tools in modern science, engineering, and business. Nearly every field that involves decision-making under constraints applies optimization techniques. Understanding these applications helps you see why the optimization concepts you're learning matter, and it prepares you to recognize optimization problems when you encounter them in your career. Economics and Finance CRITICALCOVEREDONEXAM Economics is fundamentally built on optimization problems. At its core, economic theory models two types of decision-makers: Consumers maximize utility, which means they choose consumption bundles to get the most satisfaction given their budget constraints. Mathematically, a consumer solves: $$\max{x1, x2, \ldots} u(x1, x2, \ldots) \quad \text{subject to} \quad p1 x1 + p2 x2 + \cdots \leq B$$ where $u$ is the utility function, $pi$ are prices, and $B$ is the budget. This is a constrained optimization problem you'll recognize immediately: the objective is utility, and the constraint is the budget line. The dual problem is expenditure minimization: given a desired level of utility, a consumer minimizes the cost to achieve it. These dual problems—maximization and minimization—are mathematically linked and provide deep insights into consumer behavior. Firms maximize profit, which means they choose production levels and input quantities to maximize revenue minus costs. This is similar in structure to consumer optimization but with different variables and constraints. Portfolio optimization is a particularly important application that most likely will appear on your exam. A portfolio manager must balance two competing objectives: Maximize expected return from the portfolio Minimize risk (typically measured as variance or volatility) This is fundamentally a multi-objective optimization problem. The manager cannot achieve the highest possible return without accepting higher risk. The solution is the efficient frontier—a set of portfolios where you cannot increase return without increasing risk, or decrease risk without decreasing return. Investors choose where along this frontier they prefer to be based on their risk tolerance. Operations Research CRITICALCOVEREDONEXAM Operations research (OR) is the discipline of applying mathematical optimization to real-world problems in organizations and systems. It's built entirely on optimization, and understanding the types of problems OR solves will help you recognize when optimization applies. Scheduling and routing are two classic OR problems: Scheduling determines when to perform tasks (perhaps on multiple machines or workers) to minimize completion time, cost, or lateness. For example, a manufacturing plant must decide which jobs to assign to which machines and in what order. Routing finds the best sequence of locations to visit. The famous traveling salesman problem is a routing problem: find the shortest route visiting all cities exactly once. While this sounds simple, it's computationally challenging for large numbers of cities. Inventory control balances holding costs (storing products is expensive) against ordering costs (frequent small orders are expensive, but holding large inventories ties up capital). The goal is to find the order quantity and timing that minimize total costs. Stochastic programming extends basic optimization to handle uncertainty. Real decisions often must be made before all information is known. For example, an airline must set flight schedules months in advance, but demand is uncertain. Stochastic programming models these multi-stage decisions where choices at one stage affect options available at future stages when random events become known. These problems are solved using large-scale optimization algorithms because they involve many variables and constraints representing different scenarios. Control Engineering CRITICALCOVEREDONEXAM Control engineering applies optimization at each moment of time to steer systems toward desired states. Model predictive control (MPC) is the dominant modern approach. Here's the key insight: instead of computing a control rule once (classical control), MPC solves an optimization problem repeatedly in real time. At each time step, the controller: Predicts future system behavior over some time horizon Solves an optimization problem to find the sequence of control actions that minimizes predicted error or cost Applies only the first control action from that sequence Repeats at the next time step This approach is powerful because it naturally handles constraints (actuators can only apply limited force, temperatures must stay within limits) and can explicitly account for future predicted disturbances. Chemical plants, refineries, autonomous vehicles, and power systems all use MPC. The challenge is that solving an optimization problem in real time requires fast algorithms and careful problem formulation. This is where the optimization theory you're studying becomes practically critical. Mechanics and Design CRITICALCOVEREDONEXAM Design optimization applies mathematical optimization to engineering component design. For example, aerospace engineers optimize aircraft structures to: Minimize weight (to reduce fuel consumption) Subject to constraints on strength, stiffness, and manufacturing feasibility Multidisciplinary design optimization (MDO) extends this to problems where multiple disciplines interact. An aircraft wing design affects both aerodynamics (wing shape determines lift and drag) and structures (shape also determines how stresses are distributed). These disciplines have competing objectives, so optimization must balance them. This requires careful problem formulation, sometimes solving many sub-optimization problems within an overall optimization framework. The image above shows an example of an optimized design surface—the blue surface represents the objective function (perhaps a performance measure or cost metric) that designers are optimizing over a design space. Civil Engineering and Water Resources POSSIBLYCOVEREDONEXAM Civil engineering problems often have a highly constrained, structured form that makes them ideal for linear and integer programming. <extrainfo> Water-resource allocation distributes limited water across agricultural, industrial, and municipal users to maximize benefit (or minimize shortage costs) while respecting physical constraints like available water supply and pipe capacities. Traffic management coordinates signal timing and route guidance to minimize congestion and travel time across networks. Construction scheduling sequences construction activities to minimize project duration and cost, respecting precedence constraints (you must build the foundation before the walls). These problems often involve yes/no decisions (build this project or not?) making them integer programming problems, which require specialized algorithms. </extrainfo> Machine Learning and Optimization CRITICALCOVEREDONEXAM This relationship deserves careful attention because it's central to modern data science and likely important for your exam. Machine learning fundamentally relies on optimization. When training a neural network, a decision tree, or any other machine learning model, you're solving an optimization problem: find the model parameters that minimize the gap between predicted outputs and actual outputs (the loss function). Specifically, training a machine learning model is a constrained optimization problem: $$\min{\theta} L(\theta) = \min{\theta} \frac{1}{n}\sum{i=1}^{n} \ell(f{\theta}(xi), yi)$$ where: $\theta$ represents the model parameters you're optimizing (like weights in a neural network) $L(\theta)$ is the total loss function (how wrong predictions are, on average) $\ell$ is the loss on a single example (how wrong one prediction is) $(xi, yi)$ are training examples The constraints might include regularization (preventing overfitting), bounds on parameter values, or sparsity constraints (preferring models that use fewer parameters). Why this matters for your studies: The optimization algorithms you're learning about—gradient descent, Newton's method, interior point methods—are the actual algorithms used to train machine learning models. When a data scientist says they're "training a model," they're solving an optimization problem using these techniques. Modern machine learning has also inspired new optimization algorithms and approaches. For instance, stochastic gradient descent (SGD), which updates parameters using gradients from small random batches of data rather than the entire dataset, was developed to handle the massive datasets in machine learning. This is now used far beyond ML. The reverse is also true: machine learning helps solve optimization problems. Machine learning can: Learn approximations of complex functions to speed up optimization Predict which algorithms will work best for a given problem Learn heuristics for finding good (if not optimal) solutions to hard problems quickly Computational Systems Biology POSSIBLYCOVEREDONEXAM <extrainfo> Optimization has become central to understanding and engineering biological systems: Model building and validation use optimization to find parameter values that make computational models match experimental observations. Optimal experimental design determines which experiments to run to most efficiently learn about a system, by optimizing information gain per unit cost. Metabolic engineering optimizes the genetic modifications of microorganisms (bacteria or yeast) to produce more of a desired compound (like insulin or biofuel). This involves optimizing genetic knockouts and modifications given constraints on cellular survival. Synthetic biology designs new biological systems with desired functions, formulated as optimization over genetic circuit designs. These applications are sophisticated but may be beyond the scope of introductory optimization courses. </extrainfo> Summary Optimization is everywhere. Every time you encounter a real-world decision problem involving trade-offs and constraints, there's likely an optimization formulation waiting to be discovered. The applications you've learned here—from portfolio selection to aircraft design to machine learning—show that optimization isn't just abstract mathematics. It's the language we use to solve practical problems in nearly every field.