Mixed Integer Goal Programming for Personalized Meal Optimization with User-Defined Serving Granularity
Francisco Aguilera Moreno · 2026
Read Mixed Integer Goal Programming for Personalized Meal Optimization with User-Defined Serving Granularity by Francisco Aguilera Moreno (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-13849.txt.
Introduction
Francisco Aguilera Moreno offers Mixed Integer Goal planning for custom meal tuning with user-defined serving sizes. His form uses integer counts for usable serving counts and goal planning gaps for soft nutrient targets, with inverse-target scaling evening multi-nutrient tuning. For meals with 15 or more foods, the integer solution matches the smooth optimum in every test case. Across 810 instances MIGP finds strictly better answers than goal planning with post-hoc rounding in 66 percent of cases, never worse.
Catalog record
| CREATOR | Francisco Aguilera Moreno |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79018 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.13849
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Mixed Integer Goal Programming for Personalized Meal Optimization with User-Defined Serving Granularity Francisco Aguilera Moreno Abstract Determining what to eat to satisfy nutritional requirements is one of the oldest optimization problems in operations research, yet existing formulations have two persistent limitations: continuous variables produce impractical fractional servings (1.7 eggs, 0.37 bananas), and hard nutrient constraints cause infeasibility when targets conflict. A systematic review of 56 diet optimization papers found that none combine integer programming with goal programming to address both issues. We propose Mixed Integer Goal Programming (MIGP) for personalized meal optimization. The formulation uses integer variables for practical serving counts and goal programming deviations for soft nutrient targets, with inverse-target normalization to balance multi-nutrient optimization. Per-food serving granularity allows natural units (one egg, one tablespoon of oil) without post-hoc rounding. We characterize the integrality gap in the goal programming context and identify a deviation absorption property: GP deviation variables buffer the cost of requiring integer servings, making the gap structurally smaller than in hard-constraint MIP. For meals with 15+ foods, the integer solution matches the continuous optimum in every benchmark instance. A computational evaluation across 810 instances (30 USDA foods, 9 configurations, 3 methods) shows MIGP finds strictly better solutions than GP with post-hoc rounding in 66% of cases (never worse) while maintaining 100% feasibility; hard-constraint IP achieves only 48%. Solve times stay under 100 ms for typical meal sizes using the open-source HiGHS solver. The implementation is available as an open-source Python module integrated into an interactive meal planning application. 1. Introduction 1.1. The Diet Problem Determining what to eat to satisfy nutritional requirements is one of the oldest optimization problems in operations research. Stigler [ 17 ] formulated the minimum-cost diet problem in 1945; Dantzig [ 7 ] solved it with the simplex method shortly after. Nearly 80 years later, the problem remains practically relevant: health tracking apps, meal prep services, and clinical nutrition programs all need to compose meals that meet individualized nutritional targets. The mathematical structure generalizes beyond nutrition. Resource-composition problems arise whenever discrete quantities of inputs must be combined to approximate a target composition vector: blending chemicals in manufacturing, constructing a portfolio from integer share lots, or loading vehicles to balance weight across dimensions. In each case, inputs come in discrete units, targets span multiple dimensions, and exact achievement is typically impossible. 1.2. Two Persistent Limitations Despite decades of refinement, diet optimization faces two limitations rooted in the standard LP formulation. Fractional servings. Continuous variables produce solutions like 1.7 eggs or 0.37 bananas, mathematically precise but useless for meal preparation. The literature has largely treated this as an implementation detail, rounding to integers after the fact [ 9 ] . But rounding is not neutral: it changes the nutritional profile without regard for optimality, and the rounded solution may violate the very constraints that motivated the optimization. Integer programming solves this directly. However, existing IP formulations for diet use binary selection variables (include/exclude a food) [ 3 ] rather than integer quantity variables (how many servings). The question of how much of each food (expressed in whole servings) has not been addressed with integer optimization. Hard constraints cause infeasibility. When nutrient requirements conflict or the food set is limited, hard-constraint formulations declare the problem infeasible and return nothing. Briend et al. [ 5 ] documented this in the WHO context, where iron requirements could only be met at 63% of recommendations. In interactive applications this is problematic: the system provides no guidance precisely when guidance is most needed. Goal programming [ 6 ] eliminates infeasibility by replacing hard constraints with soft targets. Deviation variables measure how far the solution falls from each target, and the objective minimizes total deviation. Gerdessen and de Vries [ 10 ] applied GP to diet optimization with good results, but their formulation uses continuous variables, inheriting the fractional serving problem. 1.3. The Gap A systematic review by Donkor et al. [ 8 ] examined 56 diet optimization papers and found that none combine integer programming with goal programming (detailed analysis in Section 2 ). Concurrently, AI/ML approaches have emerged for meal planning [ 13 , 1 , 21 ] , offering preference learning and generative capabilities but without formal optimality guarantees or transparent tradeoff reporting. These approaches are complementary to exact optimization. 1.4. Contributions We propose Mixed Integer Goal Programming (MIGP) for personalized meal optimization, filling the gap identified by Donkor et al. Our contributions: 1. Novel formulation. The first unified MIGP model for diet optimization, combining integer serving variables with GP deviation minimization. The model supports user-defined per-food serving granularity and inverse-target penalty normalization for balanced multi-nutrient optimization (Section 3 ). 2. Integrality analysis. We characterize the integrality gap in the GP context, identifying a deviation absorption property where deviation variables buffer the cost of integrality. With 15+ foods, the integer solution matches the continuous optimum. This property generalizes to any GP formulation with integer variables, not just diet optimization (Section 4 ). 3. Open-source implementation. A Python solver using HiGHS achieves sub-100 ms solve times for typical meal sizes, integrated into an interactive web application (Section 5 ). 4. Comparative evaluation. A benchmark across 810 instances shows MIGP finds strictly better solutions than GP+rounding in 66% of cases (never worse) while maintaining 100% feasibility; hard-constraint IP achieves only 48% (Section 6 ). 1.5. Paper Organization Section 2 reviews related work. Section 3 presents the MIGP formulation. Section 4 analyzes the integrality gap. Section 5 describes the implementation. Section 6 reports computational experiments. Section 7 discusses implications, limitations, and future directions. 2. Related Work 2.1. Classical Diet Optimization The diet problem is one of the earliest applications of mathematical programming. Stigler [ 17 ] formulated the problem of finding a minimum-cost diet satisfying nutritional requirements for 77 foods and 9 nutrients, solving it by manual enumeration. Dantzig [ 7 ] later applied the simplex method to this problem, establishing diet optimization as a canonical linear programming application. Subsequent work refined the LP approach for specific populations. Briend et al. [ 5 ] applied LP to complementary feeding for infants in a WHO context, demonstrating a fundamental limitation: when iron requirements could only be met at 63% of recommendations, the model declared infeasibility rather than providing a best-effort solution. Maillot et al. [ 14 ] developed individual diet modeling that translates nutrient recommendations into personalized food choices, and van Dooren et al. [ 19 , 20 ] extended the framework to incorporate sustainability criteria (cost and climate impact) alongside nutrition. Throughout this evolution, the mathematical structure remained the same: continuous LP with hard nutrient constraints. Two persistent limitations emerge. First, continuous variables produce impractical fractional servings (0.37 bananas, 1.7 eggs) that require manual rounding. Second, hard constraints cause infeasibility when nutrient requirements conflict or when the available food set cannot meet all targets simultaneously. 2.2. Goal Programming for Diet Goal programming, introduced by Charnes and Cooper [ 6 ] , replaces hard constraints with soft targets and explicit deviation variables, eliminating the infeasibility problem. Tamiz et al. [ 18 ] provide an overview of GP methodology, and Romero [ 15 ] formalizes the general achievement function structure. Gerdessen and de Vries [ 10 ] applied GP to diet optimization