Research

Innovative mathematical methods for tax system optimisation

TaxSolver is a framework developed by the Dutch Ministry of Finance that translates the tax system into mathematical formulations and searches for tax rates meeting a given set of policy objectives. Druga-Tache's research builds on it in two main directions. The first is scaling the framework up, so that it can work with national datasets rather than reduced samples. The second is widening its coverage of the tax code, particularly deductibles and the rules and exceptions that determine effective tax burdens in practice. Both directions draw on methods from mathematical optimisation, which may include problem decomposition, robust optimisation under uncertainty, and constraint learning.

The goal is to make TaxSolver a practical instrument for evidence-based policy design, able to show which reforms genuinely meet the objectives set for them by policymakers and what they mean for different households. The project is carried out in collaboration with the Dutch Ministry of Finance.

Optimising the location and capacity of healthcare facilities for time‑critical care in low‑ and middle‑income countries

Trang Luu works on mathematical methods to improve access to healthcare in low‑ and middle‑income countries. Their research focuses on designing networks of healthcare facilities so that patients reach treatment as quickly as possible. The goal is to minimise overall time‑to‑treatment, combining both travel time and waiting time, under realistic budget constraints.

To address this, Luu builds optimisation models that jointly determine where healthcare facilities should be located and how much capacity (beds, staff, equipment, etc.) each facility should have. These models are informed by datasets that combine population distributions, road networks, and detailed information on existing hospitals and their treatments, allowing system designs to be evaluated in a realistic way. While stroke care is the main application in her current projects, the modelling framework is designed to be broadly applicable, so that NGOs, governments and other partners can use it to improve geospatial accessibility and service capacity for a wide range of time‑critical healthcare services.

Optimising Large-Scale Geospatial Accessibility Problems in LMICs

This research focuses on geospatial accessibility to essential services in low‑ and middle‑income countries (LMICs). In the first line of work, I addressed large‑scale facility‑location problems through a collaboration with the Dutch Red Cross on water‑well placement in West Darfur, Sudan. The objective was to maximise the number of people within a 500 m straight‑line distance of a well. We developed a decomposition algorithm that can solve instances far beyond the reach of exact optimisation solvers. Additionally, she introduced a heat‑map‑based discretisation technique that generates candidate site regions rather than precise coordinates. This approach is especially suited to data‑scarce settings, allowing planners to refine locations later while preserving the continuous nature of the underlying problem.

The second research strand investigates flood‑resilient accessibility to critical services. Starting with a road‑selection model for Timor‑Leste that prioritised upgrades to preserve access to healthcare facilities, she later expanded the framework to incorporate multiple service types (healthcare, education, food) and to allocate upgrades preferentially to high‑poverty areas. The extended model integrates binary flood‑risk assessments with a risk‑weighted accessibility objective, enabling policymakers to evaluate trade‑offs between investment cost, service coverage, and equity under flood scenarios. Together, these contributions advance methodological tools for planning resilient infrastructure and equitable service delivery in resource‑limited environments.

Optimisation for Equitable Healthcare Access Planning

This PhD research examines how open geospatial data and optimisation methods can support more equitable healthcare access planning, particularly in low- and middle-income countries. By integrating datasets such as OpenStreetMap road networks, satellite-derived population data, and health facility registries, the research develops models that help policymakers identify underserved areas and determine optimal locations for healthcare infrastructure and services.

A key contribution is the development of the Public Infrastructure Service Access (PISA) ecosystem, a set of open-source tools that translate complex geospatial and optimisation methods into practical decision-support systems. Through real-world case studies in countries such as Timor-Leste, Vietnam, and Nepal, the research demonstrates how these tools can guide infrastructure planning, improve healthcare accessibility, and reveal critical gaps in underlying data systems.

Advancing proton therapy treatment plan optimisation under uncertainties

Various sources of uncertainty affect proton radiation therapy planning, including anatomical changes, patient setup errors, and proton range uncertainties. These uncertainties can significantly compromise treatment quality by reducing target coverage and limiting the sparing of surrounding healthy tissues.

Qiu's research focuses on minimising the negative impact of uncertainties on proton therapy through advanced optimisation models.

Domain-Driven Analytics: exploring the role of domain Intelligence in predictive and prescriptive analytics

This project explores when and how domain intelligence enhances analytical capabilities in complex, data-constrained environments. Domain intelligence may be explicit, such as formalised knowledge and operational constraints, or implicit, as seen in practitioners’ experience and decisions. However, domain intelligence remains inconsistently defined, incorporated, and evaluated in analytics, with most studies focusing on explicit forms. This project addresses these gaps by clarifying domain-related concepts, evaluating approaches to incorporating domain intelligence, and developing a framework to guide its selection and assessment.

The research investigates these questions through studies in humanitarian, healthcare, and logistics contexts. In partnership with the World Food Program (WFP), it examines how domain intelligence can improve predictions of regions at risk of child wasting. The Amazon last-mile routing study integrates tacit intelligence from drivers’ historical routes into an optimisation model. In contrast, the air-cargo congestion study applies domain-informed simulation to assess operational interventions. In collaboration with AMREF Health Africa, the thesis combines healthcare-demand estimation, routing optimisation, and operational constraints in a decision-support system for mobile-clinic deployment in Kenya. These studies demonstrate how various forms of domain intelligence can be incorporated into analytics based on the problem context.

Building on these insights, the project develops a methodological framework for selecting and assessing domain intelligence. It clarifies the role of domain intelligence in analytical modeling and examines how its incorporation affects model performance and solution relevance. For practitioners, it contributes by developing and evaluating analytics-based tools that translate domain intelligence into actionable solutions.

Mathematical optimisation under uncertainty with applications to hydrogen deployment in the Netherlands

This project is part of the broader “HyChain” research project, which investigates the deployment of hydrogen in the Netherlands. Hydrogen is a versatile energy carrier that can also be utilised as an industrial feedstock, and it may offer a viable path towards achieving a low-carbon economy. The Netherlands, as one of Europe’s largest hydrogen consumers and producers, is of particular interest due to its high demand, extensive natural gas infrastructure, and access to offshore wind resources. However, there are still many open questions regarding the deployment of hydrogen in the Netherlands.

Mathematical optimisation provides a systematic framework for answering such questions by modelling variables, objectives, and constraints in a transparent and quantitative manner. However, its application to real-world problems is complicated by uncertainty. This project focuses on two major types of uncertainty: parameter uncertainty, which arises when input data (such as costs or demand) cannot be predicted with certainty, and model uncertainty, which occurs when the mathematical formulation does not fully represent the real-world system.