We propose a worldview to facilitate the discussion of how Monte Carlo simulation experiments are conducted in a digital twin context. We use the term Monte Carlo digital twin to refer to a digital twin in which a Monte Carlo simulation model mimics the physical twin. The worldview we propose encompasses Nelson and Schmeiser's classical worldview for simulation experiments and adds key components of a Monte Carlo digital twin including a predictor, a decision maker, and bidirectional interaction between the Monte Carlo digital twin and the physical twin. We delineate the proposed worldview components in three examples.
The performance of a CPU-only implementation of the restarted GMRES algorithm with direct randomized-SVD -based preconditioning has been analyzed. The method has been tested on a set of sparse and dense matrices exhibiting varying spectral properties and compared to the ILU(0) -based preconditioning. This comparison aims to assess the advantages and drawbacks of both approaches. The trade-off between iteration-to-solution and time-to-solution metrics is discussed, demonstrating that the proposed method achieves an improved convergence rate in terms of iterations. Additionally, the method's competitiveness with respect to both metrics is discussed within the context of several relevant scenarios, particularly those where GMRES-based simulation techniques are applicable.
We consider a short-term demand forecasting problem for semiconductor supply chains. In addition to observed demand quantities, order entry information is available. We compute a combinational forecast based on an exponential smoothing technique, a long short-term memory network, and the order entry information. The weights for the different forecast sources and parameters for exponential smoothing are computed using a genetic algorithm. Computational experiments based on a rich data set from a semiconductor manufacturer are conducted. The results demonstrate that the best forecast performance is obtained if all the different forecasts are combined.
This paper investigates the sample efficient exploration policy for asynchronous Q-learning from the perspective of uncertainty quantification. Although algorithms like $\varepsilon$-greedy can balance exploration and exploitation, their performances heavily depend on hyperparameter selection, and a systematic approach to designing exploration policies remains an open question. Inspired by contextual Ranking and Selection problems, we focus on optimizing the probability of correctly selecting optimal actions (PCS) rather than merely estimating Q-values accurately. We establish a novel central limit theorem for asynchronous Q-iterations, enabling the development of two strategies: (1) an optimization-based policy that seeks an optimal computing budget allocation and (2) a parameter-based policy that selects from a parametrized family of policies. Specifically, we propose minimizing an asymptotic proxy of Q-value uncertainty with regularization. Experimental results on benchmark problems, including River Swim and Machine Replacement, demonstrate that the proposed policies can effectively identify sample-efficient exploration strategies.
The advancement of Artificial Intelligence (AI) has accelerated the transformation of simulation from a tool for analysis and design into a dynamic partner for decision-making and operation. As AI systems become more capable of learning, reasoning, and adapting, simulation is evolving into an intelligent, autonomous, and predictive framework for exploring complex futures. This paper brings together future-oriented perspectives from simulation scientists and AI experts to discuss the current AI-simulation integration and both near-term and long-term outlooks for innovation, collaboration, and disruption.
Behavioral models of component-based dynamical systems are integral to building useful simulations. Toward this goal, approaches enabled by Large Language Models (LLMs) have been proposed and developed to generate grammar-based models for Discrete Event System Specification (DEVS). This paper introduces PDEVS-LLM, an agentic framework to assist in developing Parallel DEVS (PDEVS) models. It proposes using LLMs with statecharts to generate behaviors for parallel atomic models. Enabled with PDEVS concepts, plausible facts from the whole description of a system are extracted. The PDEVS-LLM is equipped with grammars for the PDEVS statecharts and hierarchical coupled model. LLM agents assist modelers in (re-)generating atomic models with conversation histories. Examples are developed to demonstrate the capabilities and limitations of LLMs for generative PDEVS models.
Simulating realistic populations for strategic influence and social-cyber modeling requires agents that are demographically grounded, emotionally expressive, and contextually coherent. Existing agent-based models often fail to capture the psychological and ideological diversity found in real-world populations. This paper introduces AURORA, a Retrieval-Augmented Generation (RAG)-enhanced framework that leverages large language models (LLMs), semantic vector search, and salience-aware topic modeling to construct synthetic communities and personas. We compare two opinion modeling strategies and evaluate three LLMs—gemini-2.0-flash, deepseek-chat, and gpt-4o-mini—in generating emotionally and ideologically varied agents. Results show that community-guided strategies improve meso-level opinion realism, and LLM selection significantly affects persona traits and emotions. These findings demonstrate that principled LLM integration and salience-aware modeling can enhance the realism and strategic utility of synthetic populations for simulating narrative diffusion, belief change, and social response in complex information environments.
Orthopedic services are characterized by high patient volumes, long elective waits, unpredictable emergency demand, and close coupling with other hospital processes. These present significant challenges in meeting operational targets while maintaining quality of care. In healthcare, simulation has been widely used for addressing such challenges. Topic modeling is used to identify and analyze academic papers using operational-level simulation for orthopedic service delivery. We analyzed 37 papers over twenty years, combining a structured analysis with topic modeling to categorize and map applications. Despite widespread recognition of its potential, simulation remains underutilized in orthopedics, with fragmented application and limited real-world implementation. Recent trends indicate a shift toward system-wide approaches that better align with operational realities and stakeholder needs. Future research should aim to bridge methodological innovation with collaboration and practical application, such as hybrid and real-time simulation approaches focusing on stakeholder needs, and integrating relevant operational performance metrics.
This paper addresses the photolithography process scheduling problem, a critical bottleneck in both display and semiconductor production. In display manufacturing, as the number of deposited layers increases and reentrant operations become more frequent, the complexity of scheduling processes has significantly increased. Additionally, growing market demand for diverse product types underscores the critical need for efficient scheduling to enhance operational efficiency and meet due dates. To address these challenges, we propose a novel graph-based reinforcement learning framework that dynamically schedules photolithography operations in real time, explicitly considering mask locations, machine statuses, and associated transfer times. Through numerical experiments, we demonstrate that our method achieves consistent and robust performance across various scenarios, making it a practical solution for real-world manufacturing systems.
The challenge of optimizing a smooth convex functional over probability spaces is highly relevant in experimental design, emergency response, variations of the problem of moments, etc. A viable and provably efficient solver is the fully-corrective Frank-Wolfe (FCFW) algorithm. We propose an FCFW recursion that rigorously handles the zero-order setting, where the derivative of the objective is known to exist, but only the objective is observable. Central to our proposal is an estimator for the objective's influence function, which gives, roughly speaking, the directional derivative of the objective function in the direction of point mass probability distributions, constructed via a combination of Monte Carlo, and a projection onto the orthonormal expansion of an $L_{2}$ function on a compact set. A bias-variance analysis of the influence function estimator guides step size and Monte Carlo sample size choice, and helps characterize the recursive rate behavior on smooth non-convex problems.
This study presents an agent-based simulation that examines how pre-attack misinformation amplifies the effectiveness of spearphishing campaigns within organizations. A virtual organization of 235 end user agents is modeled, each assigned unique human factors such as Big Five personality traits, fatigue, and job performance, derived from empirical data. Misinformation is disseminated through Facebook, where agents determine whether to believe and spread false content using regression models from prior psychological studies. When agents believe misinformation, their psychological and organizational states degrade to simulate a worst-case scenario. These changes increase susceptibility to phishing emails by impairing security-related decision-making. Informal relationship networks are constructed based on extraversion scores, and network density is varied to analyze its effect on misinformation spread. The results demonstrate that misinformation significantly amplifies organizational vulnerability by weakening individual and collective cybersecurity-relevant decision-making, emphasizing the critical need to account for human cognitive factors in future cybersecurity strategies.
This tutorial gives an overview about the state-of-the-art of Digital Twins (DTs). We discuss what they are, what purposes they may fulfill, how they differ from related technologies, and how they may work internally. Application-wise, we focus on Digital Twins of production and logistics systems, but the majority of our explanations should also be valid beyond these domains. Finally, we discuss open issues and potential directions of future research.
Automatically generating and executing simulation experiments promises to make running simulation studies more efficient, less error-prone, and easier to document and replicate. However, during experiment generation, background knowledge is required regarding which experiments using which inputs and outputs are useful to the modeler. Therefore, we conducted an interview study to identify what types of experiments modelers perform during simulation studies. From the interview results, we defined four general goals for simulation experiments: exploration, confirmation, answering the research question, and presentation. Based on the goals, we outline and demonstrate an approach for automatically generating experiments by utilizing an explicit and thoroughly detailed conceptual model.
Liver transplantation is the second most common transplant procedure in the United States and the only curative treatment for patients with end-stage liver disease, which is one of the leading causes of death nationwide. The United Network for Organ Sharing operates the national liver transplant waiting list and allocates organs under a complex priority system based on medical urgency, geography, and waiting time. Healthcare providers accept or refuse liver offers based on transplant candidates' medical needs and donor quality, among other factors. We develop a simulation environment to assess current acceptance practices based on a Markov reward process. Our simulation framework models organ arrivals and patients' health progression as continuous-time processes and mimics how decisions are made in practice using a randomized policy. Based on our simulation framework, we provide insights and identify areas for enhancing patient management and liver offer acceptance.
Simulation-based methods for real-time decision-making have attracted increasing research attention, and such problems are usually formulated as the simulation optimization problem with covariates. There are generally two methods to address this problem. The first builds a relationship between the objective function and the covariates, allowing a solver to quickly find the solution. The second directly builds a relationship between the optimal solution and the covariates. In this paper, we focus on continuous simulation optimization with covariates and investigate neural network-based implementations of both methods. We demonstrate that even when the objective function is continuous, the optimal solution may exhibit discontinuities with respect to the covariates, limiting the applicability of the optimal solution method. In contrast, the objective function method remains effective and broadly applicable. We further establish stability conditions under which the optimal solution method is valid. Numerical experiments are conducted to support our theoretical findings.
This research presents a novel approach to spare parts inventory management by integrating real-time machine health data with dynamic, state-dependent inventory policies. Traditional static models overlook the evolving conditions of industrial machinery. Leveraging advanced digital technologies, such as those pioneered by Augury, our framework dynamically adjusts inventory levels, reducing costs and improving service. Using Markov chain modeling, simulation, and industry collaboration, we demonstrate up to 29 % cost savings with state-dependent policies over static base-stock models. Sensitivity analysis confirms the robustness of these strategies.
This study explores strategies for robust optimization of queueing performance in the presence of input model uncertainty. Ambiguity sets for Distributionally Robust Optimization (DRO) based on Wasserstein distance is preferred for general DRO settings where the computation of performance given the distribution form is straightforward. For complex queueing systems, distributions with large Wasserstein distance (from the nominal distributions) do not necessarily provide extreme objective values. Thus, the calculation of performance extremes must be done via an inner level of maximization, making DRO a compute-intensive activity. We explore approximations for queue waiting time in a number of settings and show how they can provide low-cost guidance on extreme objective values, allowing for more rapid DRO. Approximations are provided for single- and multi-server queues and queueing networks, each illustrated with an example. We also show in settings with small number of solution alternatives that these approximations lead to robust solutions.
Decision Support Systems (DSS) are a crucial component in production logistics, aiding companies in solving complex decision problems with multiple influences. This publication provides a structured review of the literature on the application of DSS in production logistics, focusing on methods for decision support, such as simulation and Artificial Intelligence (AI). The analysis considers scientific publications from 2015 to 2024, including industry use cases. Data analysis of categorizations of DSS is used. The findings highlight trends and limitations of current DSS application cases from the literature. Optimization methods, particularly heuristic and metaheuristic, are the most commonly employed decision support methods, followed by simulation. Despite the increased interest in AI technologies, their role in DSS for production logistics remains secondary. Like simulation methods, AI technologies are highly relevant when combined with optimization methods. The study provides a foundation for future research and practical advancements in decision support for manufacturing environments.
This paper applies an agent-based simulation model to examine the feasibility of battery electric trucks (BETs) in intermodal freight transportation, focusing on the Memphis hub network. Two infrastructure deployment stages, depot charging only and depot plus destination charging, are modeled and simulated using AnyLogic platform to study truck utilization patterns. Real-world manufacturing sites are chosen, and the trucks are routed along roadways using a Geographic Information System (GIS) map. Battery charge levels and charging infrastructure are modeled under both scenarios. Four electric truck models from various manufacturers including Tesla Semi, Nikola Tre, Volvo VNR, and Freightliner eCascadia are compared in terms of performance and utilization. Results showed that battery electric trucks are a feasible solution for intermodal trucking operations and transporting goods from manufacturers to destinations. This comparison also highlights effects of changing shifts and adding opportunity charging at destinations on truck utilization under different battery efficiencies and capacities.
In this work, we introduce DES-Gymnax, a novel high-performance discrete-event simulator implemented in JAX. By leveraging the just-in-time compilation, automatic vectorization, and GPU acceleration capabilities of JAX, DES-Gymnax can achieve 10x to 100x times performance improvement over traditional Python-based discrete-event simulators like Salabim. The proposed DES-Gymnax can feature a Gym-like API that facilitates seamless integration with reinforcement learning algorithms, addressing a critical gap between simulation engines and AI techniques. DES-Gymnax is validated on three benchmark models, i.e., an M/M/1 queue, a multi-server model, and a tandem queue model. Experimental results demonstrate that DES-Gymnax maintains simulation accuracy while significantly reducing execution time, enabling efficient large-scale sampling crucial for reinforcement learning applications in operations research areas. The open-source code is available in the DES-Gymnax repository (Yun, Jun, and Xiangfeng 2025).
