------------------------------------------------------------------------------------------------------------------- 1. Title of the Paper Complex marine ecological response during the Eocene-Oligocene revealed by global foraminiferal record ------------------------------------------------------------------------------------------------------------------- 2. Author Information Zhengbo Lu; State Key Laboratory of Critical Earth Material Cycling and Mineral Deposits, School of Earth Sciences and Engineering, Nanjing University; Frontiers Science Center for Critical Earth Material Cycling, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; zb_lu@foxmail.com Ke Xue; National Key Laboratory for Novel Software Technology, Nanjing University; School of Artificial Intelligence, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; xuek@lamda.nju.edu.cn Yiying Deng; School of Resources and Environmental Engineering, Hefei University of Technology; 193 Tunxi Road, Hefei, Anhui, China, 230009; yydeng12138@foxmail.com Junxuan Fan; State Key Laboratory of Critical Earth Material Cycling and Mineral Deposits, School of Earth Sciences and Engineering, Nanjing University; Frontiers Science Center for Critical Earth Material Cycling, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; jxfan@nju.edu.cn Peiyue Fang; School of Earth and Planetary Sciences, East China University of Technology; 418 Guanglan Avenue, Nanchang, Jiangxi, China, 330013; py_fang7@ecut.edu.cn Bridget S. Wade; Department of Earth Sciences, University College London; Gower Street, London, United Kingdom, WC1E 6BT; b.wade@ucl.ac.uk Laia Alegret; Department of Earth Sciences, University of Zaragoza; Pedro Cerbuna 12, Zaragoza, Spain, 50009; laia@unizar.es Michael J. Benton; School of Earth Sciences, Wills Memorial Building, University of Bristol; Queen's Road, Bristol, United Kingdom, BS8 1RJ; mike.benton@bristol.ac.uk Yuchang Wu; National Key Laboratory for Novel Software Technology, Nanjing University; School of Artificial Intelligence, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; wuyc@lamda.nju.edu.cn Chao Qian; National Key Laboratory for Novel Software Technology, Nanjing University; School of Artificial Intelligence, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; qianc@nju.edu.cn Xudong Hou; State Key Laboratory of Critical Earth Material Cycling and Mineral Deposits, School of Earth Sciences and Engineering, Nanjing University; Frontiers Science Center for Critical Earth Material Cycling, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023 Yukun Shi; State Key Laboratory of Critical Earth Material Cycling and Mineral Deposits, School of Earth Sciences and Engineering, Nanjing University; Frontiers Science Center for Critical Earth Material Cycling, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; ykshi@nju.edu.cn Peter M. Sadler; Department of Earth Sciences, University of California, Riverside; 900 University Avenue, Riverside, CA, USA, 92521; peter.sadler@ucr.edu Huiqing Xu; State Key Laboratory of Critical Earth Material Cycling and Mineral Deposits, School of Earth Sciences and Engineering, Nanjing University; Frontiers Science Center for Critical Earth Material Cycling, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023 Zhi-Hua Zhou; National Key Laboratory for Novel Software Technology, Nanjing University; School of Artificial Intelligence, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; zhouzh@nju.edu.cn Shuzhong Shen; State Key Laboratory of Critical Earth Material Cycling and Mineral Deposits, School of Earth Sciences and Engineering, Nanjing University; Frontiers Science Center for Critical Earth Material Cycling, Nanjing University; 163 Xianlin Avenue, Qixia District, Nanjing, Jiangsu, China, 210023; szshen@nju.edu.cn ------------------------------------------------------------------------------------------------------------------- 3. the name of the corresponding author (i.e., the author to whom notices will be sent concerning the competition); Chao Qian qianc@nju.edu.cn ------------------------------------------------------------------------------------------------------------------- 4. the abstract of the paper(s); The Eocene-Oligocene transition was the crucial turning point when Earth's climate shifted to its current icehouse state. Understanding how the marine biosphere responded during this transition is not well-constrained, appearing as a simple extinction pulse in low temporal resolution global compendia. Here we design an artificial-intelligence-inspired metaheuristics algorithm to construct a high-resolution global species richness history across the Eocene-Oligocene transition for the rich foraminifera fossil record with an imputed ~29,000-year resolution. The revealed diversity dynamics are complex and differ for each foraminiferal group with distinct ecology. Planktonic and shallow-water larger benthic foraminifera show steady diversity levels in the early phases of the transition in the latest Eocene after a long-term reduction, while the deeper-water small benthic foraminifera radiate notably and then decline over the same interval. In the earliest Oligocene, the planktonic and larger foraminifera suffer major species losses coincident with the first continental-scale ice sheet formed on Antarctica, while small benthic foraminifera diversity holds steady, followed by an accelerating lowering as the early Oligocene proceeds. These findings reveal complicated and ecologically differentiated environment-life processes, indicating the importance of high-resolution temporal data for dissecting out ecological responses to major environmental changes. ------------------------------------------------------------------------------------------------------------------- 5. a list containing one or more of the four letters (A, B, C, or D) that correspond to the criteria (see above) that the author claims that the work satisfies; B, C ------------------------------------------------------------------------------------------------------------------- 6. a statement stating why the result satisfies the criteria that the contestant claims (see examples of statements of human-competitiveness as a guide to aid in constructing this part of the submission); Deep-time life evolution is a fundamental problem in natural science. In its 125th anniversary issue, Science highlighted "What Determines Species Diversity?" as one of the major questions facing science [1]. Answering this question requires reconstructing biodiversity dynamics across millions to billions of years and testing how biodiversity was coupled to environmental drivers such as climate, ocean chemistry, and sea level. Computationally, the inputs are many local fossil sections, each giving only a partial order of fossil events. The output is one global event sequence. Each event is the first or last appearance of a species, so a candidate solution is a permutation of all these events. The task can therefore be formulated as a large-scale constrained black-box permutation optimization problem. A feasible sequence must satisfy basic fossil-order constraints, including first-before-last ordering within each species, coexistence constraints among species found together, and local ordering constraints within each section. The objective is to minimize the penalty, which measures how much the proposed global sequence disagrees with the local fossil records. Estimating this disagreement requires a complex computational procedure and is usually treated as a black-box evaluation. Multiple papers published in top-tier journals have demonstrated the importance of this problem. For example, a 2020 Science paper reconstructed a high-resolution biodiversity history for Cambrian to Early Triassic marine invertebrates [2] and was selected as one of the Top Ten Scientific Advances in China; and a 2024 Science paper quantified the global biodiversity of Proterozoic eukaryotes [3]. Such studies show that high-resolution deep-time biodiversity reconstruction is a major route to answering basic questions about life evolution. The key computational method behind these studies is the CONOP family of solvers, which has been the main framework in deep-time life-evolution studies. Its core search strategy is based on simulated annealing. The state-of-the-art solver CONOP.SAGA uses mutation-like operations and large-scale parallel search, enabling it to solve larger datasets, but it still faces a serious scaling challenge. For example, the 2020 Science reconstruction performed with CONOP.SAGA required about 7 million CPU hours on Tianhe-2, a world-class supercomputing platform built in China that once held the No. 1 position on the TOP500 list of supercomputing systems. Such costs show that even CONOP.SAGA cannot routinely handle richer fossil datasets or broader taxonomic analyses. As a result, much of the growing fossil record cannot yet be analyzed at comparable resolution, limiting progress across the field. Our submission describes CONOP.EA, an evolutionary optimization framework designed to overcome this bottleneck. It improves the previous CONOP.SAGA workflow by moving from a mainly mutation-driven, single-trajectory search to a population-based evolutionary search. CONOP.EA constructs a diverse initial population using stratigraphic priors and maintains a diversity-preserving archive of high-quality but structurally different candidate permutations. It then employs two mechanisms that go beyond CONOP.SAGA's memoryless search behavior. First, a learning-based self-adaptive mutation operator learns during the run which local fossil-event edits reduce the penalty, stores them, and increasingly favors similar edits later in the search, so mutation is guided by experience on the same problem instance rather than by a fixed perturbation rule. Second, a diversity-based crossover operator draws from the archive to recombine complementary event-ordering blocks from different search trajectories, rather than merely perturbing one parent sequence. Taken together, these mechanisms make CONOP.EA a more complete and efficient population-based EA framework that extends the established CONOP family workflow. These tailored EA designs make CONOP.EA highly efficient in practice. On the foraminiferal correlation dataset used in the study (approximately 40,000 fossil occurrences, 1269 species, and 161 stratigraphic records), using the same hardware and comparison parameters reported in the paper, CONOP.EA reaches a CONOP.SAGA-level objective value about 23 times faster. Under the same trial setting, it also achieves better final solutions, with an average penalty approximately 538.5 lower. These lower-penalty solutions produce global event orders that fit the fossil records more closely and thereby enable stronger scientific discovery. This combination of speed and solution quality matters because larger fossil datasets can lead to new scientific conclusions only if the optimization method can handle them in a practical amount of time. This work is a representative example of AI for Science: it shows that efficient EAs can lead to scientific discovery. CONOP.EA is not tied to foraminifera or to the Eocene-Oligocene transition; it works on the general task of turning many incomplete local event sequences into one global order. This makes it useful for many deep-time biodiversity studies and helps promote EA as a practical tool for scientific discovery. For example, in our paper, CONOP.EA integrated approximately 40,000 fossil occurrence records from 161 stratigraphic records, produced the first global high-resolution foraminiferal diversity curve for this interval, and revealed the mechanisms driving those diversity changes. Below, we describe how our results satisfy the criteria we claim: (B) Published in a peer-reviewed scientific journal as a new scientific result independent of evolutionary computation. This entry provides a very important application for evolutionary computation. It is a representative AI for Science example: an efficient EA not only improves an optimization score, but also enables a scientific discovery. In our case, CONOP.EA turns scattered fossil records into a high-resolution global timeline. The paper was published in Nature Communications as a new result in paleobiology and Earth history, not merely as a computation paper. Using CONOP.EA, the study reconstructed the first global high-resolution foraminiferal diversity curve for the Eocene-Oligocene interval, with 962 temporal levels and an imputed resolution of approximately 29.11 Kyr. This curve made it possible to reveal the mechanisms driving foraminiferal diversity change, including the distinct responses of planktonic, larger benthic, and small benthic foraminifera. The peer-reviewed publication therefore establishes that the result has independent scientific value outside the evolutionary-computation community. (C) Better than the most recent human-created solution to a long-standing or previously unsolved problem of indisputable difficulty in its field: The hard problem is global fossil correlation. Given many incomplete local fossil-event sequences, the task is to infer one global event order that best fits them all. This is a long-standing large-scale constrained permutation problem. The previous state-of-the-art solution family was CONOP, especially CONOP.SAGA. It used simulated annealing, mutation-like operations, and large-scale parallel search to build better event sequences, making high-resolution fossil correlation possible. However, it remained expensive and limited by local optima. The scale of earlier work shows the difficulty: the 2020 Science reconstruction required about 7 million CPU hours on Tianhe-2, a world-class supercomputing platform built in China that once held the No. 1 position on the TOP500 list of supercomputing systems. This level of cost makes it difficult to use richer fossil datasets or run broader analyses. CONOP.EA improves this baseline by replacing the mainly mutation-driven, single-trajectory workflow with a more complete population-based EA search. It uses domain-informed initialization, learning-based mutation, diversity-based crossover, and penalty-based selection. On the present dataset, CONOP.EA produced sequences whose average penalty, that is, average objective value, was about 538.5 lower than CONOP.SAGA under the same comparison parameters. It also reduced average wall-clock time from 13 h:09 m to 33 m on the same hardware, a roughly 23-fold speedup. Thus, under the same trial setting, it finds better solutions; to reach a CONOP.SAGA-level solution, it needs much less time. The improvement is therefore both computational and scientific: lower penalty gives a better global event order, shorter runtime makes larger fossil datasets practical, and the resulting reconstruction reveals biodiversity changes that lower-resolution approaches could not resolve. References: [1] E. Pennisi. What determines species diversity? Science, 309(5731):90, 2005. [2] J. X. Fan et al. A high-resolution summary of Cambrian to Early Triassic marine invertebrate biodiversity. Science, 367:272-277, 2020. [3] Q. Tang et al. Quantifying the global biodiversity of Proterozoic eukaryotes. Science, 386:eadm9137, 2024. [4] W. G. Kemple, P. M. Sadler, and D. J. Strauss. Extending graphic correlation to many dimensions: stratigraphic correlation as constrained optimization. In Graphic Correlation, pages 65-82. SEPM Society for Sedimentary Geology, 1995. [5] P. M. Sadler and R. A. Cooper. Best-fit intervals and consensus sequences: comparison of the resolving power of traditional biostratigraphy and computer-assisted correlation. Topics in Geobiology, 21:50-91, 2003. [6] P. M. Sadler. Constrained optimization approaches to the paleobiologic correlation and seriation problems: part one, a user's guide to the CONOP program family. Department of Earth Sciences, University of California, Riverside, CA 92521, 1998-2006, 2013. ------------------------------------------------------------------------------------------------------------------- 7. a full citation of the paper (that is, author names; title, publication date; name of journal, conference, or book in which article appeared; name of editors, if applicable, of the journal or edited book; publisher name; publisher city; page numbers, if applicable); Zhengbo Lu, Ke Xue, Yiying Deng, Junxuan Fan, Peiyue Fang, Bridget S. Wade, Laia Alegret, Michael J. Benton, Yuchang Wu, Chao Qian, Xudong Hou, Yukun Shi, Peter M. Sadler, Huiqing Xu, Zhi-Hua Zhou, and Shuzhong Shen. Complex marine ecological response during the Eocene-Oligocene revealed by global foraminiferal record. Nature Communications, volume 17, article number 3954, published online 14 March 2026. Springer Nature, London, United Kingdom. https://doi.org/10.1038/s41467-026-70541-w ------------------------------------------------------------------------------------------------------------------- 8. a statement either that "any prize money, if any, is to be divided equally among the co-authors" OR a specific percentage breakdown as to how the prize money, if any, is to be divided among the co-authors; Any prize money, if any, is to be paid to the corresponding author, Chao Qian, and will be distributed internally among the co-authors by the authors. ------------------------------------------------------------------------------------------------------------------- 9. a statement stating why the authors expect that their entry would be the "best"; We expect this entry to be a strong candidate for the "best" Humies award because it combines four strengths: an important scientific problem, a hard EA optimization task, a clear improvement over the previous human-created solution, and a peer-reviewed scientific discovery. The scientific problem is important. Scientists want to know how species diversity changes over geological time and how these changes relate to climate and environment. Computationally, this requires merging many partial local fossil orderings into one globally consistent permutation. A candidate solution is a permutation of fossil first- and last-appearance events. The objective is to minimize the penalty that measures disagreement between the proposed permutation and the fossil records. The problem also has many constraints, including first-before-last ordering within each species, coexistence constraints, and local ordering constraints within fossil sections. This is not a toy optimization problem. The CONOP family of solvers has been the state-of-the-art framework in deep-time life-evolution studies. Its core search strategy is based on simulated annealing, and CONOP.SAGA further added mutation-like operations and large-scale parallel search. These methods made high-resolution fossil correlation possible, but they face a serious scaling challenge. The 2020 Science reconstruction required about 7 million CPU hours on Tianhe-2, a world-class supercomputing platform built in China that once held the No. 1 position on the TOP500 list of supercomputing systems. Such costs make it difficult to use richer fossil datasets and analyze broader taxonomic groups. CONOP.EA directly addresses this bottleneck. It replaces the mainly mutation-driven, single-trajectory CONOP.SAGA workflow with a more complete population-based EA search. It uses domain-informed initialization, learning-based mutation, diversity-based crossover, and penalty-based selection. The reported results are concrete: under the same comparison parameters, CONOP.EA reduced the average objective value by about 538.5 penalties compared with CONOP.SAGA. It also reduced average wall-clock time from 13 h:09 m to 33 m on the same hardware, making it about 23 times faster. Thus, under a fixed trial setting, it obtains better solutions, and for a CONOP.SAGA-level target it requires far less time. The method is broadly useful rather than tailored to a single fossil group. CONOP.EA solves a general data-integration problem: many incomplete local event sequences must be assembled into one globally consistent order. This structure appears in many fossil datasets. The same framework can support biodiversity reconstruction, extinction-rate estimation, event detection, and tests of environment-life relationships in other deep-time studies. The submitted paper demonstrates the scientific discovery and the broader AI for Science significance. CONOP.EA integrated approximately 40,000 fossil occurrence records, produced the first global high-resolution foraminiferal diversity curve for this interval, and revealed the mechanisms driving its biodiversity changes. In this sense, the entry shows that efficient EAs can become practical discovery tools in a mature natural-science field. This combination of method innovation, objective performance gain, scalability, broad applicability, and independent scientific impact is why the authors expect the entry to be among the strongest candidates for the "best" award. ------------------------------------------------------------------------------------------------------------------- 10. An indication of the general type of genetic or evolutionary computation used, such as GA (genetic algorithms), GP (genetic programming), ES (evolution strategies), EP (evolutionary programming), LCS (learning classifier systems), GI (genetic improvement), GE (grammatical evolution), GEP (gene expression programming), DE (differential evolution), etc. EA (evolutionary algorithm), implemented as a population-based constrained optimization algorithm using mutation, recombination, and natural selection. It is closely related to GA-style evolutionary search over constrained sequence representations. ------------------------------------------------------------------------------------------------------------------- 11. The date of publication of each paper. If the date of publication is not on or before the deadline for submission, but instead, the paper has been unconditionally accepted for publication and is "in press" by the deadline for this competition, the entry must include a copy of the documentation establishing that the paper meets the "in press" requirement. Published online on 14 March 2026 in Nature Communications. -------------------------------------------------------------------------------------------------------------------