Scientify

Mathematical Modeling / CUMCM Paper Guide

How to Write a CUMCM Paper: Format Checklist, Abstract Writing, and Template Choice

In 2025 the China Undergraduate Mathematical Contest in Modeling (CUMCM) drew 68,311 teams from 1,837 institutions, and 294 undergraduate teams won a national first prize, about 0.48%. A paper first has to survive the format rules, then show its results within a short initial review. This guide walks through the official format specification item by item, how to write the abstract, what each body section must deliver, the violations that cost award eligibility, and official entry points for templates and past winning papers.

Short answer

A CUMCM paper follows a fixed order: declaration, identification page, a one-page abstract, the body, and appendices. The abstract states the model, key numerical results, and conclusion for every subproblem, and the body covers restatement, assumptions, modeling and solution, sensitivity analysis, and references in turn. Before submitting, check the paper against that year's official format specification, with special attention to two disqualifying rules: the appendix must contain complete, runnable source code, and no university or personal name may appear anywhere.

Review

Who reads your paper, and for how long

CUMCM uses two review levels. Regional committees review first, and regional awards generally go to no more than one third of teams. Strong regional papers then move on to national review, as described in the official national review procedures. Every paper is read by at least two groups of judges, and the first group reads fastest.

The competition charter sums up the judging basis in one sentence: reasonable assumptions, creative modeling, correct results, and clear writing. How those four criteria show up in winning papers is covered on our page about award-winning modeling papers, so this guide does not repeat them. It focuses on writing a paper that can be checked against them.

Abstract

Four abstract elements: state the model, the numbers, and the conclusion

The official specification gives the abstract one full page, including the title and keywords. Use that page to answer four questions in order, and answer every one.

  1. 01

    What problem did you solve?

    Which subproblem, what input data, and what output was requested. One sentence is enough; do not retell the background.

  2. 02

    Which model and method did you use?

    Name the model type and the solution algorithm. Leave the derivation for the body.

  3. 03

    What numerical results did you get?

    Put the key number and unit for each subproblem in the abstract instead of writing that the results were satisfactory.

  4. 04

    What do the results mean?

    State the conclusion behind the numbers in one sentence, and say whether sensitivity analysis shows it holds.

An abstract with nothing to check

"For Problem 1, we developed a reasonable mathematical model and solved it with an effective algorithm, obtaining satisfactory results. For Problem 2, we improved the previous model, and the results have practical significance." It names a method but gives no concrete problem, no numbers, and no conclusion.

An abstract a judge can verify

"For Problem 1, we formulated an integer program with scheduling order as the decision variable. Branch and bound gave an optimal total cost of RMB 18.42 million, 12.6% below the greedy baseline. Sensitivity analysis showed the optimal structure is unchanged when unit freight cost rises 10%." Problem, model, algorithm, numbers, conclusion, and robustness are all there. (Invented example; the numbers are illustrative only.)

Body

Section-by-section pass criteria and common problems

Each body section has a distinct job. Use the table to check them one at a time.

SectionPass criterionCommon problem
Problem restatementReorganize the task in your own words: the data supplied, the constraints, and the output each subproblem requires.Copying the prompt verbatim. The judge has just read it, so a copy signals that the section can be skipped.
AssumptionsEach assumption maps to a specific simplification, and you can point to where it takes effect in the model.Boilerplate such as "the data are reliable" or "other factors are ignored" that could be deleted without changing the model.
Modeling and solutionEvery figure and table is numbered and titled and cited explicitly in the text, as in "Figure 3 shows". Every symbol is defined at first use.Figures pasted in without being cited, leaving judges unsure where to look; long runs of equations with no prose between them.
Sensitivity analysisPerturb at least one key parameter, report the resulting numbers, and state whether the conclusion changes.A single line saying the model was tested and is stable, with no numbers.
ReferencesComplete entries in a standard scientific citation style, with every in-text citation matching an entry in the list.A handful of inconsistently formatted links of unclear origin added as padding.

Format

Format checklist

These items are distilled from the official Paper Format Specification (2025 revision). A 2026 revision has since been published; always check the original text for your competition year before submitting.

  • A4 paper, margins of at least 2.5 cm on all sides, bound on the left.
  • Page 1 is the declaration and page 2 the identification-number page, both in the official format without changes.
  • Page 3 is the abstract page, with title and keywords, no longer than one page; page numbering starts at 1.
  • The body starts on page 4, has no table of contents, and should stay within 20 pages where possible.
  • [Disqualifying] Appendices have no page limit but must contain all complete, runnable source code. Missing or non-runnable code costs award eligibility.
  • [Disqualifying] The paper must be fully anonymous: no university, personal name, or other identifying information anywhere in the body or appendices.
  • The electronic submission is a single PDF of no more than 20 MB, with the abstract as its first page.
  • Supporting materials are packaged as one ZIP file of no more than 20 MB.

Templates

Choosing a LaTeX template: recent maintenance before stars

Some templates online still follow the 2023 format, so parts that change from year to year, such as the declaration, the identification page, and the AI-tool usage report, may not match current rules. The table shows the maintenance status of popular GitHub templates as checked in July 2026.

qyxf/mcm-cumcm-thesis still shows up often in search results but has not been updated in seven years and should no longer be used. Whichever template you pick, check it against the current year's specification using the checklist above; confirming compliance is the team's responsibility.

RepositoryStarsStatusNotes
EmpyreanHYR/CUMCM-Latex-template—Updated 2026-06Actively maintained; includes the AI-tool usage report required by the 2025 rules
Sustainable-Enjoyment/CUMCM-LaTeX-Template—Updated 2025-08Maintained by a national first-prize winner
latexstudio/CUMCMThesis940Format current to 2023The most widely circulated legacy template; revise it against the current rules before use
personqianduixue/Math_Model4617Resource collectionPaper templates, modeling code, and study materials
Star counts and update dates were checked in July 2026 and will change over time.

Examples

Past winning papers and official problem reviews

Winning papers for the same problem show a complete write-up, and the problem setters' reviews explain how the problem was designed and what judges looked for. The official entry points below are free to browse; links are listed under Sources.

Ministry of Education student portal: winning CUMCM papers

An official section of the Ministry of Education student portal, with the 2024 and 2025 papers organized by problem.

Organizing committee archive of winning papers (2010 to 2025)

The competition website's collected page of winning papers from past years.

Problem review collection

Judging priorities in the problem setters' own words. Read the review for a problem before reading its winning papers.

Scientify

Give every number in the paper a traceable source

The format checklist only keeps a paper from being disqualified. What judges actually check is the results: whether the numbers in the abstract match the body, and whether the body's numbers can be reproduced by the appendix code. A common slip is changing a parameter on the last night without regenerating the tables and figures.

Scientify is a science agent that runs in an isolated cloud computer. Working from the contest problem, it can search background material, build the model, write and run solver code, perform sensitivity analysis, and produce figures. Data, code, logs, and result files stay in one workspace, so every table and figure in the paper traces back to the code and run that produced it. Tasks keep running after you close your laptop, and you can check progress from your phone. A complete worked case on CUMCM 2023 Problem A is on our modeling competition case page.

Sources

Frequently asked questions

How long should a CUMCM abstract be?

The official specification limits the abstract page, including title and keywords, to one page, so that is the ceiling. Fill it with the four elements of problem, method, numerical results, and conclusion, and make sure the key number for every subproblem appears.

Can we still submit if we did not finish every problem?

Yes. State in the abstract which problems you completed and what results you obtained, and present only finished, verified work in the body. The appendix code must match the paper and must run.

Does a CUMCM paper need a table of contents?

No. The official specification starts the body on page 4 with no table of contents and asks that it stay within 20 pages where possible. Remove the contents page if you reuse an older template.

Are AI tools allowed in CUMCM?

The 2025 revision requires an AI-tool usage report, and any use must be declared truthfully; follow the official notice for your year for details. Some maintained templates already include the report format. Solutions must be the team's own work, with no discussion of the problems with anyone outside the team.

How complete does the appendix code need to be?

The appendix has no page limit but must contain all complete, runnable source code. Missing or non-runnable code costs award eligibility, so run it from scratch in a clean environment before submitting and confirm its output matches the numbers in the paper.

Keep reading

What makes an award-winning modeling paperThe four judging criteria, unpacked with ten official showcase papers.The CUMCM contest, end to endRegistration, schedule, and how awards are decided.A complete CUMCM Problem A caseFrom problem to paper, with code, recomputation logs, and result tables you can inspect.

Hand Scientify the problem and keep every result reproducible

The agent models, solves, validates, and plots in a cloud computer. Code, data, and results stay in one workspace, so every number in the paper traces back to the code that produced it.

Start researching