Related work

Formalization work is growing quickly, and very little of it concerns numerical methods. Two figures, redrawn from our paper, show this. Lean libraries of numerical analysis have appeared only since the rise of AI agents: NLA-lib, a library of numerical linear algebra meant also for research done by agents, and our own VSCL. The reading list names, by topic, the work that we build on and stand beside.

Formalization work is growing, and little of it is numerical

arXiv submissions per month

29 submissions on numerical topics, against 1,271 on formalization.

  • Formalization, 1,271 in all
  • of which mention Lean or Mathlib, 806 in all
  • Numerical topics, 29 in all
0100200Feb 2024202520265130158Feb 2024: Formalization 17, of which mention Lean or Mathlib 6, Numerical topics 0Mar 2024: Formalization 16, of which mention Lean or Mathlib 5, Numerical topics 0Apr 2024: Formalization 12, of which mention Lean or Mathlib 3, Numerical topics 0May 2024: Formalization 11, of which mention Lean or Mathlib 5, Numerical topics 0Jun 2024: Formalization 14, of which mention Lean or Mathlib 6, Numerical topics 1Jul 2024: Formalization 20, of which mention Lean or Mathlib 9, Numerical topics 0Aug 2024: Formalization 11, of which mention Lean or Mathlib 5, Numerical topics 0Sep 2024: Formalization 8, of which mention Lean or Mathlib 4, Numerical topics 0Oct 2024: Formalization 27, of which mention Lean or Mathlib 11, Numerical topics 0Nov 2024: Formalization 7, of which mention Lean or Mathlib 2, Numerical topics 0Dec 2024: Formalization 15, of which mention Lean or Mathlib 3, Numerical topics 1Jan 2025: Formalization 23, of which mention Lean or Mathlib 8, Numerical topics 1Feb 2025: Formalization 26, of which mention Lean or Mathlib 12, Numerical topics 1Mar 2025: Formalization 18, of which mention Lean or Mathlib 12, Numerical topics 2Apr 2025: Formalization 23, of which mention Lean or Mathlib 8, Numerical topics 0May 2025: Formalization 31, of which mention Lean or Mathlib 17, Numerical topics 1Jun 2025: Formalization 23, of which mention Lean or Mathlib 11, Numerical topics 0Jul 2025: Formalization 25, of which mention Lean or Mathlib 10, Numerical topics 1Aug 2025: Formalization 21, of which mention Lean or Mathlib 10, Numerical topics 0Sep 2025: Formalization 35, of which mention Lean or Mathlib 17, Numerical topics 1Oct 2025: Formalization 45, of which mention Lean or Mathlib 29, Numerical topics 1Nov 2025: Formalization 32, of which mention Lean or Mathlib 19, Numerical topics 0Dec 2025: Formalization 31, of which mention Lean or Mathlib 18, Numerical topics 1Jan 2026: Formalization 30, of which mention Lean or Mathlib 18, Numerical topics 0Feb 2026: Formalization 40, of which mention Lean or Mathlib 22, Numerical topics 2Mar 2026: Formalization 60, of which mention Lean or Mathlib 40, Numerical topics 2Apr 2026: Formalization 90, of which mention Lean or Mathlib 70, Numerical topics 1May 2026: Formalization 76, of which mention Lean or Mathlib 47, Numerical topics 5Jun 2026: Formalization 108, of which mention Lean or Mathlib 81, Numerical topics 1Jul 2026: Formalization 106, of which mention Lean or Mathlib 78, Numerical topics 2Aug 2026: Formalization 112, of which mention Lean or Mathlib 90, Numerical topics 0Sep 2026: Formalization 158, of which mention Lean or Mathlib 130, Numerical topics 5

New public GitHub repositories per month

140 repositories on numerical topics with Lean 4, against 12,523 that mention Lean 4 or Mathlib.

  • Lean 4 or Mathlib, 12,523 in all
  • Autoformalization, 667 in all
  • Lean 4 with numerical topics, 140 in all
01,5003,000Jan 20242025202641742,804Jan 2024: Lean 4 or Mathlib 61, Autoformalization 3, Lean 4 with numerical topics 0Feb 2024: Lean 4 or Mathlib 48, Autoformalization 3, Lean 4 with numerical topics 0Mar 2024: Lean 4 or Mathlib 61, Autoformalization 7, Lean 4 with numerical topics 0Apr 2024: Lean 4 or Mathlib 82, Autoformalization 8, Lean 4 with numerical topics 1May 2024: Lean 4 or Mathlib 82, Autoformalization 7, Lean 4 with numerical topics 0Jun 2024: Lean 4 or Mathlib 75, Autoformalization 9, Lean 4 with numerical topics 0Jul 2024: Lean 4 or Mathlib 65, Autoformalization 1, Lean 4 with numerical topics 1Aug 2024: Lean 4 or Mathlib 76, Autoformalization 2, Lean 4 with numerical topics 1Sep 2024: Lean 4 or Mathlib 86, Autoformalization 15, Lean 4 with numerical topics 1Oct 2024: Lean 4 or Mathlib 74, Autoformalization 8, Lean 4 with numerical topics 1Nov 2024: Lean 4 or Mathlib 84, Autoformalization 4, Lean 4 with numerical topics 1Dec 2024: Lean 4 or Mathlib 87, Autoformalization 6, Lean 4 with numerical topics 2Jan 2025: Lean 4 or Mathlib 97, Autoformalization 8, Lean 4 with numerical topics 1Feb 2025: Lean 4 or Mathlib 100, Autoformalization 13, Lean 4 with numerical topics 3Mar 2025: Lean 4 or Mathlib 104, Autoformalization 13, Lean 4 with numerical topics 0Apr 2025: Lean 4 or Mathlib 105, Autoformalization 6, Lean 4 with numerical topics 0May 2025: Lean 4 or Mathlib 156, Autoformalization 17, Lean 4 with numerical topics 1Jun 2025: Lean 4 or Mathlib 131, Autoformalization 13, Lean 4 with numerical topics 0Jul 2025: Lean 4 or Mathlib 143, Autoformalization 14, Lean 4 with numerical topics 2Aug 2025: Lean 4 or Mathlib 132, Autoformalization 13, Lean 4 with numerical topics 2Sep 2025: Lean 4 or Mathlib 172, Autoformalization 10, Lean 4 with numerical topics 0Oct 2025: Lean 4 or Mathlib 165, Autoformalization 19, Lean 4 with numerical topics 3Nov 2025: Lean 4 or Mathlib 223, Autoformalization 15, Lean 4 with numerical topics 1Dec 2025: Lean 4 or Mathlib 280, Autoformalization 21, Lean 4 with numerical topics 2Jan 2026: Lean 4 or Mathlib 327, Autoformalization 33, Lean 4 with numerical topics 6Feb 2026: Lean 4 or Mathlib 459, Autoformalization 31, Lean 4 with numerical topics 6Mar 2026: Lean 4 or Mathlib 723, Autoformalization 41, Lean 4 with numerical topics 4Apr 2026: Lean 4 or Mathlib 699, Autoformalization 37, Lean 4 with numerical topics 10May 2026: Lean 4 or Mathlib 872, Autoformalization 39, Lean 4 with numerical topics 11Jun 2026: Lean 4 or Mathlib 909, Autoformalization 60, Lean 4 with numerical topics 12Jul 2026: Lean 4 or Mathlib 1,287, Autoformalization 69, Lean 4 with numerical topics 13Aug 2026: Lean 4 or Mathlib 1,754, Autoformalization 48, Lean 4 with numerical topics 14Sep 2026: Lean 4 or Mathlib 2,804, Autoformalization 74, Lean 4 with numerical topics 41
Method and data of this figure

The first panel counts arXiv submissions per month, February 2024 to September 2026, whose title or abstract mentions autoformalization, or a form of the word formalization together with a proof assistant. The dashed line counts those that mention Lean or Mathlib. The bars count submissions on floating-point arithmetic, rounding error, numerical analysis or numerical linear algebra together with a proof assistant or formal verification. The second panel counts public GitHub repositories created per month, January 2024 to September 2026, whose name, description or README mentions Lean 4 or Mathlib, autoformalization, or Lean 4 together with floating point, rounding error, numerical analysis or numerical linear algebra. Both counts were retrieved through the public arXiv and GitHub search interfaces on 2026-11-11.

arXiv submissions

MonthFormalizationof which mention Lean or MathlibNumerical topics
Feb 20241760
Mar 20241650
Apr 20241230
May 20241150
Jun 20241461
Jul 20242090
Aug 20241150
Sep 2024840
Oct 202427110
Nov 2024720
Dec 20241531
Jan 20252381
Feb 202526121
Mar 202518122
Apr 20252380
May 202531171
Jun 202523110
Jul 202525101
Aug 202521100
Sep 202535171
Oct 202545291
Nov 202532190
Dec 202531181
Jan 202630180
Feb 202640222
Mar 202660402
Apr 202690701
May 202676475
Jun 2026108811
Jul 2026106782
Aug 2026112900
Sep 20261581305

New public GitHub repositories

MonthLean 4 or MathlibAutoformalizationLean 4 with numerical topics
Jan 20246130
Feb 20244830
Mar 20246170
Apr 20248281
May 20248270
Jun 20247590
Jul 20246511
Aug 20247621
Sep 202486151
Oct 20247481
Nov 20248441
Dec 20248762
Jan 20259781
Feb 2025100133
Mar 2025104130
Apr 202510560
May 2025156171
Jun 2025131130
Jul 2025143142
Aug 2025132132
Sep 2025172100
Oct 2025165193
Nov 2025223151
Dec 2025280212
Jan 2026327336
Feb 2026459316
Mar 2026723414
Apr 20266993710
May 20268723911
Jun 20269096012
Jul 20261,2876913
Aug 20261,7544814
Sep 20262,8047441

The lines count formalization work per month, and the bars count formal work on numerical topics. Part of the 2026 rise in repositories reflects ones that coding agents generated. Redrawn from our paper on linear scales.

How little formal mathematics concerns numerical methods

Mathlib, by area · lines that begin a theorem or lemma

Numerical analysis has 13 of the 162,258 lines shown.

  • Algebra58,225
  • Analysis33,117
  • Topology18,806
  • Logic, sets, order17,256
  • Category theory12,573
  • Geometry7,662
  • Number theory5,539
  • Combinatorics5,480
  • Probability3,587
  • Numerical analysis13

Benchmarks and other libraries · items on numerical methods, of all items

Each bar is the whole of its collection. The collections differ in size and in unit.

  • miniF2F problems0 of 488
  • PutnamBench problems0 of 672
  • ProofNet problems0 of 371
  • FormalMATH problems1 of 5,560
  • Archive of Formal Proofs entries12 of 1,040
  • Coq and Rocq packages19 of 593
Method and data of this figure

Mathlib is counted at commit 021ce68 of 2026-10-07. The 13 lines on numerical analysis are in two files: an experimental floating-point model and the error bound of the trapezoidal rule. Mathlib's infrastructure directories (19,742 lines) are left out of the panel, so its areas add up to 162,258 of about 182,000. The second panel counts problems in four Lean benchmarks, entries in Isabelle's Archive of Formal Proofs, and released Coq and Rocq packages, and how many of each concern numerical methods. The one FormalMATH problem runs three steps of gradient descent in exact arithmetic. The counts come from keyword searches whose hits were read one by one, and the units differ from row to row.

Mathlib, by area

Arealines that begin a theorem or lemma
Algebra58,225
Analysis33,117
Topology18,806
Logic, sets, order17,256
Category theory12,573
Geometry7,662
Number theory5,539
Combinatorics5,480
Probability3,587
Numerical analysis13

Benchmarks and other libraries

CollectionOn numerical methodsAll
miniF2F problems0488
PutnamBench problems0672
ProofNet problems0371
FormalMATH problems15,560
Archive of Formal Proofs entries121,040
Coq and Rocq packages19593

The first panel counts lines that begin a theorem or lemma in Mathlib, by area, where a keyword search finds 13 for numerical analysis. The second panel shows how many items of four Lean benchmarks and of two other collections concern numerical methods. Redrawn from our paper.

Reading list

A listing is not an endorsement in either direction.

Machine-checked theorems about whole algorithms

Most of these works prove rounding-error results for whole algorithms and programs in a proof assistant. They cover the Cholesky factorization, inner and matrix products, the Jacobi iteration, a wave-equation program, an ODE solver, and fast Fourier transforms, and some connect their theorems to C code. Machine-checked versions of the whole-algorithm theorems in textbooks remain few.

Automatic tools for rounding-error bounds

Some tools bound the rounding error of a given program automatically. They handle straight-line code and loops with a fixed bound, and they validate their results in different ways, for example with proofs checked in Coq, HOL Light, or PVS, or with a formally verified certificate checker. Type systems such as Bean track backward error through a program.

Formal floating-point arithmetic

Rounding-error proofs rest on assumptions about how each arithmetic operation rounds. Formal arithmetic libraries help justify those assumptions for particular formats, and some also prove bounds for sums and dot products. In Lean, three such libraries were published in 2026.

AI-assisted formalization pipelines

These projects describe ways to organize AI-assisted formalization and to review its results, for textbooks, course material, and a catalogue of open problems. Some divide the work among agents in separate roles, and others run it as an open protocol or platform to which anyone's agents can contribute.

Checking statements against their sources

The Lean comparator checks that a proof establishes a statement as written, not that the statement is faithful to its source. Palomar and Tau Ceti address faithfulness directly, through AI review of statements and through review among agents under rubrics written by people. In Choir's demonstration project, a condition omitted when two theorems were transcribed was found by a later review and corrected.

Connecting guarantees to running code

A theorem about an algorithm does not by itself establish a guarantee for a particular kernel. Evidence about a kernel can be a program proof, a sound output checker, or a test. The first three works give an example of each. The last two are AI-generated solvers that come with Lean files.

AI systems in mathematics and its formalization

These are reports on mathematics done with AI systems, most of them from AI laboratories and companies, and several come with Lean formalizations.

Formal libraries of mathematics

Machine-checked formalizations rely on formal libraries, which are collections of definitions, theorems and lemmas that a proof assistant checks. Most of the AI systems above build on Mathlib, the main library of Lean.

Benchmarks for formal theorem proving

Benchmarks hold formal problems for AI systems to prove. Across four widely used Lean benchmarks, one of the 7,091 problems concerns a numerical method.

Numerical software

Netlib is a collection of mathematical software, papers and databases. LAPACK, the reference library of dense linear algebra, is distributed there, and its test suite has checked the library's numerical behavior for three decades. The LAPACK Working Notes are a series of reports on its development.

Other software by our group

RandBLAS and RandLAPACK are C++ libraries for randomized numerical linear algebra (RandNLA). Members of our group are among their developers.

  • RandBLASGitHub repositoryExtends the BLAS tradition to randomized operations, such as sketching dense and sparse data.
  • RandLAPACKGitHub repositoryDriver algorithms and building blocks for RandNLA, including randomized SVD and QR with column pivoting.