The Future Roadmap of Mathematics: Open Problems, Emerging Fields, and the Next Century

2. Part I: The State of Mathematics in the 2020s

1.1 The Scale of the Enterprise

The Mathematical Reviews database (MathSciNet) indexed over 4 million items by 2024, covering publications from some 2,500 journals. The Mathematics Subject Classification (MSC 2020) lists 63 top-level categories and over 5,000 subcategories. The number of research mathematicians worldwide is estimated at 60,000–100,000, with roughly 30,000 new PhD dissertations per year. The sheer scale makes it impossible for any individual to command even a small fraction of the total landscape — a situation radically different from 1900, when Hilbert could plausibly survey the entire field.

1.2 The Fragmentation Problem

Modern mathematics suffers from extreme specialization. A typical research paper is read in full by perhaps 10–50 people worldwide. Cross-subfield communication has become difficult: an algebraic geometer may not follow developments in combinatorics, and vice versa. This fragmentation has real costs. It delays the recognition of deep connections (the Langlands program took decades to be understood across its relevant fields), it creates redundant rediscoveries, and it makes the field less accessible to outsiders, including the applied scientists and engineers who depend on mathematical tools.

Against this fragmentation, several unifying currents are visible. Category theory and its higher-categorical generalizations provide a common language that connects algebra, topology, geometry, and logic. The Langlands program links number theory, representation theory, and algebraic geometry. Probabilistic methods have invaded analysis, combinatorics, group theory, and even number theory. And the rise of computational tools — both computer algebra systems and proof assistants — is creating a shared infrastructure that cuts across specializations.

1.3 Key Metrics of the Field

1.4 The Major Breakthroughs of 2010–2025

The period from 2010 to 2025 has been exceptionally productive. Major results include:

• Yitang Zhang (2013) — bounded gaps between primes, proving there are infinitely many pairs of primes differing by at most 70 million (later reduced to 246 by Maynard and the Polymath project)
• Peter Scholze (2012–) — perfectoid spaces, a new framework connecting p-adic Hodge theory, algebraic geometry, and commutative algebra; Scholze’s condensed mathematics program with Dustin Clausen
• Maryam Mirzakhani (2014) — Fields Medal for work on dynamics and geometry of Riemann surfaces and their moduli spaces
• Terence Tao et al. (2015–) — significant progress on the Erdős discrepancy problem (solved), polynomial expansion, and random matrix theory
• June Huh (2022) — Fields Medal for bringing Hodge theory into combinatorics, proving the Rota–Welsh conjecture and the log-concavity conjectures for matroids
• James Maynard (2022) — Fields Medal for advances in analytic number theory, including primes with missing digits
• Maryna Viazovska (2022) — Fields Medal for proving the optimal sphere packing in dimensions 8 and 24 using modular forms
• Liquid tensor experiment (2022) — Scholze’s challenge to formalize a key theorem in condensed mathematics verified in Lean, a landmark in formal verification
• Geordie Williamson (2017–) — used deep learning to discover patterns in Kazhdan–Lusztig polynomials, a pioneering AI-mathematics collaboration
• DeepMind’s FunSearch (2023) — LLM-guided search discovered new solutions to the cap set problem, the first time an LLM contributed to a genuine mathematical discovery
• Kevin Buzzard / Lean community (2020–) — formalization of significant mathematics in Lean 4, mathlib grows to 180,000+ theorems
• Terence Tao’s Polymath-style LLM experiments (2024–) — exploration of AI as a mathematical collaborator

3. Part II: The Millennium Problems — Status Report

In 2000, the Clay Mathematics Institute named seven Millennium Prize Problems, each carrying a $1 million prize. They represent some of the deepest questions in mathematics. As of 2026, one has been solved, one has strong partial results, and five remain wide open.

2.1 Beyond the Millennium Problems

The Millennium list, while iconic, does not capture the full landscape of important open problems. Other problems of comparable depth and difficulty include:

• The Langlands Conjectures — the grand unifying vision linking number theory, representation theory, and algebraic geometry (see Part III)
• The ABC Conjecture — Mochizuki’s claimed proof (2012) via Inter-Universal Teichmüller Theory (IUT) remains controversial; most experts outside Mochizuki’s circle do not accept it. Scholze and Stix identified a critical gap in Corollary 3.12. The problem remains effectively open.
• Twin Prime Conjecture — infinitely many primes p such that p+2 is also prime. Zhang’s bounded gaps result is the strongest evidence; gap reduced to 246 but the final step to 2 seems to require fundamentally new ideas.
• Goldbach’s Conjecture — every even integer > 2 is the sum of two primes. Helfgott (2013) proved the weak (ternary) version. The binary version remains open.
• Collatz Conjecture — Tao (2019) proved that almost all Collatz orbits attain almost bounded values, the strongest result to date, but a full proof seems beyond current methods.
• Invariant Subspace Problem — solved negatively for general Banach spaces (Enflo, 1976; Read, 1985) but remains open for Hilbert spaces.
• Schanuel’s Conjecture — a sweeping conjecture in transcendental number theory that, if proved, would settle almost all open transcendence questions.

4. Part III: Number Theory — The Langlands Program and Beyond

3.1 The Langlands Program: The Grand Unified Theory of Mathematics

Robert Langlands’s 1967 letter to André Weil proposed a network of conjectures connecting automorphic forms, Galois representations, and L-functions. Nearly sixty years later, the Langlands program has become the single most ambitious and interconnected research program in pure mathematics. It has absorbed huge swaths of number theory, algebraic geometry, and representation theory, and its completion would unify large parts of mathematics.

3.2 The Three Langlands Programs

3.3 Arithmetic Geometry: The Scholze Revolution

Peter Scholze’s introduction of perfectoid spaces (2012) has reshaped arithmetic geometry. Perfectoid spaces provide a way to “tilt” between characteristic 0 and characteristic p, transferring problems from one world to the other. This has led to:

• Proof of the monodromy–weight conjecture for complete intersections
• The Scholze–Clausen “condensed mathematics” program, which replaces topological spaces with condensed sets, providing a cleaner foundation for functional analysis within algebra
• New approaches to local Langlands via the Fargues–Fontaine curve
• The “geometrization of the local Langlands correspondence” (Fargues–Scholze, 2021)

Scholze’s work represents a structural transformation comparable to Grothendieck’s schemes: it provides a new language in which old problems become tractable. The condensed mathematics program, in particular, may eventually subsume large parts of analysis within an algebraic framework.

3.4 Analytic Number Theory

The classical territory of primes, L-functions, and sieve methods remains intensely active. Key frontiers include:

• Gaps between primes — the Maynard–Tao theorem gives bounded gaps; the twin prime conjecture remains the goal
• Primes in arithmetic progressions — the Green–Tao theorem (2004) proved arbitrarily long arithmetic progressions in the primes; extensions to polynomial progressions are under investigation
• Distribution of primes — the “alternative” to the Riemann Hypothesis via zero-density estimates and large sieve methods
• Automorphic forms and L-functions — subconvexity bounds, moments of L-functions, quantum unique ergodicity
• Additive combinatorics — the Kelley–Meka breakthrough (2023) on sets without three-term arithmetic progressions, dramatically improving Roth’s theorem

3.5 The Arithmetic of Elliptic Curves

Elliptic curves remain central objects in number theory, connecting BSD, Langlands, Iwasawa theory, and computational number theory. The Bhargava program has transformed the statistical study of elliptic curves: Bhargava and Shankar proved that the average rank of elliptic curves over ℚ is bounded (at most 7/6 under BSD), and that a positive proportion have rank 0 or 1. Whether elliptic curves of arbitrarily large rank exist is still unknown — the current record is rank 29 (Elkies, 2024).

5. Part IV: Geometry, Topology, and the Shape of Space

4.1 Low-Dimensional Topology

After Perelman’s resolution of the Poincaré conjecture and the geometrization program, the focus in 3-manifold topology has shifted to:

• Hyperbolic geometry — Agol’s proof of the virtual Haken conjecture (2012) resolved the last of Thurston’s major conjectures. The study of hyperbolic 3-manifolds continues via arithmetic invariants, quantum invariants, and computational methods.
• 4-manifold topology — remains the most mysterious dimension. The smooth Poincaré conjecture in dimension 4 is open (exotic ℝ⁴ structures exist — a phenomenon unique to dimension 4). The classification of smooth 4-manifolds is far from complete.
• Knot theory — Khovanov homology and its variants categorify the Jones polynomial. Connections to gauge theory (Witten, Floer) and to quantum computing (anyonic models) keep the field interdisciplinary.

4.2 Higher-Dimensional Geometry

The minimal model program (MMP) in birational algebraic geometry aims to classify algebraic varieties, generalizing the classification of surfaces. Birkar’s proof of the boundedness of Fano varieties (BAB conjecture, Fields Medal 2018) was a landmark. The MMP in dimension 4 and higher is largely complete in characteristic 0 but remains challenging in positive characteristic.

4.3 Symplectic and Contact Geometry

Symplectic geometry, born from classical mechanics, has become a central area of pure mathematics through the work of Gromov (pseudo-holomorphic curves, 1985), Floer (Floer homology), and Fukaya (Fukaya categories and mirror symmetry). Key open problems include:

• The Arnold conjecture in full generality (minimum number of fixed points of Hamiltonian diffeomorphisms)
• Homological mirror symmetry (Kontsevich): the equivalence between the Fukaya category of a symplectic manifold and the derived category of coherent sheaves on its mirror. Proved in special cases (Seidel, Sheridan, Abouzaid) but open in general.
• The Weinstein conjecture (existence of closed Reeb orbits) — proved in dimension 3 by Taubes (2007), open in higher dimensions

4.4 Metric Geometry and Optimal Transport

Optimal transport, originating with Monge (1781) and revived by Brenier, Villani, and others, has become a unifying framework connecting PDEs, probability, geometry, and machine learning. The Lott–Sturm–Villani theory defines curvature bounds for metric measure spaces using optimal transport, extending Riemannian geometry to singular spaces. Applications to machine learning (Wasserstein GANs, Sinkhorn distances) have created an unusually direct pipeline from pure mathematics to industry.

4.5 Sphere Packing and Discrete Geometry

Viazovska’s solution of the sphere-packing problem in dimensions 8 (2016) and 24 (2016, with Cohn, Kumar, Miller, and Radchenko) using modular forms was a tour de force. The packing problem remains open in all other dimensions ≥ 4. In dimension 3, the Kepler conjecture was proved by Hales (1998, formally verified in 2014 via the Flyspeck project). The question of optimal packing density in general dimensions is connected to lattice theory, coding theory, and theoretical physics (the cosmological constant via the “swampland”).

6. Part V: Algebra — From Groups to Infinity-Categories

5.1 Higher Category Theory

The most significant structural development in algebra since 2000 is the maturation of higher category theory. Where classical category theory deals with objects, morphisms, and composition, higher category theory introduces 2-morphisms (morphisms between morphisms), 3-morphisms, and so on, up to ∞-categories where this hierarchy is infinite.

Jacob Lurie’s foundational works Higher Topos Theory (2009) and Higher Algebra (2017) provided the technical foundations. The ∞-categorical language has become the standard framework for:

• Derived algebraic geometry (Toën, Vezzosi, Lurie)
• Topological field theories (Lurie’s proof of the cobordism hypothesis)
• Stable homotopy theory (the chromatic program)
• The geometric Langlands program (Gaitsgory et al.)

The formalization of ∞-categories is itself a major ongoing project. Riehl and Verity’s “model-independent” approach, and the connections to homotopy type theory, aim to make the subject accessible beyond the small community of specialists.

5.2 Representation Theory

Representation theory — the study of symmetry through linear algebra — remains one of the most interconnected areas of mathematics. Key frontiers:

• The Langlands program for p-adic groups — understanding the smooth representations of p-adic reductive groups
• Modular representation theory — Williamson’s disproof of the expected bounds in Lusztig’s conjecture (2013) revealed that the subject is more complex than expected
• Geometric representation theory — using algebraic geometry to construct and study representations; perverse sheaves, D-modules, and the geometric Satake correspondence
• Categorification — replacing algebraic structures with categories (e.g., Khovanov homology categorifying the Jones polynomial, Elias–Williamson’s proof of the Kazhdan–Lusztig conjecture via Soergel bimodules)

5.3 Homological and Commutative Algebra

Derived categories and derived algebraic geometry have transformed commutative algebra. The perfectoid revolution (Scholze) has imported new techniques into commutative algebra, including the solution of the direct summand conjecture (André, 2018, using perfectoid methods) and the homological conjectures program. The theory of tight closure and its connections to singularity theory and birational geometry remain active areas.

5.4 Group Theory After the Classification

The Classification of Finite Simple Groups (CFSG), a theorem whose proof spans tens of thousands of pages by hundreds of authors, was effectively completed by 2004. The ongoing “Gorenstein–Lyons–Solomon” revision project aims to produce a self-contained, streamlined proof (~5,000 pages projected). Beyond CFSG, group theory research focuses on:

• The subgroup structure and maximal subgroups of the simple groups
• Profinite groups and their role in Galois theory
• Geometric group theory — the study of groups as geometric objects (Gromov’s program). Agol’s virtual Haken theorem used geometric group theory in an essential way.
• Computational group theory — algorithms for group-theoretic problems, with applications to cryptography (lattice-based and group-based cryptosystems)

7. Part VI: Analysis, PDEs, and the Navier–Stokes Frontier

6.1 The Regularity Problem for Navier–Stokes

The Millennium Prize problem asks: given smooth initial data, does the 3D incompressible Navier–Stokes equation always have a smooth solution for all time? The key developments:

• Tao’s 2016 result: For an “averaged” version of Navier–Stokes, finite-time blowup is possible. This shows that no proof based solely on energy estimates can work — the specific algebraic structure of the nonlinearity must be used.
• Buckmaster–Vicol (2019): Non-uniqueness of weak (Leray–Hopf) solutions, using convex integration techniques developed from the Onsager conjecture program. This challenges the physical relevance of weak solutions.
• The Onsager conjecture: Isett (2018) proved that there exist weak solutions of the Euler equations with Hölder exponent < 1/3 that dissipate energy, confirming Onsager’s 1949 prediction. The sharp boundary at exponent 1/3 connects to turbulence theory.

6.2 Harmonic Analysis and Dispersive PDEs

The school of harmonic analysis centered around the work of Bourgain, Tao, and their collaborators has produced powerful techniques for studying dispersive equations (Schrödinger, wave, KdV). Key open problems:

• The restriction conjecture and the Kakeya conjecture — linked to deep questions in geometric measure theory. The polynomial method (Guth, 2015) has produced the best known bounds.
• Decoupling theory (Bourgain–Demeter, 2015) has resolved several long-standing problems and connected harmonic analysis to number theory (via Vinogradov’s mean value theorem).
• Global well-posedness and scattering for critical dispersive equations in high dimensions

6.3 Geometric Analysis and Ricci Flow

After Perelman’s use of Ricci flow to prove the Poincaré conjecture, geometric flows have become a major tool. Current frontiers include:

• Ricci flow in higher dimensions — Brendle and Schoen’s differentiable sphere theorem (2009)
• Kähler–Ricci flow and the Yau–Tian–Donaldson conjecture on K-stability and the existence of Kähler–Einstein metrics (solved by Chen–Donaldson–Sun, 2015)
• Mean curvature flow and the study of singularities (Colding–Minicozzi)
• Spectral geometry: the relationship between the geometry of a manifold and the spectrum of its Laplacian

6.4 Probability and Stochastic Analysis

The boundary between analysis and probability has dissolved. Key developments:

• Stochastic PDEs — Hairer’s theory of regularity structures (Fields Medal 2014) provided a general framework for making sense of ill-posed SPDEs (like the KPZ equation). This has opened up the mathematical study of physical systems with rough noise.
• Random matrix theory — connections to number theory (zeros of the Riemann zeta function), combinatorics, and quantum chaos
• Schramm–Loewner evolution (SLE) — the mathematical framework for conformally invariant random curves; connections to conformal field theory

8. Part VII: Combinatorics, Probability, and the Discrete Revolution

7.1 The Rise of Combinatorics

Combinatorics has undergone a transformation from a collection of clever tricks to a deep, structurally rich field at the center of mathematics. The driving forces include: the interaction with computer science (complexity theory, algorithms), the infusion of algebraic and topological methods (Huh, Adiprasito, Katz), and the demands of data science and machine learning.

7.2 Extremal and Additive Combinatorics

7.3 Algebraic Combinatorics

The injection of algebraic geometry into combinatorics, pioneered by June Huh, has been transformative. By associating algebraic varieties (or Hodge-theoretic structures) to combinatorial objects like matroids and graphs, Huh and collaborators proved long-standing conjectures:

• The log-concavity of the characteristic polynomial of matroids (Huh 2012, Adiprasito–Huh–Katz 2018)
• The Dowling–Wilson conjecture (Huh–Wang, 2017)
• The top-heavy conjecture for geometric lattices (Braden–Huh–Matherne–Proudfoot–Wang, 2022)

This work demonstrates that deep algebraic structures underlie combinatorial phenomena that appear elementary on the surface. The next frontier is to find a purely combinatorial proof of these results — which would constitute a major advance in understanding.

7.4 Probabilistic Combinatorics and Random Graphs

The probabilistic method, pioneered by Erdős and systematized by Alon and Spencer, remains one of the most powerful tools. Current frontiers include: the study of random graphs above and below the giant component threshold, random regular graphs, percolation on general graphs, and the emerging connections to statistical physics (phase transitions, spin glasses). The spread of randomized methods into algebra, geometry, and number theory is one of the defining trends of 21st-century mathematics.

9. Part VIII: Computation, Complexity, and P vs NP

8.1 Computational Complexity Theory

The P vs NP question — whether every problem whose solution can be verified efficiently can also be solved efficiently — is not just a question about computers. It is a question about the nature of mathematical proof itself: is finding a proof as easy as checking one?

The barrier results are sobering:

The Geometric Complexity Theory (GCT) program of Mulmuley and Sohoni proposes to use algebraic geometry and representation theory to prove lower bounds. The idea is to embed computational problems into spaces of tensors and polynomials and then use the symmetry of these spaces (via representation theory) to prove that certain computations require high complexity. This is a century-scale research program.

8.2 Quantum Computing and Complexity

Quantum computing has created entirely new complexity classes (BQP, QMA) and raised new mathematical questions:

• Quantum supremacy: Google’s 2019 experiment and subsequent claims. The mathematical question is whether there exist problems in BQP \\ BPP — i.e., problems that quantum computers can solve efficiently but classical ones cannot. Shor’s algorithm (factoring) provides a conditional separation.
• Quantum error correction: the theory of topological quantum error-correcting codes connects deep mathematics (topology, knot invariants) to practical engineering
• MIP* = RE (2020): Ji, Natarajan, Vidick, Wright, and Yuen proved that the class of problems verifiable by multi-prover interactive proofs with entangled provers equals the class of recursively enumerable languages. This resolved Tsirelson’s problem and the Connes embedding conjecture (negatively). It is one of the most dramatic interactions between theoretical computer science and pure mathematics in recent decades.

8.3 Algorithmic Mathematics

The boundary between “existence proof” and “algorithm” is increasingly porous. Constructive mathematics, long a philosophical minority, is gaining practical importance through:

• Proof assistants that produce executable code (Lean, Coq extracting programs from proofs)
• Algorithmic algebraic geometry (Gröbner bases, computational homotopy theory)
• Parameterized complexity — classifying problems by the structure of inputs, not just their size
• Fine-grained complexity — understanding the exact polynomial exponents of fundamental algorithms

10. Part IX: Mathematics and Physics — The Quantum Frontier

9.1 Quantum Field Theory and Mathematics

The interaction between mathematics and physics has been intensely productive since the 1980s, when string theory, conformal field theory, and topological quantum field theory generated a flood of mathematical conjectures and results. Key active areas:

• Mirror symmetry: the Kontsevich homological mirror symmetry conjecture (1994) proposes an equivalence between symplectic and algebraic geometry. The Strominger–Yau–Zaslow conjecture provides a geometric picture. Partial results by Seidel, Sheridan, Abouzaid, Ganatra, Pardon, and Shende.
• Gauge theory and topology: Donaldson invariants, Seiberg–Witten invariants, and Floer homology provide invariants of 3- and 4-manifolds derived from physics. The “Atiyah–Floer conjecture” connecting these is a major open problem.
• Topological quantum field theories (TQFTs): Lurie’s proof of the cobordism hypothesis (2009) classifies fully extended TQFTs and connects quantum field theory to higher category theory.
• Conformal field theory (CFT): the conformal bootstrap program, connecting CFT to number theory (modular forms) and probability (SLE). The rigorous construction of interacting CFTs in dimensions > 2 is a major open problem.

9.2 String Theory and Mathematics

Whether or not string theory describes physical reality, it has been an extraordinarily productive source of mathematical conjectures. Key contributions include:

• Calabi–Yau manifolds and the classification of compact geometries
• The monstrous moonshine conjecture (Borcherds, Fields Medal 1998) connecting the monster group to modular functions, and its string-theoretic explanation
• The Gopakumar–Vafa conjecture on BPS state counting
• The AGT correspondence (Alday–Gaiotto–Tachikawa) linking 4D gauge theories, 2D CFTs, and geometric representation theory
• Mathieu moonshine and umbral moonshine — unexpected connections between string compactifications and sporadic groups

9.3 Mathematical General Relativity

The mathematical study of Einstein’s field equations has seen major progress:

• The stability of Minkowski spacetime (Christodoulou–Klainerman, 1993) and ongoing work on the stability of Kerr black holes (the Kerr stability conjecture)
• The Penrose singularity theorems and their relationship to the cosmic censorship conjectures — both weak (singularities are hidden behind horizons) and strong (generic singularities are spacelike). Both remain open.
• Positive mass theorem (Schoen–Yau, Witten) and its generalizations

11. Part X: AI, Proof Assistants, and the Mechanization of Mathematics

10.1 The Proof Assistant Revolution

Proof assistants — software that allows mathematicians to write proofs in a formal language and have them verified by a computer — are transforming how mathematics is done. The key systems:

10.2 The Mathlib Phenomenon

Lean’s community library mathlib deserves special attention. As of 2026, it contains approximately 180,000 formally verified theorems and definitions, covering undergraduate and much of graduate-level mathematics: group theory, ring theory, topology, measure theory, analysis, linear algebra, number theory, and category theory. It is doubling in size roughly every 18 months.

The implications are profound:

• Verification at scale: Any theorem in mathlib has been machine-checked. Human errors — which do exist in the published literature — are structurally eliminated.
• Composability: Formalized results can be combined mechanically, enabling chain-of-reasoning proofs that would be impractical to verify by hand.
• Training data: The formalized corpus serves as training data for AI systems that generate proofs (see below).
• Social infrastructure: mathlib has created a community of hundreds of contributors, blurring the boundary between “doing mathematics” and “programming.”

10.3 AI for Mathematics

The intersection of machine learning and mathematics is evolving rapidly. The current landscape:

10.4 The Long-Term Vision: Mathematics as a Human–Machine Collaboration

The most likely future is not AI replacing mathematicians but AI augmenting them:

• The “proof compiler” model: Mathematicians write high-level proof sketches; AI fills in the details and formalizes them in Lean or Coq. This would dramatically accelerate the pace of formalization.
• The “conjecture engine” model: AI systems scan large datasets (knot tables, number-theoretic functions, algebraic invariants) and propose conjectures for human verification. The mathematician becomes a curator and interpreter of machine-generated hypotheses.
• The “referee assistant” model: AI checks submitted papers for logical errors before human review. This could address the refereeing crisis (long delays, unreliable reviews) that plagues mathematical journals.
• The “Bourbaki 2.0” model: A fully formalized, machine-verified library of all known mathematics, searchable and composable — a 21st-century successor to the Bourbaki project, but written in Lean rather than French.

12. Part XI: Applied Mathematics — Biology, Climate, and Finance

11.1 Mathematical Biology

Biology is becoming the dominant application domain for new mathematics, overtaking physics. Key areas:

• Genomics and topological data analysis (TDA): persistent homology and other tools from algebraic topology are used to analyze the “shape” of high-dimensional biological data. Applications include protein folding (AlphaFold’s success is partly a mathematical achievement), gene expression landscapes, and tumor evolution.
• Epidemiological modeling: COVID-19 exposed both the power and the limitations of mathematical models. The next generation of models integrates network theory, stochastic processes, agent-based simulations, and data assimilation.
• Neuroscience: the mathematical foundations of neural networks (both biological and artificial) involve dynamical systems, information theory, statistical mechanics, and category theory. The mathematical theory of deep learning is still in its infancy.
• Evolutionary dynamics: mathematical population genetics, game theory on graphs, and the stochastic dynamics of evolution remain active areas with deep connections to probability theory.

11.2 Climate and Earth System Modeling

Climate modeling is one of the most computationally intensive applications of mathematics. The mathematical challenges include:

• Multi-scale PDE systems coupling atmosphere, ocean, ice, and carbon cycle
• Uncertainty quantification in chaotic systems
• Data assimilation — combining sparse observational data with model predictions using Bayesian methods and ensemble Kalman filters
• The “digital twin” Earth project (e.g., Destination Earth) requiring exascale computing and new numerical methods

11.3 Machine Learning Theory

The theoretical foundations of deep learning are one of the great open problems of applied mathematics. Why do neural networks generalize so well despite being massively overparameterized? Key research directions:

• Neural tangent kernel theory (Jacot et al., 2018) — connecting wide neural networks to kernel methods
• Mean-field theory — treating neural network training as an optimization over probability measures
• Information geometry — the Fisher information metric on parameter space
• Implicit regularization — understanding why gradient descent finds solutions that generalize, without explicit regularization
• Transformers and attention — the mathematical theory of transformer architectures is essentially nonexistent; developing one is a major open problem
• Scaling laws — the empirical power-law relationship between model size, data, and performance lacks a rigorous mathematical explanation

11.4 Mathematical Finance

After the 2008 crisis exposed the limitations of the Black–Scholes framework and Gaussian copula models, mathematical finance has shifted toward:

• Rough volatility models (Gatheral, Jaisson, Rosenbaum) — using fractional Brownian motion with Hurst parameter H ≈ 0.1
• Market microstructure — the mathematics of order books, high-frequency trading, and optimal execution
• Systemic risk and network models of financial contagion
• Cryptocurrency and DeFi — automated market makers, mechanism design, MEV (maximal extractable value) as a game-theoretic problem

13. Part XII: The Sociology of Mathematics — Who Does It, How, and Where

12.1 The Geography of Mathematics

Mathematics has historically been concentrated in a small number of countries: France, Germany, the UK, the US, and Russia/USSR dominated the 20th century. The 21st century is seeing significant geographic diversification:

12.2 The Diversity Problem

Mathematics has a severe diversity problem. Women constitute approximately 15–20% of mathematics PhDs in the US and an even lower proportion of tenured faculty. The Fields Medal has been awarded to one woman (Mirzakhani, 2014) out of 64 recipients. Racial and ethnic diversity is also extremely low in most countries.

The causes are structural: pipeline leakage at every stage (undergraduate, graduate, postdoc, tenure), implicit bias in hiring and evaluation, the “genius culture” that valorizes innate talent over persistence, and the extreme time demands of an academic career that disproportionately affect those with caregiving responsibilities.

12.3 The Publishing Crisis

Mathematical publishing faces multiple crises:

• Slow refereeing: Average time from submission to publication in top journals is 2–4 years. Some papers wait 5+ years. Referees are unpaid volunteers, overworked and underincentivized.
• Journal costs: Elsevier, Springer, and other commercial publishers charge universities thousands of dollars per subscription. The open-access movement (arXiv, Diamond OA journals like Algebraic Geometry, Compositio, Annals) is growing but incomplete.
• The arXiv problem: Most mathematicians post preprints to arXiv before (or instead of) journal submission. This makes journals function primarily as certification mechanisms rather than dissemination tools. The question is whether journals can be replaced by formal verification + community review.
• Errors in the literature: The rate of serious errors in published mathematics is estimated at 1–5%. Formal verification could eliminate this, but the formalization cost is currently high (~10x the effort of writing an informal proof).

12.4 The Competition Pipeline

Mathematical olympiads (IMO, Putnam) play an outsized role in identifying and training young mathematicians. However, the relationship between competition success and research success is weaker than often assumed. Many Fields Medalists were not IMO gold medalists; conversely, many IMO gold medalists do not pursue research careers. The competition system also reinforces the “speed and cleverness” model of mathematics at the expense of the “depth and patience” model that characterizes research.

14. Part XIII: The Next Century — Ten Theses on the Future of Mathematics

Thesis 1: Formal Verification Will Become Standard

By 2050, major journals will require or strongly encourage formally verified proofs for submitted papers. The current trajectory of Lean/mathlib, combined with AI-assisted autoformalization, will reduce the cost of formalization to 2–3x the effort of writing an informal proof (down from ~10x today). This will not happen suddenly; it will begin with subfields where formalization is most natural (algebra, combinatorics) and gradually spread.

Thesis 2: AI Will Not Replace Mathematicians but Will Radically Change Their Work

The most important mathematical breakthroughs require conceptual leaps — the invention of new frameworks, the recognition of deep analogies, the ability to ask the right questions. These remain beyond current AI capabilities. However, AI will handle routine proof steps, verify arguments, search for counterexamples, and generate conjectures from data. The mathematician of 2050 will spend less time on technical computation and more time on conceptual architecture.

Thesis 3: The Langlands Program Will Be Substantially Completed

The geometric Langlands conjecture proof (Gaitsgory et al., 2024) is a turning point. The classical and p-adic cases will follow, though perhaps not in full generality within this century. The key will be finding the right categorical framework — likely ∞-categorical — that makes the conjectures “obvious” from the right perspective.

Thesis 4: The Riemann Hypothesis Will Remain Open for Decades

Unlike the Langlands program, where there is a clear direction of attack, the Riemann Hypothesis has no credible proof strategy. It may require entirely new mathematics — perhaps ideas from quantum field theory, spectral theory, or a yet-uninvented framework. It is plausible that RH will be the last of the Millennium Problems to be resolved. It is also plausible (though unlikely) that it is undecidable.

Thesis 5: P vs NP Will Remain Open Even Longer

The barrier results in complexity theory are more fundamental than those facing RH. The Geometric Complexity Theory program is a century-scale enterprise. Unless a completely unexpected approach emerges, P vs NP will remain open through the 21st century. However, partial results (conditional separations, circuit lower bounds for restricted models) will continue to advance.

Thesis 6: Applied Mathematics Will Become the Largest Branch

The demand for mathematical expertise in machine learning, computational biology, climate modeling, quantum computing, and finance will grow faster than the demand for pure mathematical research. By 2050, applied and computational mathematics will employ more researchers than pure mathematics. This is not a crisis for pure mathematics — it is a natural rebalancing. The deepest applications will continue to draw on pure mathematical ideas, as optimal transport, algebraic topology (TDA), and random matrix theory already demonstrate.

Thesis 7: Category Theory Will Become the Universal Language

Higher category theory and derived algebraic geometry are already the lingua franca of advanced algebraic geometry, topology, and mathematical physics. As formalization progresses (and proof assistants are built on type theory, which is categorically native), categorical thinking will spread to analysis, probability, and applied mathematics. The “unreasonable effectiveness of category theory” will be a defining theme of 21st-century mathematics.

Thesis 8: The Distinction Between Pure and Applied Mathematics Will Dissolve

The 20th-century separation of pure and applied mathematics, driven by Bourbaki’s influence and Cold War funding structures, is already weakening. Number theory is applied in cryptography. Algebraic topology is applied in data analysis. Representation theory is applied in quantum computing. The mathematicians of 2050 will not identify as “pure” or “applied” but as specialists in specific problems that draw on whatever tools are needed.

Thesis 9: New Foundations May Emerge

Homotopy type theory (HoTT) offers an alternative to set-theoretic foundations that is natively categorical, constructive, and amenable to computer formalization. Whether HoTT or a descendant will replace ZFC as the working foundation of mathematics remains unclear, but the pressure from formalization and computation is pushing foundations in a more constructive, type-theoretic direction. The question “what are the right foundations?” — once considered settled — is open again.

Thesis 10: Mathematics Will Remain a Human Activity

Despite all the changes — AI, formalization, interdisciplinary pressure, the erosion of traditional boundaries — mathematics will remain fundamentally a human creative endeavor. The experience of mathematical understanding, the aesthetic pleasure of an elegant proof, the sudden flash of insight that resolves a long-standing confusion — these are cognitive experiences that machines can simulate but (as far as we know) do not share. Mathematics will be augmented by machines, but the drive to understand — to see why, not just that — will remain human.

15. Interactive Timeline: Key Events and Projected Milestones

16. Active Research Frontiers: Searchable Table

17. Research Activity by Field (Estimated Relative Volume, 2024)

18. Bibliography

Surveys and General References

• Avigad, Jeremy. “The Mechanization of Mathematics.” Notices of the AMS 65, no. 6 (2018): 681–690.
• Buzaglo, Meir. The Logic of Concept Expansion. Cambridge University Press, 2002.
• Gowers, Timothy, ed. The Princeton Companion to Mathematics. Princeton University Press, 2008.
• Jaffe, Arthur, and Frank Quinn. “Theoretical Mathematics: Toward a Cultural Synthesis of Mathematics and Theoretical Physics.” Bulletin of the AMS 29 (1993): 1–13.
• Villani, Cédric. Birth of a Theorem: A Mathematical Adventure. Farrar, Straus and Giroux, 2015.
• Zelmanov, Efim. “Mathematics: State of the Art.” ICM 2022 plenary lecture.

Number Theory and Langlands

• Frenkel, Edward. Love and Math: The Heart of Hidden Reality. Basic Books, 2013.
• Gaitsgory, Dennis, et al. “Proof of the Geometric Langlands Conjecture.” Preprint, 2024. arXiv:2405.03599.
• Scholze, Peter. “Perfectoid Spaces.” Publ. Math. IHÉS 116 (2012): 245–313.
• Scholze, Peter, and Dustin Clausen. “Condensed Mathematics and Complex Geometry.” Lecture notes, 2022.
• Taylor, Richard. “Automorphy for Some l-adic Lifts of Automorphic mod l Galois Representations.” Publ. Math. IHÉS 108 (2008): 1–181.
• Bhargava, Manjul, and Arul Shankar. “Binary Quartic Forms Having Bounded Invariants.” Annals of Mathematics 181 (2015): 191–242.

Geometry and Topology

• Agol, Ian. “The Virtual Haken Conjecture.” Documenta Mathematica 18 (2013): 1045–1087.
• Hales, Thomas, et al. “A Formal Proof of the Kepler Conjecture.” Forum of Mathematics, Pi 5 (2017).
• Kontsevich, Maxim. “Homological Algebra of Mirror Symmetry.” ICM 1994 proceedings.
• Viazovska, Maryna. “The Sphere Packing Problem in Dimension 8.” Annals of Mathematics 185 (2017): 991–1015.

Algebra and Category Theory

• Lurie, Jacob. Higher Topos Theory. Princeton University Press, 2009.
• Riehl, Emily, and Dominic Verity. Elements of ∞-Category Theory. Cambridge University Press, 2022.
• Huh, June. “Combinatorics and Hodge Theory.” ICM 2022 proceedings.

Analysis and PDEs

• Hairer, Martin. “A Theory of Regularity Structures.” Inventiones Mathematicae 198 (2014): 269–504.
• Tao, Terence. “Finite Time Blowup for an Averaged Three-Dimensional Navier–Stokes Equation.” JAMS 29 (2016): 601–674.
• Buckmaster, Tristan, and Vlad Vicol. “Nonuniqueness of Weak Solutions to the Navier–Stokes Equation.” Annals of Mathematics 189 (2019): 101–144.
• Bourgain, Jean, and Ciprian Demeter. “The Proof of the l² Decoupling Conjecture.” Annals of Mathematics 182 (2015): 351–389.

Combinatorics

• Campos, Marcelo, et al. “An Exponential Improvement for Diagonal Ramsey.” Preprint, 2023. arXiv:2303.09521.
• Kelley, Zander, and Raghu Meka. “Strong Bounds for 3-Progressions.” Preprint, 2023. arXiv:2302.05537.
• Adiprasito, Karim, June Huh, and Eric Katz. “Hodge Theory for Combinatorial Geometries.” Annals of Mathematics 188 (2018): 381–452.

Computation and Complexity

• Aaronson, Scott. Quantum Computing Since Democritus. Cambridge University Press, 2013.
• Arora, Sanjeev, and Boaz Barak. Computational Complexity: A Modern Approach. Cambridge University Press, 2009.
• Ji, Zhengfeng, et al. “MIP* = RE.” Communications of the ACM 64 (2021): 131–138.
• Mulmuley, Ketan, and Milind Sohoni. “Geometric Complexity Theory I.” SIAM Journal on Computing 31 (2001): 496–526.

AI and Proof Assistants

• Buzzard, Kevin. “The Future of Mathematics?” ICM 2022 invited lecture.
• Davies, Alex, et al. “Advancing Mathematics by Guiding Human Intuition with AI.” Nature 600 (2021): 70–74.
• Romera-Paredes, Bernardino, et al. “Mathematical Discoveries from Program Search with Large Language Models.” Nature 625 (2024): 468–475.
• Trinh, Trieu, et al. “Solving Olympiad Geometry without Human Demonstrations.” Nature 625 (2024): 476–482.
• The mathlib community. “The Lean Mathematical Library.” CPP 2020 proceedings.

Mathematical Physics

• Witten, Edward. “Quantum Field Theory and the Jones Polynomial.” Communications in Mathematical Physics 121 (1989): 351–399.
• Costello, Kevin, and Owen Gwilliam. Factorization Algebras in Quantum Field Theory. Cambridge University Press, 2017.

Applied Mathematics

• Villani, Cédric. Topics in Optimal Transportation. AMS, 2003.
• Belkin, Mikhail, et al. “Reconciling Modern Machine Learning Practice and the Bias–Variance Trade-Off.” PNAS 116 (2019): 15849–15854.
• Gatheral, Jim, Thibault Jaisson, and Mathieu Rosenbaum. “Volatility Is Rough.” Quantitative Finance 18 (2018): 933–949.