OPENPROBLEMBENCH · 2026DETERMINISTIC VERIFICATION

Open problems, finite certificates

OPBench is an extensible benchmark and public research interface for AI attempts on open mathematical problems—centered on outputs that independent programs can verify.

SELECT A CONJECTURE
FRONTIER NEWS

AI-assisted mathematics, with the verification status attached

Published attempts, expert review, exact certificates, and formal proofs are labeled separately.

Jacobian Conjecture · Tangent-Sweep Follow-up

A geometric account extends the counterexample mechanism

A follow-up preprint develops a self-contained tangent-sweep explanation of the recent counterexample and constructs families in every dimension above two with arbitrarily large geometric degree. It is presented as a research preprint, distinct from the separately formalized finite certificate.

GEOMETRIC FOLLOW-UP PREPRINTarXiv preprint
Hessian Conjecture · Claude Fable 5

A five-variable counterexample follows the Jacobian breakthrough

A new preprint derives an exact integer-polynomial counterexample to the Hessian conjecture from the Alpöge–Fable construction, with all defining identities checked in exact rational arithmetic.

EXACT-ARITHMETIC PREPRINTarXiv preprint
Dinitz–Garg–Goemans Conjecture · GPT-5.6 Pro

A finite unsplittable-flow counterexample awaits specialist review

A released construction separates fractional cost 58 from every admissible unsplittable solution of cost at least 60. The finite certificate is checkable, but no stable scholarly manuscript or independent specialist review is yet available.

PENDING INDEPENDENT REVIEWMathematics & Science Breakthrough Tracker
Černý Conjecture · Codex with GPT-5.6 Sol

The one-cluster case receives an AI-assisted proof

A preprint proves the Černý bound for synchronizing one-cluster automata and a sharper parameterized result. The author states that the proof arose through interaction with Codex and was then verified by the author.

AUTHOR-VERIFIED PREPRINTarXiv preprint
Chernoff Density Conjecture · GPT-5.6 Sol

A 2014 strong log-concavity conjecture is proved

A statistical-theory note proves that Chernoff’s density is strongly log-concave and reports that GPT-5.6 Sol generated the proof in its entirety. The result is public as a research preprint.

PUBLIC PREPRINTarXiv preprint
Jacobian Conjecture · Claude Fable 5

An explicit counterexample appears in complex dimension three

Levent Alpöge released a degree-seven polynomial map with constant Jacobian −2 and three distinct inputs sharing one image. Exact-arithmetic and independent Lean checks now make the finite certificate directly auditable.

INDEPENDENTLY CHECKED CERTIFICATEJacobian Conjectures Atlas
Cycle Double Cover Conjecture · GPT-5.6 Sol Ultra

Independent expositions clarify an AI-generated proof

After OpenAI released a concise proof, graph theorists published detailed expositions of the reduction and linear-algebra argument. The theorem says every bridgeless graph has a cycle double cover.

EXPERT-EXPOUNDED PROOFIndependent mathematical exposition
Derivative-Free Optimization Problem · GPT-5.6 Sol Pro

A 30-year oracle-complexity gap closes up to logarithmic factors

A 148-minute AI-assisted session produced the central lower-bound argument for derivative-free convex optimization. The author checked it and supplied a Lean 4 formalization; the preprint remains unreviewed.

FORMALIZED · NOT PEER REVIEWEDarXiv preprint + Lean artifact
Sabidussi Compatibility Conjecture · GPT-5.6 Sol

A graph-decomposition conjecture is proved and formalized in Lean

The proof partitions the edges of an even multigraph into compatible circuits and strengthens the statement to a four-colouring theorem. A public Lean 4 formalization accompanies the preprint.

FORMALIZED PREPRINTarXiv preprint + Lean artifact
Benjamini–Hochberg FDR Conjecture · GPT-5.6 Pro

Interval arithmetic certifies failure under correlated tests

A factor-model construction proves that the procedure can exceed its nominal false-discovery rate for correlated two-sided Gaussian tests. The paper says GPT-5.6 Pro obtained the proof and the author carefully checked it.

AUTHOR-CHECKED CERTIFICATEarXiv preprint
Erdős–Szemerédi Sum–Product Conjecture · GPT-5.5 Pro

Seven autonomous runs reproduce a disproof over the reals

A three-stage agent generated correct proofs in seven of eight trials, including constructions unlike the public human proof. The author independently verified every included proof and released the full traces.

HUMAN-VERIFIED REPRODUCTIONarXiv preprint + released traces
Jamming Exponent Identity · Claude Sonnet 4.6 + Opus 4.7

AI assistance helps close a long-standing physics derivation

Giorgio Parisi and Francesco Zamponi report a proof of an identity between critical exponents of jamming, with Claude used to explore and refine the argument before the authors verified the final derivation.

AUTHOR-VERIFIED DERIVATIONarXiv preprint
Erdős + OEIS Problems · AlphaProof Nexus

Formal proof search settles dozens of open entries

The AlphaProof Nexus study reports machine-checked Lean proofs for nine of 353 open Erdős problems and 44 of 492 open OEIS problems, pairing neural search with a kernel-verifiable endpoint.

FORMALLY VERIFIEDarXiv preprint + Lean proofs
Planar Unit-Distance Conjecture · OpenAI Reasoning Model

A number-theoretic construction disproves a central geometry conjecture

An AI-generated construction refutes the conjectured near-linear upper bound for unit distances in the plane. Mathematicians checked the result and extracted the unexpected bridge from algebraic number theory.

MATHEMATICIAN-VERIFIEDOpenAI report + mathematical checks
Kissing Number Problem · AlphaEvolve

A 2026 impact review tracks a machine-found lower-bound improvement

DeepMind’s 2026 review revisits AlphaEvolve’s 11-dimensional construction, which raised the known kissing-number lower bound from 592 to 593. This is a record improvement, not a solution of the full problem.

BOUND IMPROVEMENT · NOT SOLVEDGoogle DeepMind impact review
Directed-Cycle Decomposition Problem · Claude Opus 4.6

Claude finds the construction behind Knuth’s open problem

Claude discovered a general decomposition of a three-dimensional directed grid into Hamiltonian cycles after a guided computational search. Donald Knuth checked the construction and wrote the rigorous proof.

CHECKED BY DONALD KNUTHDonald Knuth, Stanford
First Proof · OpenAI Reasoning Model

OpenAI releases attempts for all ten research-level proof problems

OpenAI published complete attempts for the ten First Proof problems. Expert feedback judged at least five attempts highly likely to be correct, while clearly leaving the remaining submissions under review rather than treating model output as a verdict.

10 ATTEMPTS · EXPERT REVIEWOpenAI · First Proof
Erdős Open Problems · Aletheia / Gemini Deep Think

Human–AI collaboration advances several open Erdős problems

Google DeepMind reports progress on four research problems, combining long-context mathematical reasoning with expert steering and verification. The cases illustrate collaboration rather than unattended theorem production.

EXPERT-STEERED AND CHECKEDGoogle DeepMind
COMPLEX POLYNOMIAL MAPS

Jacobian Conjecture

Can a polynomial map be locally invertible everywhere yet fail to be globally injective? OPBench asks models for explicit maps and finite collision witnesses that an offline symbolic program can check.

Resolved for dimensions ≥ 3 in 2026; the plane case remains open

05
problems
05
models
50
records
01
offline passes
problems/jacobian_conjecture.jsonlresults/jacobian_conjecture/
SOURCE · ℂ³TARGET · ℂ³LOCAL · EACH SHEET INVERTIBLEGLOBAL · THREE POINTS, ONE IMAGE
nonzero constant Jacobian collision fiber
F(p1)=F(p2)=F(p3),detJF=2F(p_1)=F(p_2)=F(p_3),\qquad \det J_F=-2

The 2026 three-dimensional certificate combines a nonzero constant Jacobian with three rational points in one fiber.

00 / STATEMENT

For a polynomial map over the complex numbers,

F=(F1,,Fd):CdCd,detJFC×F=(F_1,\ldots,F_d):\mathbb C^d\longrightarrow\mathbb C^d,\qquad \det J_F\in\mathbb C^\times

the conjecture asserted that FF is a polynomial automorphism. A counterexample supplies distinct points pqp\ne q with F(p)=F(q)F(p)=F(q).

Every displayed result is checked by an offline symbolic program. No LLM judge decides the algebraic verdict.

01 / MATHEMATICAL ATLAS

From the conjecture to the first counterexample

The frontier changed in July 2026: dimension three is false, stabilization settles every higher dimension, and dimension two remains the open core.

Progress timeline

  1. Keller formulates the problem

    A nonzero constant Jacobian is conjectured to force polynomial invertibility.

    Original paper
  2. Degree reduction reshapes the search

    Bass, Connell, and Wright reduce the general problem to special cubic-homogeneous maps.

    Bass–Connell–Wright
  3. A three-dimensional certificate is independently formalized

    An explicit degree-seven map has determinant −2 and three distinct rational preimages of one point; Isabelle/HOL verifies the finite claims.

    Archive of Formal Proofs
  4. The geometry and higher-degree families are developed

    A self-contained tangent-sweep account gives examples in every dimension above two and of arbitrarily large geometric degree.

    arXiv:2608.00222
02 / BENCHMARK

Five problems, one capability curve

The suite moves from unconstrained construction to degree, fiber-size, and open two-dimensional frontiers. Every task uses the same exact certificate protocol.

Shared mathematical context

Let F=(F1,,Fd):CdCdF=(F_1,\ldots,F_d):\mathbb{C}^d\to\mathbb{C}^d be a polynomial map. The statement under investigation asserts that if detJFC×\det J_F\in\mathbb{C}^{\times} is constant, then FF is injective and hence admits a polynomial inverse. A counterexample must satisfy detJFC×\det J_F\in\mathbb{C}^{\times} while admitting distinct points pqp\ne q with F(p)=F(q)F(p)=F(q).

Each problem is evaluated both without a hint and with a reference-direction hint.
P1
Exploration Level

Open Construction

Complex dimension 3 · no degree bound

CapabilityNovel algebraic construction

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3. No degree or sparsity restriction is imposed.

ℂ³ polynomial map det J_F ∈ ℂ× ≥ 2-point collision Novelty reported separately
Why it matters

Tests invention together with exact certificate construction.

Full problem prompt

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3. No degree or sparsity restriction is imposed.

Hint & boundary

Let u=1+xyu=1+xy. A known three-dimensional construction is
F1=u3z+y2u(4+3xy),F2=y+3xu2z+3xy2(4+3xy),F3=2x3x2yx3z.F_1=u^3z+y^2u(4+3xy),\qquad F_2=y+3xu^2z+3xy^2(4+3xy),\qquad F_3=2x-3x^2y-x^3z.
It satisfies detJF=2\det J_F=-2, and
(0,0,14), (1,32,132), (1,32,132)(0,0,-\tfrac14),\ (1,-\tfrac32,\tfrac{13}{2}),\ (-1,\tfrac32,\tfrac{13}{2})
all map to (14,0,0)(-\tfrac14,0,0). The submission must be a genuinely new counterexample algebraically inequivalent to this reference. Reformulations, variable or component permutations, reparametrizations, and constructions obtainable by invertible algebraic coordinate changes in the source or target are inadmissible.

01 / 05
P2
Constrained Level

Degree-Seven Rediscovery

Complex dimension 3 · degree ≤ 7

CapabilityConstraint-aware rediscovery

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3 with maxidegFi7\max_i\deg F_i\le 7. Supply at least two distinct algebraic points p1,p2p_1,p_2 in one common fiber, so that F(p1)=F(p2)F(p_1)=F(p_2).

ℂ³ polynomial map deg F ≤ 7 ≥ 2-point collision Novelty reported separately
Why it matters

A controlled target tests whether a model can reconstruct the known mechanism.

Full problem prompt

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3 with maxidegFi7\max_i\deg F_i\le 7. Supply at least two distinct algebraic points p1,p2p_1,p_2 in one common fiber, so that F(p1)=F(p2)F(p_1)=F(p_2).

Hint & boundary

Let u=1+xyu=1+xy. A known construction is
F1=u3z+y2u(4+3xy),F2=y+3xu2z+3xy2(4+3xy),F3=2x3x2yx3z.F_1=u^3z+y^2u(4+3xy),\qquad F_2=y+3xu^2z+3xy^2(4+3xy),\qquad F_3=2x-3x^2y-x^3z.
Its component degrees are (7,6,4)(7,6,4), detJF=2\det J_F=-2, and the three points (0,0,14)(0,0,-\tfrac14), (1,32,132)(1,-\tfrac32,\tfrac{13}{2}), and (1,32,132)(-1,\tfrac32,\tfrac{13}{2}) share the image (14,0,0)(-\tfrac14,0,0). The submission must be a genuinely new counterexample algebraically inequivalent to this reference. Reformulations and constructions obtainable by invertible algebraic coordinate changes are inadmissible.

02 / 05
P3
Research Level

Lower-Degree Frontier

Complex dimension 3 · improve degree 7

CapabilityRecord-level degree reduction

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3 whose polynomial degree degF:=maxidegFi\deg F:=\max_i\deg F_i is strictly smaller than the currently known minimum 77. Supply at least two distinct algebraic points p1,p2p_1,p_2 in one common fiber, so that F(p1)=F(p2)F(p_1)=F(p_2).

ℂ³ polynomial map deg F < 7 ≥ 2-point collision Record target
Why it matters

Beating degree seven would expose a simpler counterexample geometry.

Full problem prompt

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3 whose polynomial degree degF:=maxidegFi\deg F:=\max_i\deg F_i is strictly smaller than the currently known minimum 77. Supply at least two distinct algebraic points p1,p2p_1,p_2 in one common fiber, so that F(p1)=F(p2)F(p_1)=F(p_2).

Hint & boundary

For reference only, let u=1+xyu=1+xy and consider
F1=u3z+y2u(4+3xy),F2=y+3xu2z+3xy2(4+3xy),F3=2x3x2yx3z.F_1=u^3z+y^2u(4+3xy),\qquad F_2=y+3xu^2z+3xy^2(4+3xy),\qquad F_3=2x-3x^2y-x^3z.
This map has detJF=2\det J_F=-2, three known collision points, and degF=7\deg F=7. The task is to improve the currently known minimum, so the reference construction itself does not qualify.

03 / 05
P4
Research Level

Four-Sheet Frontier

Generic fiber degree 4 · degree ≤ 11

CapabilityGeneric-fiber optimization

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3 with generic fiber degree deggenF=4\deg_{\mathrm{gen}}F=4 and polynomial degree degF:=maxidegFi11\deg F:=\max_i\deg F_i\le 11. The currently known construction with generic fiber degree 44 has polynomial degree 1212; improve this bound and supply four pairwise-distinct algebraic points p1,p2,p3,p4p_1,p_2,p_3,p_4 in one common fiber.

Generic fiber degree = 4 target deg F ≤ 11 4-point witness Exact degree reported separately
Why it matters

Tests joint control of fiber geometry and algebraic complexity.

Full problem prompt

Construct a counterexample F:C3C3F:\mathbb{C}^3\to\mathbb{C}^3 with generic fiber degree deggenF=4\deg_{\mathrm{gen}}F=4 and polynomial degree degF:=maxidegFi11\deg F:=\max_i\deg F_i\le 11. The currently known construction with generic fiber degree 44 has polynomial degree 1212; improve this bound and supply four pairwise-distinct algebraic points p1,p2,p3,p4p_1,p_2,p_3,p_4 in one common fiber.

Hint & boundary

For structural reference only, the first known three-dimensional map
F1=(1+xy)3z+y2(1+xy)(4+3xy),F2=y+3xz(1+xy)2+3xy2(4+3xy),F3=2x3x2yx3zF_1=(1+xy)^3z+y^2(1+xy)(4+3xy),\quad F_2=y+3xz(1+xy)^2+3xy^2(4+3xy),\quad F_3=2x-3x^2y-x^3z
satisfies detJF=2\det J_F=-2 and has three supplied collision points mapping to (14,0,0)(-\tfrac14,0,0). It is not a solution to this task: the target is a four-sheeted generic fiber with degF11\deg F\le 11, improving the currently known degree-1212 construction.

04 / 05
P5
Open-Problem Level

Two-Dimensional Frontier

Complex dimension 2 · no degree bound

CapabilityOpen-problem discovery

Construct a counterexample F:C2C2F:\mathbb{C}^2\to\mathbb{C}^2. No degree or sparsity restriction is imposed.

ℂ² polynomial map det J_F ∈ ℂ× ≥ 2-point collision Open frontier
Why it matters

Dimension two is the remaining unresolved Jacobian frontier.

Full problem prompt

Construct a counterexample F:C2C2F:\mathbb{C}^2\to\mathbb{C}^2. No degree or sparsity restriction is imposed.

Hint & boundary

For reference only, the known three-dimensional map
F1=(1+xy)3z+y2(1+xy)(4+3xy),F2=y+3xz(1+xy)2+3xy2(4+3xy),F3=2x3x2yx3zF_1=(1+xy)^3z+y^2(1+xy)(4+3xy),\quad F_2=y+3xz(1+xy)^2+3xy^2(4+3xy),\quad F_3=2x-3x^2y-x^3z
has detJF=2\det J_F=-2 and three collision points. Because it is a map on C3\mathbb{C}^3, it is only a structural reference for this two-dimensional task.

05 / 05

Novelty and exact generic-fiber degree are reported separately when they are not machine-verifiable; neither is delegated to an LLM judge.

03 / EVALUATION

Exact outcomes across tasks, models, and hint modes

Outcome types are derived from deterministic verifier fields, while transport failures are separated from mathematical failures.

50 records
Offline pass rate2.0%1 / 50
Output parsed74%37 records
Mathematically valid2.0%1 records
Total inference time25.8 hverification 59.4 s
Total tokens992.3Kreasoning 692.3K
Outcome type statisticsDeterministic attribution · no LLM judge
Output format failedA candidate was present, but it did not match the declared machine-readable schema.21
API or model response failureNo evaluable response was produced because the request or upstream service failed.9
Jacobian condition failedThe determinant is not a nonzero constant.7
Collision certificate failedThe submitted points do not share a common image.6
No evaluable certificate submittedThe required structured certificate block is absent.4
Collision points are not distinctRepeated inputs cannot establish non-injectivity.1
Certificate failed deterministic verificationThe offline program could not confirm all required conditions.1
Verified counterexample (research qualifiers reported separately)The finite algebraic certificate and machine-checkable task constraints passed.1
Model × problem outcome matrixEach cell: NO HINT on the left; HINT on the right.
verified constraint miss mathematical error protocol error response failureNO HINTHINT · KNOWN COUNTEREXAMPLE PROVIDED
MODELdeterministic outcome
P1Open Construction
P2Degree-Seven Rediscovery
P3Lower-Degree Frontier
P4Four-Sheet Frontier
P5Two-Dimensional Frontier
Detailed model comparisonPasses are separated by hint condition; averages use the active filters.
ModelRecordsPassPass rateNO HINTHINTParsedAvg tokensAvg time
P1HINT ONRUN 01

Claude-Opus-4.8-Thinking

temperature 1 · top_p 0.95 · max_tokens 128,000

OFFLINEFAIL
Inference 6.8 minVerification 1.0 s38.1K tokens0 symbolic ops
DETERMINISTIC OUTCOME ATTRIBUTIONOutput format failed

A candidate was present, but it did not match the declared machine-readable schema.

certificate has unsupported keys: ['reason']
certificate has unsupported keys: ['reason']
results/jacobian_conjecture/claude-opus-4-8-thinking/jacobian_conjecture_1_hint_01.jsonSource result
Loading record…

An OFFLINE PASS means that the submitted finite object satisfies every machine-checkable condition declared for this task. Research qualifications outside deterministic verification are reported separately; model text is retained for auditability and is not endorsed by this site.

04 / SYMBOLIC LAB

Test a candidate counterexample interactively

Enter a three-variable polynomial map and explicit collision points. The browser checks the stated Jacobian and collision constraints locally; it does not use an LLM judge or assess novelty.

Deterministic constraint checks onlyInputs stay in the browser. No LLM judge and no novelty judgment.
NO LLM JUDGENO NOVELTY JUDGMENTOffline verifier
3-variable polynomial verifier
01
Polynomial mapUse x, y, z, i; write multiplication with *
02
Candidate collision pointsFractions and complex values are supported, e.g. 1/2 and 2*i
p1
p2
p3
03
Deterministic resultIn-browser symbolic differentiation and numerical substitution

Edit the expressions, then compute the Jacobian matrix and collision result.

Scope note: this interactive tool accepts rational and real coefficients plus explicit complex i syntax. The benchmark verifier accepts more general algebraic-coefficient certificates. Browser collision checks use a 10⁻⁹ numerical tolerance; reported scores always come from the exact offline program.

05 / COMMUNITY

Keep the questions worth pursuing

Share a missing constraint, a counterexample direction, or a possible verification gap. Every message is moderated before publication; durable questions can inform the next benchmark release.

01Interactive submissionMarkdown research note
02Secure recordUTC timestamp · abuse controls
03Strong-model reviewSafety · relevance · category
04Human decisionExplicit approval required
05Categorized publicationOnly approved messages

Public messages

Only approved messages are shown

Community backend unavailable

The mathematics and benchmark results remain available. This panel reconnects automatically when the deployment backend returns.

SUBMIT A QUESTION

Submit a research note

MessageMarkdown and LaTeX are supported

The server records an exact UTC timestamp and applies origin, payload, rate, fingerprint, duplicate, honeypot, and content checks. Contact email stays private. AI never publishes automatically; human approval is mandatory.

06 / REFERENCES

Sources and related work for the Jacobian Conjecture

The benchmark is built on a much larger mathematical record. We thank the authors, editors, maintainers, and institutions who made these works and specialist resources available.

  1. 01

    Ganze Cremona-Transformationen

    Paper1939Monatshefte für Mathematik und Physik
    Ott-Heinrich Keller

    The original paper that formulated the polynomial Jacobian problem now bearing Keller's name.

  2. 02

    The Jacobian Conjecture: Reduction of Degree and Formal Expansion of the Inverse

    Paper1982Bulletin of the American Mathematical Society
    Hyman Bass, Edwin H. Connell, David Wright

    The foundational reduction showing why special cubic-homogeneous maps capture the general conjecture.

  3. 03

    An Effective Approach to Keller's Jacobian Conjecture

    Paper1983Mathematische Annalen
    Ludwik M. Drużkowski

    A further reduction to power-linear cubic maps that continues to shape computational searches.

  4. 04

    Polynomial Automorphisms and the Jacobian Conjecture

    Book2000Birkhäuser, Progress in Mathematics 190
    Arno van den Essen

    A self-contained monograph and extensive reference for polynomial automorphisms and the conjecture.

  5. 05

    Jacobian Conjecture bibliography and problem page

    Website
    Tzuong-Tsieng Moh · Purdue University

    A long-running specialist bibliography and historical entry point to the Jacobian Conjecture literature.

07 / GLOBAL REACH BACKEND OFFLINE

A growing, global research audience

Cumulative visits are recorded by the first-party Cloudflare backend and aggregated only by country. Raw IP addresses are never stored or displayed.

0cumulative visits

The first country-level visit will light up the map.

Fiji: 0Tanzania: 0W. Sahara: 0Canada: 0United States of America: 0Kazakhstan: 0Uzbekistan: 0Papua New Guinea: 0Indonesia: 0Argentina: 0Chile: 0Dem. Rep. Congo: 0Somalia: 0Kenya: 0Sudan: 0Chad: 0Haiti: 0Dominican Rep.: 0Russia: 0Bahamas: 0Falkland Is.: 0Norway: 0Greenland: 0Fr. S. Antarctic Lands: 0Timor-Leste: 0South Africa: 0Lesotho: 0Mexico: 0Uruguay: 0Brazil: 0Bolivia: 0Peru: 0Colombia: 0Panama: 0Costa Rica: 0Nicaragua: 0Honduras: 0El Salvador: 0Guatemala: 0Belize: 0Venezuela: 0Guyana: 0Suriname: 0France: 0Ecuador: 0Puerto Rico: 0Jamaica: 0Cuba: 0Zimbabwe: 0Botswana: 0Namibia: 0Senegal: 0Mali: 0Mauritania: 0Benin: 0Niger: 0Nigeria: 0Cameroon: 0Togo: 0Ghana: 0Côte d'Ivoire: 0Guinea: 0Guinea-Bissau: 0Liberia: 0Sierra Leone: 0Burkina Faso: 0Central African Rep.: 0Congo: 0Gabon: 0Eq. Guinea: 0Zambia: 0Malawi: 0Mozambique: 0eSwatini: 0Angola: 0Burundi: 0Israel: 0Lebanon: 0Madagascar: 0Palestine: 0Gambia: 0Tunisia: 0Algeria: 0Jordan: 0United Arab Emirates: 0Qatar: 0Kuwait: 0Iraq: 0Oman: 0Vanuatu: 0Cambodia: 0Thailand: 0Laos: 0Myanmar: 0Vietnam: 0North Korea: 0South Korea: 0Mongolia: 0India: 0Bangladesh: 0Bhutan: 0Nepal: 0Pakistan: 0Afghanistan: 0Tajikistan: 0Kyrgyzstan: 0Turkmenistan: 0Iran: 0Syria: 0Armenia: 0Sweden: 0Belarus: 0Ukraine: 0Poland: 0Austria: 0Hungary: 0Moldova: 0Romania: 0Lithuania: 0Latvia: 0Estonia: 0Germany: 0Bulgaria: 0Greece: 0Turkey: 0Albania: 0Croatia: 0Switzerland: 0Luxembourg: 0Belgium: 0Netherlands: 0Portugal: 0Spain: 0Ireland: 0New Caledonia: 0Solomon Is.: 0New Zealand: 0Australia: 0Sri Lanka: 0China: 0Italy: 0Denmark: 0United Kingdom: 0Iceland: 0Azerbaijan: 0Georgia: 0Philippines: 0Malaysia: 0Brunei: 0Slovenia: 0Finland: 0Slovakia: 0Czechia: 0Eritrea: 0Japan: 0Paraguay: 0Yemen: 0Saudi Arabia: 0Antarctica: 0N. Cyprus: 0Cyprus: 0Morocco: 0Egypt: 0Libya: 0Ethiopia: 0Djibouti: 0Somaliland: 0Uganda: 0Rwanda: 0Bosnia and Herz.: 0Macedonia: 0Serbia: 0Montenegro: 0Kosovo: 0Trinidad and Tobago: 0S. Sudan: 0
No visitsMore visits

Map geometry: Natural Earth / World Atlas · public domain