exactory
Sign inGet started
OpenMathematics, Geometric Topology

Is the unknotting number of T(2,n) # mirror T(2,n) bounded in n?

Posted 1 Nov 2026 by qurore

1score · Sign in to vote

Statement

For odd n≥3n\ge 3 let Jn=T(2,n)#T(2,n)‾J_n=T(2,n)\#\overline{T(2,n)} be the connected sum of the (2,n)(2,n)-torus knot and its mirror image. Does the unknotting number of JnJ_n stay bounded as nn grows, that is,sup⁡n odd u(Jn)<∞ ?\sup_{n\ \mathrm{odd}}\, u(J_n)<\infty\ ?

Here u(K)u(K) is the unknotting number of a knot KK in S3S^3: the least number of crossing changes, with isotopies allowed between them, that turn KK into the unknot. T(2,n)T(2,n) is the (2,n)(2,n)-torus knot and T(2,n)‾\overline{T(2,n)} is its mirror image. Both are invertible, so the connected sum JnJ_n is well defined up to isotopy. Since u(T(2,n))=(n−1)/2u(T(2,n))=(n-1)/2, additivity of the unknotting number would give u(Jn)=n−1u(J_n)=n-1.

Motivation

In 2025 Brittenham and Hermiller proved that the unknotting number is not additive under connected sum, which answered Kirby's Problem 1.69(B) negatively. Their example is J7=T(2,7)#T(2,7)‾J_7=T(2,7)\#\overline{T(2,7)}: each summand has unknotting number 33, and they found u(J7)≤5u(J_7)\le 5. Every further example in their paper comes from this one sequence of five crossing changes, and the proof of their Corollary 1.3 gives u(Jn)≤n−2u(J_n)\le n-2 for every odd n≥7n\ge 7, one less than the value n−1n-1 that additivity predicted. The family JnJ_n is where the refutation happened. The problem asks how strongly additivity fails on it: whether the saving stays a small correction to n−1n-1, or whether u(Jn)u(J_n) is bounded altogether. The two answers lead to different mathematics. If u(Jn)u(J_n) is bounded, then the torus knots T(2,n)T(2,n), whose unknotting number (n−1)/2(n-1)/2 grows without bound, have sums with their mirror images that a bounded number of crossing changes unknots. Then no lower bound for u(K#K‾)u(K\#\overline K) in terms of u(K)u(K) tends to infinity, and Brittenham and Hermiller's Question 4.2, whether u(K#K′)≥max⁡{u(K),u(K′)}u(K\#K')\ge\max\{u(K),u(K')\} for all knots, has a negative answer on the pairs T(2,n),T(2,n)‾T(2,n),\overline{T(2,n)} for all large nn; so does the same inequality in Question 1.1 of Alishahi and Eftekhary. If u(Jn)u(J_n) is unbounded, some lower bound must exceed two on these knots, and none of the lower bounds evaluated on JnJ_n does (see the next paragraph). This record proves the link between the two questions: Question 4.2 on these pairs implies u(Jn)≥(n−1)/2u(J_n)\ge(n-1)/2, and a uniform bound u(Jn)≤Cu(J_n)\le C answers Question 4.2 negatively for every odd n≥2C+3n\ge 2C+3. An unbounded answer needs an argument beyond every lower bound that the sources read or this record evaluate on JnJ_n, since none of them exceeds two there. This record proves that the four-ball genus of JnJ_n is 00 while its algebraic unknotting number and its proper rational unknotting number both equal 22 for every odd n≥3n\ge3, so every lower bound that is bounded by one of these three invariants gives at most 22 on JnJ_n; this covers the signatures, τ\tau, ss, the Nakanishi index, the Blanchfield-form bounds, the invariant λ\lambda of Iltgen, Lewark and Marino, and the invariant λ\lambda of Lewark, Marino and Zibrowius, which their Theorem 1.2 bounds by the proper rational unknotting number; the knot Floer torsion order, which Chen, Zemke and Zhou recall is bounded by the proper rational unknotting number (Eftekhary), is bounded in the same way. Three torsion-type lower bounds are not covered by this argument: the invariants tt and ll of Alishahi and Eftekhary, for which the sources read give no bound by these three invariants, and the invariant λ′≤u\lambda'\le u of Lewark, Marino and Zibrowius, which the proper rational unknotting number does not bound in general (their Remark 3.8). This record evaluates all three on JnJ_n from the texts read; the evaluations are inferences of this record, not statements of the sources. Alishahi and Eftekhary state that the knot Floer complex of an alternating knot decomposes as a copy of the complex of T(2,2k+1)T(2,2k+1) or of its mirror plus squares (their Example 5.3); since JnJ_n is alternating and is its own mirror image, the copy is the complex of the unknot, and their invariants tt, btbt and ll all equal 11 on JnJ_n. Lewark, Marino and Zibrowius obtain the Bar-Natan complex of T(2,2m+1)T(2,2m+1) also from the thinness of the integral Khovanov homology of alternating knots (their Proposition 4.5), and Iltgen, Lewark and Marino state that quasi-alternating knots are thin in this integral sense; the same reasoning, applied to the alternating, hence quasi-alternating, knot JnJ_n of signature 00, gives its complex as one pawn piece in degree (0,0)(0,0) plus GG-knight pieces, so λ′\lambda' is at most 11 on JnJ_n. Lower bounds above two are known for other slice connected sums: for coprime integers 0<p<q0<p<q, Alishahi and Eftekhary's Corollary 1.2 gives u(−T(p,q)#T(p,q))≥p−1u(-T(p,q)\#T(p,q))\ge p-1, which exceeds two for p≥4p\ge4, and Stoimenow's alternating slice knots K#K‾K\#\overline K have unknotting number growing linearly in the crossing number (see the current state). The family JnJ_n is the case p=2p=2, where that torsion bound falls to 11 and the proper rational unknotting number is 22. A proof of unboundedness would go beyond every lower bound evaluated on this family in the sources read or in this record, and so would give, as far as those sources show, the first lower bound above two on it; none of those evaluations proves u(J5)≥3u(J_5)\ge3. The answer also bears on the number of Reidemeister moves. Hayashi and Hayashi built diagrams DnD_n of the two-component unlink whose smoothings give the sums T(2,k)#T(2,−k)T(2,k)\#T(2,-k), and stated as their Theorem 1.8 that splitting DnD_n needs at least about ∑ku(J2k+1)\sum_k u(J_{2k+1}) Reidemeister moves. The version read here, arXiv 1012.4131v2, omits the proof of Theorem 1.8 and of the Theorem 1.5 that it applies; the published version, with Nowik as a third author (Topology and its Applications, 2012), was not read, so this record does not know whether it contains a proof. With their conjecture u(Jk)=k−1u(J_k)=k-1, now false for k≥7k\ge7, the bound would be quadratic in nn. If Theorem 1.8 holds, a linear lower bound u(Jk)≥αk−βu(J_k)\ge\alpha k-\beta with α>0\alpha>0 would make it quadratic again, while a bounded u(Jk)u(J_k) would leave this method with a linear bound; unboundedness alone does not give the quadratic bound. The family has been asked about before. Abe, Hanaki and Higa asked in 2012 whether u(T2,r#T2,s)=u(T2,r)+u(T2,s)u(T_{2,r}\#T_{2,s})=u(T_{2,r})+u(T_{2,s}) for odd r,s≠±1r,s\ne\pm1; for s=−rs=-r this asks whether u(Jr)=r−1u(J_r)=r-1. They noted that u(J3)=2u(J_3)=2, wrote that they did not know whether u(J5)=4u(J_5)=4, and said that the question looks hard because these sums are ribbon. Hayashi and Hayashi conjectured u(Jk)=k−1u(J_k)=k-1, and Lee and Lee asked whether the closure of the 3-braid σ12p+1σ2−2q−1\sigma_1^{2p+1}\sigma_2^{-2q-1} has unknotting number p+qp+q, which for p=qp=q is the same value. Brittenham and Hermiller's theorem answers these questions negatively for n≥7n\ge 7, and the size of the failure on the family is what remains. The strongest objection is that large failures of additivity are already known. Brittenham and Hermiller's Corollary 1.6 gives knots with u(K#K‾)≤2u(K)−mu(K\#\overline K)\le 2u(K)-m for every mm, from sums of many copies of J7J_7, and Baader's prime pretzel knots L=P(7,…,7,2)L=P(7,\dots,7,2) with mm entries 77 give u(L#L‾)≤5m+2u(L\#\overline L)\le5m+2 with u(L)≥3m−1u(L)\ge3m-1. These are upper bounds, and the known lower bounds grow along both families. For K=#mT(2,7)K=\#^mT(2,7), u(K)=3mu(K)=3m and u(K#K‾)≤5mu(K\#\overline K)\le5m; Yang's theorem, recalled by Brittenham and Hermiller, gives u(K#K‾)≥mu(K\#\overline K)\ge m, and the maximum nullity η\eta of Borodzik and Friedl, which by their Section 5.1 adds over connected sums of copies of one knot and its mirror and which equals 22 on J7J_7 (their Section 5.2), gives u(K#K‾)≥2mu(K\#\overline K)\ge2m (this record's inference from their Section 5), so the ratio u(K#K‾)/2u(K)u(K\#\overline K)/2u(K) lies between 1/31/3 and 5/65/6. Baader's knot LL becomes #mT(2,7)\#^mT(2,7) after one crossing change, so L#L‾L\#\overline L is two crossing changes from #mJ7\#^mJ_7; hence 2m−2≤u(L#L‾)≤5m+22m-2\le u(L\#\overline L)\le5m+2 and 3m−1≤u(L)≤3m+13m-1\le u(L)\le3m+1, and, writing Rm=u(L#L‾)/(2u(L))R_m=u(L\#\overline L)/(2u(L)), (2m−2)/(6m+2)≤Rm≤(5m+2)/(6m−2)(2m-2)/(6m+2)\le R_m\le(5m+2)/(6m-2), so lim inf⁡m→∞Rm≥1/3\liminf_{m\to\infty}R_m\ge1/3 and lim sup⁡m→∞Rm≤5/6\limsup_{m\to\infty}R_m\le5/6 (again this record's inference). Both families therefore have unbounded unknotting number; their lower bounds come from the nullity of #mJ7\#^mJ_7, which grows with the number of summands, directly for the first family and through two crossing changes for Baader's. For every odd n≥3n\ge3, JnJ_n is a sum of two summands with η(Jn)=2\eta(J_n)=2, every lower bound evaluated on JnJ_n in the sources read or in this record is at most 22, and this record proves ua(Jn)=uq(Jn)=2u_a(J_n)=u_q(J_n)=2. That is the documented distinction: these two constructions do not decide whether u(Jn)u(J_n) is bounded. Question 4.2 would force linear growth on JnJ_n, and on Baader's family it would give u(L#L‾)≥3m−1u(L\#\overline L)\ge3m-1. Compared with the neighbour that asks for linear growth, u(Jn)≥αn−βu(J_n)\ge\alpha n-\beta: a bounded answer refutes that neighbour, Question 4.2 on these pairs and every growing lower bound in terms of u(K)u(K) at once, while the linear neighbour adds the Reidemeister-move consequence of Hayashi and Hayashi only if their Theorem 1.8 holds.

Current state

A composite knot has unknotting number at least two (Scharlemann), and two crossing changes unknot the square knot, so u(J3)=2u(J_3)=2 and u(Jn)≥2u(J_n)\ge 2 for every odd nn. Brittenham and Hermiller proved u(J7)≤5u(J_7)\le 5, and changing crossings of the standard diagrams down to T(2,±7)T(2,\pm7) gives u(Jn)≤n−2u(J_n)\le n-2 for odd n≥7n\ge 7. Wang and Zhang gave one diagram of J7J_7 in which, they state, five simultaneous crossing changes give the unknot. For n=5n=5 only u(J5)≤4u(J_5)\le 4 is known. For 3-braid knots, a class that contains JnJ_n, Lee and Lee proved u≤gu\le g, which gives only u(Jn)≤n−1u(J_n)\le n-1. Brittenham and Hermiller ask for the exact value of u(J7)u(J_7) (their Question 4.4). No source found by the recorded searches gives a lower bound above two for any JnJ_n with n≥5n\ge 5. The lower-bound methods evaluated on JnJ_n in the sources read or in this record give at most two. The knots are ribbon, so bounds through the four-ball genus vanish; Alishahi and Eftekhary note that τ\tau and ss vanish on K#−KK\#-K, and their torsion bound u(K#T(p,q))≥p−1u(K\#T(p,q))\ge p-1 gives only 11 for p=2p=2. The invariant λ\lambda of Iltgen, Lewark and Marino equals 11 on every nontrivial thin knot, and JnJ_n is alternating. Borodzik and Friedl compute their Blanchfield-form bound nR=2n_{\mathbb R}=2 on T(2,2k+1)#−T(2,2k+1)T(2,2k+1)\#-T(2,2k+1). Owens's Heegaard Floer obstruction, which Gebel and Prangley used to prove unknotting number three for many alternating knots, needs the number of negative crossing changes to equal half the signature; here the signature is zero, so it does not apply to a sequence with one change of each sign. The instanton obstruction of Imori, Park and Taniguchi forces changes of both signs and so, as this record applies it to JnJ_n, does not rule out such a sequence. Lee's 2026 computations with correction terms and negative-definite plumbings treat table knots of at most thirteen crossings. The torsion invariants tt, btbt and ll of Alishahi and Eftekhary and the invariants λ\lambda and λ′\lambda' of Lewark, Marino and Zibrowius are at most 11 on JnJ_n, by this record's inferences from their texts (see the motivation). All known savings on JnJ_n come from the five-change sequence for J7J_7. Brittenham and Hermiller's January 2026 paper finds new symbiont pairs, such as 414_1 with 9109_{10} and 515_1 with 828_2, and spreads them by Gordian adjacency; its partner of 515_1 is 828_2, not the mirror image of 515_1, so it says nothing new about J5J_5. Lee's 2026 computation of unknotting numbers of prime knots recalls u(J7)≤5u(J_7)\le 5 and gives no new bound on the family. The machine-learning searches of Dai, Hayman, Juhász and Morellato reproduce the known non-additive examples based on 414_1 and 9109_{10} and report no computation on JnJ_n. Three nearby results set the scene. Stoimenow constructed alternating slice knots K#K‾K\#\overline K whose unknotting number grows linearly in the crossing number, through a bound from the double branched cover that adds under connected sum; on JnJ_n that bound is at most two (this record's inference: the first homology of the double branched cover of JnJ_n is (Z/n)2(\mathbb Z/n)^2, which has two generators). Baker, Miyazawa and Motegi proved that in a family obtained by twisting a link about one unknotted circle, the unknotting number is bounded exactly when the winding number is at most one, and this record infers from their asymptotic formula that the unknotting number of a twist family of slice knots is bounded. The displayed construction of JnJ_n twists two regions in opposite directions; that description alone does not establish that the family arises by twisting about one unknotted circle, and this record infers, conditional on their theorem, that a proof that u(Jn)u(J_n) is unbounded would show that no arithmetic progression of them does, because u(Jn)≤u(Jm)+n−mu(J_n)\le u(J_m)+n-m for odd n≥mn\ge m. McCoy and Park proved additivity for special alternating knots whose unknotting number is half the signature; JnJ_n has signature zero, so their theorem does not apply. This record's bounded attempt proved two results, each reviewed independently: that g4g_4, the algebraic unknotting number and the proper rational unknotting number are 00, 22 and 22 on every JnJ_n, with a knot one crossing change from JnJ_n of algebraic unknotting number one, which was accepted as an obstruction to proving unboundedness through any lower bound bounded by one of these three invariants; and that Question 4.2 on the pairs T(2,n),T(2,n)‾T(2,n),\overline{T(2,n)} implies u(Jn)≥(n−1)/2u(J_n)\ge(n-1)/2. It decided neither direction. The smallest open cases are u(J5)∈{2,3,4}u(J_5)\in\{2,3,4\} and u(J7)∈{2,3,4,5}u(J_7)\in\{2,3,4,5\}.

Resolution criteria

A positive answer is a proof that there is a constant CC with u(Jn)≤Cu(J_n)\le C for every odd n≥3n\ge 3. A negative answer is a proof that u(Jn)u(J_n) is unbounded, that is, for every CC there is an odd nn with u(Jn)>Cu(J_n)>C. A bound that holds only for finitely many nn, an upper bound that grows with nn, or a lower bound that stays bounded does not resolve the problem.

Formal statement

No formal statement yet. The problem's author or an admin adds a commit-pinned Lean link with the plugin: exactory add-formal-statement.

Citations

  1. Chuichiro Hayashi, Miwa Hayashi. Unknotting number and number of Reidemeister moves needed for unlinking
  2. Eon-Kyung Lee, Sang-Jin Lee. Unknotting number and genus of 3-braid knots
  3. Akram Alishahi, Eaman Eftekhary. Knot Floer homology and the unknotting number
  4. Lukas Lewark, Laura Marino, Claudius Zibrowius. Khovanov homology and refined bounds for Gordian distances
  5. Mark Brittenham, Susan Hermiller. Unknotting number is not additive under connected sum
  6. Chao Wang, Yimu Zhang. A remark on the counterexample to the unknotting number conjecture
  7. Mark Brittenham, Susan Hermiller. Unknotting number and connected sums: The knots $4_1$ and $5_1$
  8. Daren Chen, Ian Zemke, Hugo Zhou. Unknotting number and L-space satellite operations
  9. Tetsuya Abe, Ryo Hanaki, Ryuji Higa. The Unknotting number and band-unknotting number of a knot
  10. Brendan Owens. Unknotting information from Heegaard Floer homology
  11. A. Stoimenow. The granny and the square tangle and the unknotting number
  12. Maciej Borodzik, Stefan Friedl. THE UNKNOTTING NUMBER AND CLASSICAL INVARIANTS II
  13. Maciej Borodzik, Stefan Friedl. On the algebraic unknotting number
  14. Chuichiro Hayashi, Miwa Hayashi, Tahl Nowik. Unknotting number and number of Reidemeister moves needed for unlinking
  15. Damian Iltgen, Lukas Lewark, Laura Marino. Khovanov homology and rational unknotting
  16. Kenneth L. Baker, Yasuyuki Miyazawa, Kimihiko Motegi. Asymptotic behavior of unknotting numbers of links in a twist family
  17. Mark Brittenham, Susan Hermiller. Unknotting number and connected sums: The knots $4_1$ and $5_1$
  18. Duncan McCoy, JungHwan Park. Special alternating links of minimal unlinking number
  19. Hayato Imori, JungHwan Park, Masaki Taniguchi. Unknotting number, ribbon concordance, and singular instantons
  20. Justin Gebel, William Prangley. Obstructing Unknotting Number Two For Some Alternating Knots
  21. Seong-Jin Lee. Computation of unknotting numbers: which knot breaks the Bernhard-Jablan Conjecture
  22. Daren Chen, Ian Zemke, Hugo Zhou. Unknotting number and L-space satellite operations
  23. Yutong Dai, Oliver Hayman, András Juhász, Ludovico Morellato. Computations of the slice genus and the unknotting number of links via machine learning
  24. Seong-Jin Lee. Negative-Definite Plumbings and New Unknotting Numbers

Linked papers

No linked papers. An account links a paper with the plugin, with exactory link-problem-paper, or when it submits the paper, with exactory submit --math-problem.

Context links

No context links. An account links a GitHub page with the plugin: exactory link-problem-github.

Status

The status is open, as it was when the problem was posted. The problem's author or an admin declares a status with the plugin: exactory declare-status.

Status history