WorksMathematics
Determinants of amphichiral knots of braid index at most three
Abstract
We show that an odd number n is the determinant of an amphichiral knot of braid index at most three if and only if both n and n + 4 are sums of two squares. The proof rests on the known description of amphichiral knots of braid index three as closures of alternating 3-braid words with a rotation or reflection symmetry (Birman–Menasco, Stoimenov, Jung). Determinant formulas for closed 3-braids are also known (e.g. through the reduced Burau representation at t = -1); what is added here is their evaluation on the two symmetric families, which gives the condition on n + 4, and an explicit construction for the converse. For amphichiral knots of braid index three this describes which determinants occur, which bears on the determinant part of a question of Jung. We did not find the statement in the literature we searched. As a consequence, for an odd n with n + 4 not a sum of two squares, every amphichiral knot with determinant n has braid index at least five; we test this on 4,999 amphichiral knots of Burton's census with up to 19 crossings whose determinant satisfies this hypothesis.
Statement
A knot is amphichiral if it is isotopic to its mirror image, amphichiral if it is isotopic to the reverse of its mirror image, and amphichiral if one of the two holds. The determinant of a knot is odd, and for every amphichiral knot it is a sum of two squares [8 Remark (7), via Goeritz].
Theorem 1. .
Here 0 counts as a square (so 5, the determinant of the figure-eight knot, belongs to the right-hand side: , ). Among the odd sums of two squares the first ones excluded by the condition on are
This work was developed in collaboration with the AI system Claude Code (Anthropic): the choice of problem, the direction of the work and the computing hardware are due to the author (a sculptor); the formalisation, the algorithms and the proofs were developed by Claude Code (Anthropic) working with the author. The collaboration is stated as a fact, not a disclaimer; details are in the section on tool use at the end. 17, 29, 53, 65, 73. Braid index 1 and 2 contribute only the unknot (), since the torus knots , , are chiral.
Corollary 2. If is odd and is not a sum of two squares, every amphichiral knot with determinant has braid index at least 5.
Proof. By Theorem 1 the braid index is at least 4, and the braid index of an amphichiral knot is odd [9 Theorem A] (the proof is one line: by the theorem of Dynnikov and Prasolov [5 Theorem 9] (the Jones conjecture) all minimal braids of a knot have the same exponent sum, so a minimal braid of an amphichiral knot has exponent sum 0 [9 Lemma 3.3 and eq. (1)], and the parity of the permutation of a braid whose closure is a knot then makes the number of strands odd). □
Jung asks [9 Question 2] whether invariants such as the determinant can be “detected or computed from the braid word”. For amphichiral knots of braid index three, Lemmas 4 and 5 below compute the determinant from the word, and Theorem 1 describes the values that occur. (The author of [13, 14, 15, 16, 17] is the same person; we keep the spelling of each publication.)
Conventions and what is used from the literature
, , . For a sequence of positive integers, even, the alternating word of is . Let , , and . These are the matrices of the reduced Burau representation at (, , so and ). For a closed 3-braid knot with Burau matrix at one has : Datta [4 §5] states for the closure of any 3-braid, its writhe and the trace of its matrix (with our convention), and it also follows from ; we checked it against SnapPy on random 3-braid knots. For all entries are nonnegative and , so and . General determinant formulas for closed 3-braids are in [3 Theorem 3.1, from Qazaqzeh–Chbili]. (indices mod ) and .
Types. is of positive type if for some odd , and of negative type if for all for some odd (equivalently, is rotated by an even amount).
From the literature. (i) Birman and Menasco classified closed 3-braids [1]: every conjugacy class of contains a Schreier normal form, unique up to cyclic permutation of the syllables after , either (, “generic”) or one of , , , [1 pp. 104–105]; oriented links of braid index are determined by the conjugacy class of a 3-braid representative up to the unknot, the torus links of type and the flype case, in which a word and its reverse represent the same link [1 pp. 76–77, 106–107]. Their Corollary 7.1 B [1 p. 107] describes the amphichiral generic closures by the condition that agrees with the array up to an even rotation, i.e. (positive type), and ; it has no flype clause—, the closure of , is amphichiral but only of negative type—and does not treat amphichirality. (ii) Amphichiral 3-braid links. Stoimenov [13 Theorem 1.1(a)] proves that an orientedly amphicheiral 3-braid link (positively or negatively [13 Definition 2.2]) is the closure of an alternating 3-braid whose Schreier vector admits a “dihedral (anti)symmetry”, i.e. can be turned into by cyclic permutation and inversion [13 p. 448]; a rotation by changes all signs iff is odd, and a reflection does iff is odd, so these are exactly our two types. His proof bypasses the method of [1] (his Remark 3.2; it goes back to [16 Theorem 6.2]), relying on [1] only for where flypes can occur (for the equality of writhes he cites [1] or, alternatively, his own earlier work). For prime knots, Jung [9 Theorems 4.13 and B, the symmetry type in Theorem C] gives the same description, with the flype case ; in his terms positive type is “odd period” and negative type “palindrome” [9 §4, eq. (2)]. The two families, with the proof that their closures are amphichiral (our Lemma 3), also appear (for words with at least two pairs of syllables) as family (3) in Lisca’s theorem on 3-braid knots of finite concordance order [11 Theorem 1.1]; he does not consider which determinants occur. That the alternating half follows from [1] is also noted in [15 Prop. 7.1(2)]. For completeness we give the short derivation from [1] in Proposition 7.
Lemmas
Lemma 3 (symmetry). Up to conjugation, the mirror image followed by conjugation with sends the alternating word of to that of , and reading the word backwards sends it to that of . Rotation of by an even amount is conjugation. Hence a closure of an alternating word of positive type is amphichiral and one of negative type is amphichiral.
Proof. The mirror image of is ; conjugation by exchanges and , giving , a cyclic permutation of the alternating word of . Reading backwards gives , a cyclic permutation of the alternating word of ; its closure is the knot with reversed orientation (rotate the closed braid by about a line that meets the braid axis at a right angle: the letters keep their signs and indices and their cyclic order is reversed). If with odd, then with even, so the mirror image is isotopic to the knot. If is rotated by an even amount, the mirror image and the reverse are closures of conjugate words.
Lemma 4 (negative type). If is of negative type and the closure of its alternating word is a knot, it has and for a nonnegative integer matrix with .
Proof. Let with odd, , and . Then satisfies , i.e. with . The cyclically rotated product of starting at index equals if is even and if is odd (since ); in both cases . Let with for even and for odd. Since and the indices and have different parity, is the product of the remaining factors of , so . is a product of nonnegative unimodular matrices, so and . Hence and .
Lemma 5 (positive type). If is of positive type and the closure of its alternating word is a knot, its determinant is a sum of two squares, and is a sum of two squares.
Proof. Let with odd. The self-rotations of form a subgroup of , so is a self-rotation; is odd and with even. Let be the product of the first factors of . The -th block of factors starts at index , which has the parity of , so the blocks are alternately and , and with . Write with odd, , (), , and . From : and for . Hence equals for , equals for , and for , by induction using and ,
so . Finally for and for .
Lemma 6 (converse). If is odd and and are sums of two squares, there is a sequence of negative type whose alternating word closes to a knot with determinant .
Proof. Write , with . Each representation has one odd and one even component; let be the odd ones. If , let be the odd components (then and ); otherwise let be the even ones (then and ). Put
, , , . These are nonnegative integers with , (in both cases : in the first , in the second is odd), so , and (as ).
Row dominance. A nonnegative integer matrix with has one row entrywise the other: otherwise either , and , which equals 1 only for ; or , and ; and both rows dominating each other gives . If the top row dominates, , otherwise , with nonnegative, and entry sum smaller by resp. , which is ; so the process terminates at , and is a product of factors and . Collecting equal neighbours, with all ; it starts with because . Then is of negative type with as in Lemma 4, so its determinant is .
Knot. is odd, so has order 3 in ; since and modulo 2 send the generators to two distinct transpositions, this is the permutation of the braid up to an automorphism of , and it is a 3-cycle. □
Proof of Theorem 1
Proposition 7 ([13 Thm. 1.1(a)]; for prime knots [9 Theorems 4.13 and B]). An amphichiral knot of braid index 3 is the closure of an alternating word of positive or negative type.
Derivation from [1]. Let be the Schreier normal form of a 3-braid representing . The closures of and are not knots (the permutation is not a 3-cycle), and , close to torus knots , , which have braid index for ; for they cannot be amphichiral of braid index 3: would be a minimal braid with exponent sum resp. , while a minimal braid of an amphichiral knot has exponent sum 0 (by Dynnikov–Prasolov [5 Theorem 9], as in [J, Lemma 3.3 and eq. (1)]). (They are chiral in any case; their signatures were computed by Hirzebruch, Murasugi and Shinohara, as cited in [6 §2].) So $W=C^k\sigma_1^{-p_1}\sigma_2^{q_1}\cdots\sigma_1^{-p_r}\sigma_2^{q_r}$ with array $A=(p_1,q_1,\ldots,p_r,q_r)$. Let $W^*$ be the mirror word and $W^\leftarrow$ the word read backwards. If $K$ is $+$amphichiral, $W^*$ and $W$ represent the same oriented knot; if $-$amphichiral, $W^*$ and $W^\leftarrow$ do. By the classification {citep}[1 pp. 76–77, 106–107] (braid index 3; in the flype case a word and its reverse represent the same knot) this gives $W^*\sim W$ or $W^*\sim W^\leftarrow$ in both cases. The normal form of $W^*$ is $C^{-k}$ with array $\operatorname{rot}_1(A)$ {citep}[1 p. 107]`, that of is with array (reading backwards gives , cyclically ), and two generic normal forms are conjugate iff the powers of agree and the arrays agree up to an even rotation. Hence , and or is rev rotated by an even amount. With , is the mirror of the alternating word of , i.e. Lemma 3 conjugate to the alternating word of , and both conditions are invariant under rotation.
Proof of Theorem 1. “”: braid index 1 gives , ; braid index 2 gives no amphichiral knot; braid index 3 is covered by Proposition 7 with Lemmas 4 and 5. “”: Lemma 6 gives a word of negative type closing to a knot with determinant ; the knot is amphichiral by Lemma 3 and has braid index .
Remarks
Remark 8 (composite knots and small determinants). The closure of (, odd) is the connected sum of with its mirror image, with determinant . For — the three values excluded in Stoimenow’s conjecture on prime alternating achiral knots [14] — the only sequences of positive or negative type, up to rotation, are , , : the unknot, and the connected sums of resp. with their mirror images. The search is finite because : the trace is a polynomial in the exponents with nonnegative coefficients and constant term 2 (all exponents 0 give the identity), in which every product of one - and one -exponent occurs with coefficient 1.
Remark 9 (Stoimenow’s conjecture). Stoimenow conjectured [14 Conjecture 4.20, Conjecture 4.2 in arXiv v3]: an odd is the determinant of a prime alternating achiral knot iff is a sum of two squares and ; he settled the with a representation by coprime components and the squares, and reports the conjecture true for . Up to 10,693 (the largest determinant among the verified amphichiral prime alternating knots of Burton’s census, which go up to 18 crossings — alternating amphichiral knots have even crossing number) there remain 169 odd not settled by him (sums of two squares that are not squares and have no representation by coprime components); for 74 of them is not a sum of two squares, so by Corollary 2 any witness has braid index at least 5. Ferudun [7] realizes every left open by Stoimenow (unrefereed). Of the 74, 26 have no witness in the census: 6309, 6413, 7281, 7301, 7857, 8361, 8433, 8477, 8793, 8973, 8993, 9065, 9081, 9117, 9225, 9245, 9297, 9477, 9621, 9873, 10125, 10161, 10377, 10413, 10469, 10629. Forty-one further alternating census knots with 14 to 18 crossings, amphichiral according to SnapPy’s numerical symmetry group, stayed undecided in our verified computation; none of their determinants is among these 26. For 25 of the 26 we found witnesses that are closures of 5-braid words , where is a word in and is read backwards with each replaced by ; they are prime (verified hyperbolic), alternating (reduced alternating diagrams of 20 to 26 crossings), verified amphichiral, with Morton–Franks–Williams bound 5, so of braid index exactly 5; details in the deposit [Z]. The remaining value, , is covered only by [7]. No knot that is the closure of a word with of at most 20 letters (all words for each ; their closures are braid diagrams of crossings) has determinant 6413.
Remark 10 (the alternating half without the classification). That an amphichiral knot of braid index 3 is alternating also follows without the classification theorem of [1]. The non-generic normal forms are excluded as in the first lines of the proof of Proposition 7. For a generic form write , ; Erle’s Theorem 2.6(iv) [6] states , which equals , the exponent sum; his printed proof treats with and with (the remaining cases are left as an exercise); we use only the case , , and only that there, which does not depend on the sign convention for . For a minimal braid of an amphichiral knot (by Dynnikov–Prasolov [5 Theorem 9], as in [9 Theorem 3.2, Lemma 3.3, eq. (1)]) and . Suppose ; then . If , ; if , the mirror of , conjugated by and cyclically rotated, is the generic form with and the roles of the and exchanged, whose is , and changes sign under mirroring. Either way , a contradiction; so and is alternating. The same argument with the extreme degrees of the Jones polynomial in place of Dynnikov–Prasolov is Chbili’s [3 Proposition 5.1]. Stoimenov’s proof of [13 Theorem 1.1] also bypasses the classification, as described in §2.
Remark 11 (context). Datta [4] relates links of braid index to binary quadratic forms of discriminant , the trace of the Burau matrix at ; for the knots here and the discriminant is . He does not consider amphichirality. For in Lemma 5 the alternating word is of the form , with obtained from by (as in [17 p. 10]), and the square determinant can be explained as Stoimenow does for the Turk’s head braids with odd (this is the case ; the closure is a knot iff ), through strong achirality and Hartley–Kawauchi [17 Theorems 3.1, 3.2].
Checks
Braid index test of “”. Contrapositive: an amphichiral knot whose determinant has not a sum of two squares must have braid index . (a) KnotInfo [10], crossings, tabulated braid index: of the 78 amphichiral knots, the 22 of braid index all satisfy the condition, and the 22 that violate it (a different set of knots) all have braid index ; without amphichirality the implication fails (107 chiral knots of braid index 3 have not a sum of two squares, the first being 5₂). (b) Burton’s census [2], crossings, amphichirality verified with SnapPy [12] (knots left undecided by the verified computation are not counted): 3,049 alternating and 1,950 non-alternating knots violate the condition. The Morton–Franks–Williams lower bound is for all alternating ones and for 1,892 of the others; the remaining 58 have Jones span smaller than the crossing number, so they are not alternating (the census crossing numbers are exact), and by the argument of Remark 10, which does not use the classification of [1], they do not have braid index 3. None of the 4,999 has braid index 3; with the odd braid index [9 Theorem A], all have braid index . These checks are consistent with Theorem 1; by their nature they cannot prove it.
Other checks. Lemmas 4, 5, 6 and the type conditions were checked by exhaustive enumeration on small boxes, and the relation against SnapPy for random 3-braid knots; scripts and logs are in the deposit [Z].
Tool use
This work was prepared with substantial use of Claude Code (Anthropic), an agentic AI system based on large language models, working together with the author.
Mathematics. The proofs were worked out by Claude Code under the author’s direction.
Writing and typesetting. The text and the LaTeX source were written by Claude Code, in English.
Code. The checks were written and run by Claude Code in Python, with SnapPy [12], Sage and the KnotInfo database [10]; the 5-braid search of Remark 9 ran as a CUDA kernel through CuPy on a GPU.
Checking. The argument and drafts of this note were read by separate Claude Code agents in independent cold reads and by outside AI readers (Claude in a web browser, DeepSeek and ChatGPT). Their findings were checked and incorporated where they held.
Literature. The search was done by a Claude Code agent (web and bibliographic databases), with the author by hand (Google Scholar, Math-Overflow) and, by Claude Code, through the arXiv API; an outside AI reader (Claude in a web browser) suggested further sources. [17] was found in the author’s search and [13] was pointed out by an outside AI reader (Claude in a web browser), both only after the argument was written.
The author takes full responsibility for all contents.
Data and code
Scripts and their outputs are in the deposit [Z] on Zenodo; its README describes which files support the computational checks.
Wasserburg am Inn, Bavaria, Germany
Email address: [email protected]
References
- [1][BM] J. S. Birman, W. W. Menasco, Studying links via closed braids III: Classifying links which are closed 3-braids, Pacific J. Math. 161 (1993) 25–113.DOI
- [2][Bu] B. A. Burton, census of prime knots up to 19 crossings, data files https://people.smp.uq.edu.au/BenjaminBurton/knots/, linked from https://regina-normal.github.io/data.html (downloaded 2026).
- [3][Ch] N. Chbili, A note on the Jones polynomials of 3-braid links, Siberian Math. J. 63 (2022) 983–994.DOI
- [4][D] A. Datta, A novel connection between integral binary quadratic forms and knot polynomials, arXiv:2204.13660v1 (2022).arxiv.org/abs/2204.13660
- [5]I. Dynnikov, M. Prasolov, Bypasses for rectangular diagrams. Proof of Jones’ conjecture and related questions, arXiv:1206.0898v2 (2013); journal reference given there: Proc. of Moscow Math. Soc. 74 (2013), no. 1, 115–173; English translation: Trans. Moscow Math. Soc. 74 (2013) 97–144.arxiv.org/abs/1206.0898
- [6]D. Erle, Calculation of the signature of a 3-braid link, Kobe J. Math. 16 (1999) 161–175.DOI
- [7]A. Ferudun, Determinants of prime alternating achiral knots: a partial answer to a problem of Stoimenow, Zenodo, 10.5281/zenodo.23482759 (2026).DOI
- [8]R. Hartley, A. Kawauchi, Polynomials of amphicheiral knots, Math. Ann. 243 (1979) 63–70.DOI
- [9]Hyungseok Jung, Amphichiral knots: odd braid index and symmetry classification in the three-braid case, arXiv:2609.12295v1 (2026).arxiv.org/abs/2609.12295
- [10]C. Livingston, A. H. Moore, KnotInfo: Table of Knot Invariants, knotinfo.org (accessed 2026).
- [11]P. Lisca, On 3-braid knots of finite concordance order, Trans. Amer. Math. Soc. 369 (2017), no. 7, 5087–5112; arXiv:1504.03129v3.arxiv.org/abs/1504.03129
- [12]M. Culler, N. M. Dunfield, M. Goerner, J. R. Weeks, SnapPy, version 3.3.2.
- [13]A. Stoimenov, The classification of partially symmetric 3-braid links, Open Math. 13 (2015) 444–470, 10.1515/math-2015-0043.DOI
- [14]A. Stoimenow, Square numbers, spanning trees and invariants of achiral knots, Comm. Anal. Geom. 13 (2005) 591–631; Conjecture 4.20 there; also arXiv:math/0003172v3, Conjecture 4.2 there.
- [15]A. Stoimenow, Square numbers and polynomial invariants of achiral knots, Math. Z. 255 (2007) 703–719.DOI
- [16]A. Stoimenow, Non-triviality of the Jones polynomial and the crossing numbers of amphicheiral knots, arXiv:math/0606255v2 (2007).arxiv.org/abs/math/0606255
- [17]A. Stoimenow, Wheel graphs, Lucas numbers and the determinant of a knot, preprint, Max-Planck-Institut für Mathematik (2000).