Cubic Jordan algebras are not a series
Abstract.
The idea of the exceptional series is that the exceptional simple Lie algebras should form a series. Since all four simple Lie algebras in the fourth row of the Freudenthal magic square are exceptional it is natural to ask if the remaining rows form a series. A stronger version of this question is that, for the first two rows (corresponding to and ), there is a category defined by a presentation which is a reasonable candidate for the series. Our main results show that neither of these candidates is a series but each consists of a finite set of points. In each case the series is defined by a parameter and we show that the relations imply that this parameter satisfies a polynomial. These two results were obtained by a computer calculation. Our calculation is supported by the website, www.brucewestbury.uk, for inspection, and the calculations are certified by Lean 4.
Contents
1. Introduction
The classification of complex simple Lie algebras is due to Killing in 1888. The classification consists of four infinite families, known as the classical Lie algebras, and five others, known as the exceptional Lie algebras.
A more recent development was the discovery of the Vogel plane, partially published in [Vog11]. Vogel proved (independently of Killing’s classification) that the complement of the trivial representation in the symmetric square of the adjoint representation of a simple Lie algebra (other than ) has either two or three composition factors. Furthermore, for the exceptional algebras, the number of composition factors is two. The four infinite families of classical Lie algebras are often regarded as forming a series; for example, the ring of symmetric functions can be regarded as the character ring of where is a formal variable. The radical idea in Vogel’s work is that the exceptional Lie algebras might also form a series.
In earlier work,[Fre64], Freudenthal gave a uniform matrix construction of the exceptional Lie algebras. This construction is now known as the Freudenthal magic square. The fourth row of the magic square consists of four of the exceptional Lie algebras. Since Vogel had proposed that the fourth row might form a series it is natural to ask about the remaining rows. The result announced in [Thu04] is the negative answer that the first two rows of the magic square do not form a series. This result was obtained by a computer calculation which cannot be inspected and which has not been reproduced. This work reconstructs the calculation from first principles using diagrammatic reduction rules. produces a transparent reduction system, and supplies independently checkable certificates for the computation.
This work is supported by the website www.brucewestbury.uk which has the role of an Appendix and presents details of the calculations. This work is further supported by Lean code which independently certifies the reduction steps. The coding was done by ChatGPT and Claude.
Acknowledgment
The author would like to thank Dylan Thurston for valuable conversations.
2. Cubic Jordan algebras
The classical Lie algebras all have matrix constructions; so the classification of simple Lie algebras raised the problem of giving matrix constructions of the exceptional Lie algebras. The Lie algebra was shown to be the derivation algebra of the octonions by Elie Cartan in 1913. Then matrix constructions of and were given in [Fre53] by constructing actions on the Jordan algebra of Hermitian matrices with entries in the octonions. This algebra is known as the Albert algebra.
2.1. Construction
An accessible account of these constructions is given in [Ada82]. Let be a (unital) composition algebra with involution and norm . The condition for a composition algebra is
An element is real if and imaginary if . Every has a real part and an imaginary part with .
Let be a composition algebra of dimension . Let be the space of Hermitian matrices with entries in . The space has dimension .
Define the Jordan product of by
It is clear that the Jordan product is commutative and power associative; also, the unit matrix is a unit for the Jordan product. The Jordan identity is for all . The Jordan product also satisfies the Jordan identity. This is clear if is associative, and, remarkably, also holds for , the octonions.
The Jordan algebra is a cubic Jordan algebra. An element can be written as
where the diagonal entries are scalars and the remaining entries are in . The norm, , is given explicitly by
Define a symmetric inner product using the matrix trace by
The properties of the inner product are that it is non-degenerate and invariant where invariance is the condition that,for ,
Let be the derivation algebra of . Then preserves the inner product. Let be the subspace of trace-free matrices. Then decomposes as a sum of irreducible -modules
where is the subspace of scalar matrices. The projection map , is a map of -modules.
Define a product on by
This is a map of -modules, .
Define , the norm of , by . Then the fundamental identity is that, for .
2.2. Cubic forms
The abstract theory of cubic Jordan algebras is given in the text books [Jac68], [Spr97], [KMRT98], [McC04, Part II 4], [GPR24, VI].
A Jordan algebra is a commutative algebra that satisfies the Jordan identity. The polarised Jordan identity is:
A cubic Jordan algebra is a Jordan algebra, , together with a linear form , a quadratic form and a cubic form such that a generic element satisfies
Define the adjoint by
Then . Let be a symmetric cubic form on a finite dimensional vector space . Then the first linearisation is which is the coefficient of in the expansion of , so
Note that is quadratic in and linear in . The second linearisation is
which comes from the expansion
Note that is linear in all three variables.
The cubic form can be recovered from the polarisation (if 6 is invertible) since
A basepoint is an element with . Define linear forms
Define a quadratic form and its polarisation
These satisfy
Assume the bilinear trace, , is a nondegenerate bilinear form. Then we define the adjoint by
This satisfies .
The identity appears in [McC04, Problem 4.5].
A Jordan cubic norm is a cubic norm that satisfies
We can construct a cubic Jordan algebra from a Jordan cubic norm. The map is quadratic. The linearisation is a bilinear product
This satisfies .
The Jordan product is determined by the sharp product
The polarisation of the fundamental identity, using the symmetry of the product and the inner product, is
| (1) |
for .
The conclusion is Springer’s result that a cubic Jordan algebra has a Jordan cubic norm (by construction) and that a Jordan cubic norm together with a basepoint is a cubic Jordan algebra.
2.3. Lie algebras
Let be a Jordan cubic norm on . Define the structure group to be the subgroup of which preserves . Note that the structure group is an algebraic group.
Let be a cubic Jordan algebra. Then we have the structure group and the derivation algebra, . Define the structure algebra, , to be the Lie algebra of the structure group of the Jordan cubic norm. By construction, is a subalgebra of .
The action of on preserves the unit, . Define to be the subspace orthogonal to , equivalently, the kernel of . Then , as a module for , decomposes as . Also, , as a module for , decomposes as .
Restricting to the kernel of ,
and then projecting to the orthogonal complement to gives . This shows that the sharp product is a bilinear product on and that is the Lie algebra of derivations.
Then the quadratic norm on satisfies .
3. Derivation algebras
3.1. String diagrams
The string diagrams are discussed in [Cvi08, Chapter 19] (our convention is ), [GSZ22] (our convention is and ) and [MST24, §6.2.3,§6.7].
In §2 we described the algebraic theory of a cubic Jordan algebra. In this section we consider the algebraic theory for the trace-free subspace of a cubic Jordan algebra and describe the PROP for this theory in terms of string diagrams. The data for this theory is a vector space with a non-degenerate symmetric inner product and a fully symmetric cubic form . This data determines a bilinear multiplication which satisfies
The cobordism category of trivalent graphs is a strict rigid symmetric monoidal category and has monoid of objects (with addition) and so is a PROP. The cubic norm on the trace-free subspace of a cubic Jordan algebra satisfies the fundamental relation:
In order to write this in terms of string diagrams we first take the polarisation. The polarisation of the fundamental relation is:
The polarisation introduces coefficients. In order to polarise a cubic norm we require 6 to be invertible. Note that we can rescale a trivalent vertex. The fundamental relation fixes a scale.
In addition to the fundamental relation we also impose the reduction rules:
The first reduction rule defines the loop value. The loop value is an element in the coefficient ring. Let be the ring obtained from by inverting 6 and adjoining the formal variable . Then the coefficient ring, , is a commutative -algebra.
The second reduction rule is independent of the fundamental relation and excludes an infinite sequence of cubic Jordan algebras, see [MST24, Example 6.60] for details. The third and fourth reduction rules are consequences of the fundamental relation and the first two reduction rules; see [Cvi08, (19.2),(19.17)] and [GSZ22] and Example 3.6 below.
The last reduction rule is
| (2) |
Note that the last reduction rule is invariant under rotation.
The PROP defined by these relations has a second interpretation as the category of invariant tensors of a trace-free cubic Jordan algebra regarded as a representation of the automorphism group.
3.2. Confluence
We distinguish the reduction rules from the fundamental relation. Each reduction rule is oriented and substitutes one diagram by a linear combination. This defines a rewriting system on linear combinations of diagrams. A reduction step consists of making the substitution in one term.
Proposition 3.1.
Proof.
This is an application of Newman’s Lemma (aka Bergman’s Diamond Lemma). It is sufficient to show that the rewriting system is terminating and locally confluent.
Each term in the support of the linear combination has fewer vertices than the original term. Hence, by well-founded induction, the rewriting system is terminating.
The proof that the rewriting system is locally confluent is a direct calculation. The overlaps are shown in Figure 2. ∎
Corollary 3.2.
Every linear combination of string diagrams has a normal form which is a linear combination of string diagrams of girth at least five.
Proof.
The normal form is given by applying reduction rules. Since the rewriting system is confluent the result is independent of the choice of the sequence of reduction steps. The result follows from the observation that the string diagrams which have no reduction are the string diagrams of girth at least five. ∎
3.3. Closed graphs
Let be the cobordism category of trivalent graphs. For any commutative -algebra, , let be the free -linear category on . In this section we prove:
Theorem 3.3.
The -linear tensor ideal of generated by the fundamental relation and the reduction rules contains where is the empty diagram and is the polynomial
Corollary 3.4.
Let be a field whose characteristic is not 2 or 3. The dimension of a trace-free cubic Jordan algebra over is an element of .
Proof.
A trace-free cubic Jordan algebra over gives a -linear symmetric monoidal functor from to the category of finite dimensional super vector spaces over . Put . The homomorphism maps to the super dimension of the trace-free cubic Jordan algebra and also must map to 0. ∎
Remark 3.5.
Consider as a family of trace-free cubic Jordan algebras over parametrised by the affine line . Then Theorem 3.3 says that this is a family parametrised by the quotient ring .
We evaluate trivalent graphs. Let be a trivalent graph. An evaluation of is a relation of the form where and is the empty graph. We refer to as the evaluation. For example, the evaluation of the loop is . The evaluation of a trivalent graph is the product of the evaluations of the connected components, so we assume is connected.
Applying Corollary 3.2, the normal form of a trivalent graph is a linear combination of trivalent graphs of girth at least five. The method for finding evaluations of connected trivalent graphs is inductive on the number of vertices. There are no connected trivalent graphs with less than 10 vertices. This implies that the normal form of a trivalent graph with less than 10 vertices is its evaluation.
A source is a connected graph with one vertex of valence four and all other vertices of valence three. Given a source we can substitute the six-term relation for the four valent vertex to get a relation which is a linear combination of six trivalent graphs. Note that this is well-defined as the six-term relation is invariant under permutation of the boundary points, by construction.
Example 3.6.
The third and fourth reduction rules are obtained from the sources
If the source has trivalent vertices then three of the terms have vertices and three have vertices. The normal form of this relation is a linear combination of connected trivalent graphs of girth at least five. Furthermore, each term of the relation has at most vertices.
This gives an inductive procedure for finding evaluations. Assume we have found evaluations of all trivalent graphs of girth at least five with fewer than vertices. Given a source with trivalent vertices we construct the relation and its normal form. The normal form is a linear combination of trivalent graphs where each term has girth at least five and at most vertices. If any term has a connected component with fewer than vertices then, by the inductive assumption, we have found the evaluation and we substitute the evaluation for the component. The result of these substitutions is a linear combination of connected trivalent graphs of girth at least five and precisely vertices.
Example 3.7.
The Petersen graph is the unique (up to isomorphism) connected trivalent graph with 10 vertices. There is also a unique (up to isomorphism) source with 8 trivalent vertices. Substituting the six term relation gives a linear combination of six connected trivalent graphs. One of the terms is the Petersen graph and the other five terms have girth at most four. Therefore the normal form of this relation is the evaluation of the Petersen graph.
The number of trivalent graphs of girth at least five and the number of sources is shown in Table 1. For more terms in the first sequence see A014372.
We put and, for we generated all sources with trivalent vertices (up to isomorphism) and computed the linear combination of connected trivalent graphs of girth at least five and vertices. We observed that for each connected trivalent graph there was a source that gave an evaluation. For substituting these evaluations in the remaining relations gave 0. However for 145 of the 335 sources gave a non-zero polynomial. For all sources which gave a non-zero polynomial the polynomial was of the form
where are positive integers. These are all associates of as elements of .
Following [Thu04], we construct a trace-free Jordan algebra for each root of . This shows that none of the roots of will be eliminated by continuing the calculation to .
-
•
The trace-free cubic Jordan algebras for are the algebras for a composition algebra constructed in §2.
-
•
The trace-free cubic Jordan algebra for is the two dimensional irreducible representation of the symmetric group .
-
•
The case is the degenerate case where all diagrams are zero.
-
•
The trace-free cubic Jordan algebra for is the super space of dimension and the automorphism group is the orthosymplectic super group . The PROP reduces to a specialisation of the Brauer category.
-
•
The trace-free cubic Jordan algebra for is the super space of dimension and the automorphism group is . The PROP reduces to the Temperley-Lieb category.
The six diagrams in the fundamental relation generically span a free -module of rank 5. This -module has a symmetric inner product, defined diagrammatically. This inner product is degenerate for and is non-degenerate otherwise. For there are further relations give by the null space of the inner product.
The trace-free Jordan algebra for is constructed in [DG02]. Let be the group of order 2. The centraliser of is , the centre of . Let be the adjoint representation of and the adjoint representation of . The automorphism group of is . The centraliser of is the symmetric group .
There is an isomorphism of -modules where is the seven dimensional fundamental representation of and is the two dimensional irreducible representation. The vector space is a trace-free cubic Jordan algebra whose automorphism group is .
4. Structure algebras
4.1. String diagrams
The string diagrams are discussed in [Cvi08, §2.3,Chapter 18] (our convention is ).
In §2 we described the algebraic theory of a cubic norm. In this section we consider the algebraic theory for the cubic norm of a cubic Jordan algebra and describe the PROP for this theory in terms of string diagrams. The data for this theory is a vector space, , together with a fully symmetric cubic form . This data determines a linear map (where is the dual vector space): which satisfies
The diagram category, , is the cobordism category of oriented trivalent graphs such that every trivalent vertex is either a source or a sink, see Figure 3. The category is a rigid symmetric category and the monoid of objects is the free monoid on two generators. This makes a coloured PROP with two colours. The two generating objects are a dual pair. Note that any cycle has even length.
The polarisation of the fundamental relation is the seven-term relation shown in Figure 4
The polarisation introduces coefficients. In order to polarise a cubic norm we require 6 to be invertible. Note that we can rescale a trivalent vertex. The fundamental relation fixes a scale.
In addition to the fundamental relation we also impose the reduction rules:
The first reduction rule defines the loop value. The loop value is an element in the coefficient ring. Let be the ring obtained from by inverting 6 and adjoining the formal variable . Then the coefficient ring, , is a commutative -algebra.
The coloured PROP defined by these relations has a second interpretation as the category of invariant tensors of a cubic norm regarded as a representation of the structure group.
4.2. Confluence
We distinguish the reduction rules from the fundamental relation. Each reduction rule is oriented and substitutes one diagram by a linear combination. This defines a rewriting system on linear combinations of diagrams. A reduction step consists of making the substitution in one term.
Proposition 4.1.
The rewriting system arising from the reduction rules in Figure 5 is confluent.
Proof.
This is an application of Newman’s Lemma (aka Bergman’s Diamond Lemma). It is sufficient to show that the rewriting system is terminating and locally confluent.
Each term in the support of the linear combination has fewer vertices than the original term. Hence, by well-founded induction, the rewriting system is terminating.
The proof that the rewriting system is locally confluent is a direct calculation. The overlaps are shown in Figure 6. ∎
Corollary 4.2.
Every linear combination of string diagrams has a normal form which is a linear combination of string diagrams of girth at least six.
Proof.
The normal form is given by applying reduction rules. Since the rewriting system is confluent the result is independent of the choice of the sequence of reduction steps. The result follows from the observation that the string diagrams which have no reduction are the string diagrams of girth at least six. ∎
4.3. Closed graphs
Let be the cobordism category of oriented trivalent graphs. For any commutative -algebra, , let be the free -linear category on . In this section we prove:
Theorem 4.3.
Corollary 4.4.
Let be a field whose characteristic is not 2 or 3. The dimension of a cubic norm over is an element of .
Proof.
A cubic norm over gives a -linear symmetric monoidal functor from to the category of finite dimensional super vector spaces over . Put . The homomorphism maps to the super dimension of the cubic norm and also must map to 0. ∎
Remark 4.5.
Consider as a family of cubic norms over parametrised by the affine line . Then Theorem 4.3 says that this is a family parametrised by the quotient ring .
We evaluate oriented trivalent graphs. Let be an oriented trivalent graph. An evaluation of is a relation of the form where and is the empty graph. We refer to as the evaluation. For example, the evaluation of the loop is . The evaluation of a trivalent graph is the product of the evaluations of the connected components, so we assume is connected.
Applying Corollary 4.2, the normal form of an oriented trivalent graph is a linear combination of trivalent graphs of girth at least six. The method for finding evaluations of connected trivalent graphs is inductive on the number of vertices. There are no connected trivalent graphs with less than 14 vertices. This implies that the normal form of a trivalent graph with less than 14 vertices is its evaluation.
A oriented source is a connected oriented graph with one vertex of valence two connected to a vertex of valence four and all other vertices of valence three. Given an oriented source we can substitute the seven-term relation for the vertex of valence two connected to the vertex of valence four to get a relation which is a linear combination of seven trivalent oriented graphs. Note that this is well-defined as the seven term relation is invariant under permutation of four of the five boundary points, by construction. If the source has trivalent vertices then four of the terms have vertices and three have vertices. The normal form of this relation is a linear combination of connected trivalent oriented graphs of girth at least six. Furthermore, each term of the relation has at most vertices.
This gives an inductive procedure for finding evaluations. Assume we have found evaluations of all trivalent graphs of girth at least six with fewer than vertices. Given an oriented source with trivalent vertices we construct the relation and its normal form. The normal form is a linear combination of trivalent oriented graphs where each term has girth at least six and at most vertices. If any term has a connected component with fewer than vertices then, by the inductive assumption, we have found the evaluation and we substitute the evaluation for the component. The result of these substitutions is a linear combination of connected trivalent oriented graphs of girth at least six and precisely vertices.
Example 4.6.
The Heawood graph is the unique (up to isomorphism) connected trivalent bipartite graph with 14 vertices. There is also a unique (up to isomorphism) source with 11 trivalent vertices. Substituting the seven term relation gives a linear combination of seven connected trivalent bipartite graphs. One of the terms is the Heawood graph and the other six terms have girth at most four. Therefore the normal form of this relation is the evaluation of the Heawood graph.
The number of trivalent oriented graphs and the number of oriented sources is shown in Table 2. For more terms in the first sequence see A260811.
We put and, for we generated all oriented sources with trivalent vertices (up to isomorphism) and computed the linear combination of connected trivalent oriented graphs of girth at least six and vertices. We observed that for each connected trivalent oriented graph there was an oriented source that gave an evaluation. For substituting these evaluations in the remaining relations gave 0. However for 149 of the 406 sources gave a non-zero polynomial. For all oriented sources which gave a non-zero polynomial the polynomial is an associate of as elements of .
There is a strict rigid symmetric monoidal functor from the Structure PROP to the Derivation PROP. This functor is -linear after the substitution . This predicts that divides . A direct comparison of the two polynomials gives
This constructs a cubic norm for all solutions of except and . The functor for sends all diagrams to 0.
We explain the factor by constructing a pair of -linear strict rigid monoidal functors from the PROP to the category of finitely generated free super -modules. Let be a free -module of rank 3 considered as an odd super module. Since is odd, the determinant identifies and the vector cross product is an identification . The structure group is . Since and are inequivalent representations of this constructs a pair of functors.
References
- [Ada82] John Frank Adams. Lectures on Lie groups. University of Chicago Press, 1982.
- [Cvi08] Predrag Cvitanović. Group Theory. Princeton University Press, 2008. doi:10.1515/9781400837670.
- [DG02] Pierre Deligne and Benedict H. Gross. On the exceptional series, and its descendants. Comptes Rendus Mathematique, 335(11):877–881, 2002. doi:10.1016/s1631-073x(02)02590-6.
- [Fre53] Hans Freudenthal. Zur ebenen oktavengeometrie. In Indagationes Mathematicae (Proceedings), volume 56, pages 195–200, 1953. doi:10.1016/S1385-7258(53)50025-3.
- [Fre64] Hans Freudenthal. Lie groups in the foundations of geometry. Advances in Mathematics, 1(2):145–190, 1964. doi:10.1016/0001-8708(65)90038-1.
- [GPR24] Skip Garibaldi, Holger P. Petersson, and Michel L. Racine. Albert algebras over commutative rings : the last frontier of Jordan systems. New mathematical monographs. Cambridge University Press, Cambridge, United Kingdom ; New York, NY, first edition. edition, 2024.
- [GSZ22] Raj Gandhi, Alistair Savage, and Kirill Zainoulline. Diagrammatics for . Bulletin of the London Mathematical Society, 2022. doi:10.1112/blms.12733.
- [Jac68] Nathan Jacobson. Structure and representations of Jordan algebras, volume 39. American Mathematical Soc., 1968.
- [KMRT98] Max-Albert Knus, Alexander Merkurjev, Markus Rost, and Jean-Pierre Tignol. The book of involutions, volume 44 of Colloquium Publications. American Mathematical Soc., 1998.
- [McC04] Kevin McCrimmon. A taste of Jordan algebras. Springer, 2004.
- [MST24] Scott Morrison, Noah Snyder, and Dylan P. Thurston. Towards the quantum exceptional series, 2024. URL: https://arxiv.org/abs/2402.03637, arXiv:2402.03637.
- [Spr97] Tonny A Springer. Jordan algebras and algebraic groups. Springer Science & Business Media, 1997.
- [Thu04] Dylan Thurston. The and families have a finite number of points, 2004. URL: https://www.brucewestbury.uk/papers/F4E6.pdf.
- [Vog11] Pierre Vogel. Algebraic structures on modules of diagrams. Journal of Pure and Applied Algebra, 215(6):1292–1339, 2011. doi:10.1016/j.jpaa.2010.08.013.