Cartesian Coupling in Angular Momentum

• Cartesian coupling is the reformulation of spherical-harmonic expressions into completely symmetric traceless Cartesian tensors, providing a clear geometric interpretation of angular momentum.
• It employs algorithmic assembly and contraction rules to replace summations over magnetic indices with efficient tensor operations.
• This method improves computational efficiency and symbolic clarity in fields like atomistic machine learning, despite the rapid combinatorial growth of terms.

Searching arXiv for the provided topic and closely related work to ground the article in published research. Cartesian coupling, in the angular-momentum sense developed through Cartesian harmonic tensors, is the rewriting of coupled spherical-harmonic expressions in terms of completely symmetric traceless Cartesian tensors built from unit vectors. In this formulation, each $Y^\ell_m(\hat{\mathbf a})$ is replaced by the rank-$\ell$ irreducible Cartesian tensor $a^{\{\ell\}}$, and couplings of spherical harmonics associated with different solid angles are reduced by tensor products, contractions, and traceless symmetrization until rotational invariants appear as explicit polynomials in scalar products of the participating unit vectors (Parke, 2023).

1. Irreducible Cartesian tensors as the basis of Cartesian coupling

Parke’s construction identifies Cartesian harmonic tensors as the Cartesian realization of the usual rank-$\ell$ irreducible angular objects. For a unit vector $\widehat{\mathbf a}$, the rank-$\ell$ tensor is obtained from the direct product $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$ by projection to the completely symmetric traceless part. The explicit closed form is

$$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$

where the braces denote the completely symmetrized sum over $\ell-2r$ vector factors and $r$ Kronecker deltas, each distinct term counted once (Parke, 2023).

The normalization factor

$\ell$0

is chosen so that contraction with the unit vector yields the next lower-rank tensor. In this setting, complete symmetry means invariance under arbitrary permutations of the $\ell$1 Cartesian indices, while tracelessness means contraction over any index pair vanishes. The alternating sum in the defining expression implements the traceless projection concretely: the $\ell$2 term is the raw symmetric product, the $\ell$3 term subtracts traces, the $\ell$4 term restores over-subtracted double traces, and so on.

Representation-theoretically, these tensors carry the same $\ell$5-dimensional irreducible representation of $\ell$6 as the spherical harmonics $\ell$7. The paper justifies this by counting independent components: a symmetric rank-$\ell$8 tensor in $\ell$9 dimensions has $a^{\{\ell\}}$0 components, the trace conditions remove $a^{\{\ell\}}$1, and for $a^{\{\ell\}}$2 the result is $a^{\{\ell\}}$3. This is the precise basis for replacing spherical-basis manipulations by Cartesian ones.

2. Explicit construction and low-rank forms

The low-rank tensors make the construction transparent. For rank $a^{\{\ell\}}$4,

$a^{\{\ell\}}$5

For rank $a^{\{\ell\}}$6,

$a^{\{\ell\}}$7

the familiar symmetric traceless dyad. For rank $a^{\{\ell\}}$8,

$a^{\{\ell\}}$9

For rank $\ell$0, the same pattern continues with a quartic monomial, six single-trace subtraction terms, and three double-trace restoration terms (Parke, 2023).

These examples exhibit the general architecture of Cartesian coupling at the tensor level: the leading product in $\ell$1 is followed by all admissible trace-subtraction structures built from $\ell$2, with coefficients fixed by irreducibility and normalization. The corresponding combinatorics grow rapidly. The number of distinct terms in

$\ell$3

is

$\ell$4

This counts the number of distinct placements of $\ell$5 pairwise contractions among $\ell$6 indices, leaving $\ell$7 vector slots.

A plausible implication is that the practical difficulty of Cartesian coupling is not conceptual but combinatorial: the irreducible object has only $\ell$8 independent components, yet its explicit symmetrized-traceless realization contains rapidly proliferating brace terms.

3. Coupling rules and scalar reduction

The paper’s central motivation is the coupling of spherical harmonics associated with different solid angles to total angular momentum zero. In the Cartesian formulation, the operative rules are straightforward. First, associate each $\ell$9 with its Cartesian harmonic tensor $\widehat{\mathbf a}$0. Second, form tensor products of these irreducible tensors. Third, reduce the products by contraction and traceless symmetrization into irreducible Cartesian tensors, in direct analogy with angular-momentum coupling. Fourth, when the target is a scalar, fully contract all indices to obtain rotational invariants built from $\widehat{\mathbf a}$1 and vector components, which reduce to dot products of unit vectors (Parke, 2023).

The only explicit worked coupling printed in the paper is a nested spherical-harmonic scalar reduction whose final result is an expression entirely in terms of scalar products such as $\widehat{\mathbf a}$2, $\widehat{\mathbf a}$3, and $\widehat{\mathbf a}$4. The text notes a probable typographical inconsistency in the variable labels, but the formal point is unaffected: the coupled scalar is rewritten as a geometric polynomial invariant rather than as a sum over magnetic quantum numbers.

This is precisely the Cartesian coupling payoff. The basis-dependent $\widehat{\mathbf a}$5-summation machinery is replaced by contractions among irreducible Cartesian tensors, and the final scalar forms are geometrically transparent. The paper further states that earlier Lehman–Parke work supplied closed-form expressions for tensor-product decomposition coefficients analogous to those of Racah, as well as explicit scalar couplings of up to five spherical harmonics. The present paper, however, concentrates on generating the irreducible Cartesian tensors that serve as the input to those reductions rather than tabulating the coupling coefficients themselves.

4. Algorithmic assembly of arbitrary-rank tensors

A central contribution is an explicit indexing scheme for symbolic generation of the brace terms. For a contribution with $\widehat{\mathbf a}$6 vector factors and $\widehat{\mathbf a}$7 deltas, vector indices are labeled $\widehat{\mathbf a}$8, $\widehat{\mathbf a}$9, and the two indices of the $\ell$0-th delta are $\ell$1, $\ell$2. The allowed ranges are

$\ell$3

$\ell$4

$\ell$5

with $\ell$6. These inequalities generate each distinct symmetrized term exactly once and prevent duplication (Parke, 2023).

The paper presents this as an Algebraic Assembly Language method and compares it with an independent Mathematica construction based on nested sums and Boole[...] constraints. For $\ell$7, the Mathematica implementation generates $\ell$8 terms, in agreement with

$\ell$9

That case takes about $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$0 seconds in Mathematica on the author’s machine, whereas the compiled assembly version takes less than $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$1 milliseconds, roughly a factor of $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$2 faster.

The sample outputs make the assembly logic concrete. For $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$3, the code prints the expected quartic term, six single-trace terms with coefficient $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$4, and three double-trace terms with coefficient $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$5, with the overall normalization omitted. For $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$6, it prints the $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$7 brace contribution with $$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$8 distinct terms. The validation strategy is structural and combinatorial: agreement of explicit term lists, agreement of term counts, and recovery of the known rank-$$\widehat{\mathbf a}\otimes\cdots\otimes\widehat{\mathbf a}$$9, rank-$$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$0, and rank-$$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$1 formulas.

5. Relation to spherical-basis methods and later Cartesian tensor coupling

The relation to ordinary spherical harmonics is conceptual rather than componentwise in the paper: Cartesian harmonic tensors “correspond to the traditional spherical harmonics $$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$2, but with their base vectors transformed from polar to Cartesian form.” The paper does not provide an explicit matrix $$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$3 mapping $$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$4 to Cartesian components, nor does it develop a standalone Cartesian Clebsch–Gordan algebra in full (Parke, 2023).

Later work in equivariant atomistic machine learning makes this implicit coupling structure explicit. “Cartesian-nj: Extending e3nn to Irreducible Cartesian Tensor Product and Contracion” defines Cartesian-3j and Cartesian-nj objects as direct analogues of Wigner-3j and generalized Clebsch–Gordan/Wigner-nj coefficients for irreducible Cartesian tensors, with

$$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$5

as the tensor-product channel rule (Xu et al., 18 Dec 2025). That paper also defines Cartesian harmonics as fully symmetric traceless irreducible Cartesian tensors generated from unit vectors, using the normalization factor

$$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$6

A different later use appears in CEITNet, where “Cartesian local-environment many-body coupling” denotes coupling of Cartesian tensor channels assembled from neighbor directions rather than coupling in an irreducible spherical basis. There the interaction is performed in channel space by Cartesian tensor operations such as matrix multiplication and outer products, while equivariance is inherited from the geometric tensor basis (Jin et al., 4 Feb 2026).

This suggests that the angular-momentum formulation of Cartesian coupling in Parke’s sense and later Cartesian tensor couplings in machine learning share a common structural intuition: replace basis-index summations by explicit tensor algebra in Cartesian space, then organize the result by irreducibility, contraction, or both.

6. Scope, advantages, and limitations

The main practical advantage of Cartesian coupling is geometrical transparency. Scalar couplings emerge as explicit polynomials in dot products of unit vectors, which is especially useful when many coupled harmonics must ultimately reduce to a scalar. The formalism also makes direct contact with symbolic computation, because tensor contractions and trace subtractions can be generated explicitly and checked mechanically (Parke, 2023).

The paper’s limitations are equally clear. It is focused narrowly on three-dimensional Cartesian harmonic tensors as building blocks for angular-coupling calculations. It does not provide explicit basis-conversion formulas between the $$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$7-basis and Cartesian components. It does not tabulate Cartesian Clebsch–Gordan coefficients in the text itself. It refers the broader coupling theory back to earlier Lehman–Parke and older Coope–Snider–McCourt work. And its explicit brace expansion suffers a combinatorial explosion even though the number of independent tensor components remains only $$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$8.

A common misconception is to treat Cartesian coupling as merely a change of notation from $$a^{\{\ell\}}= \left[ \frac{(2\ell-1)!!}{\ell!}\right] \sum_{r=0}^{\lfloor \ell/2 \rfloor } (-1)^r \left[ \frac{(2\ell-2r-1)!!}{(2\ell-1)!!}\right] \left\{ a\cdots (\ell-2r)\cdots a\,\delta\cdots (r)\cdots \delta \right\},$$9 to tensors. The construction is more specific than that. It is a concrete irreducible-tensor program built from complete symmetrization, explicit detracing, and algorithmic term generation, with the explicit aim of converting spherical-harmonic couplings into scalar and tensor invariants expressed through Cartesian geometry. In that restricted but technically important sense, Cartesian coupling is not an alternative representation layered on top of the same computations; it is a different computational route to the same angular objects, optimized for contraction, scalar reduction, and symbolic manipulation.