Monomial Divisibility Diagrams (MDDs)

• Monomial Divisibility Diagrams (MDDs) are algebraic and combinatorial structures that represent divisibility relations among monomials via Hasse diagrams and DAGs.
• They serve as optimized data structures for fast membership testing and insertion in monomial ideals, crucial for signature-based Gröbner basis algorithms.
• MDDs reveal important poset topology properties such as shellability and homological invariants, aiding advances in computational algebra and algebraic geometry.

Monomial Divisibility Diagrams (MDDs) are algebraic and combinatorial structures that encode the divisibility relations among monomials in multivariate polynomial rings, as well as optimized data structures for representing monomial ideals. MDDs are foundational in the study of poset topology and are deployed as practical data structures in fast computational algebra, particularly for signature-based Gröbner basis algorithms. The terminology encompasses both the Hasse diagram of a poset of exponent vectors under a proper divisibility order, and, computationally, the shared subtree DAG representation of monomial ideals supporting insertion and membership in logarithmic time.

1. Proper Divisibility Posets and MDD Hasse Diagrams

Given $a = (a_1, \ldots, a_n) \in \mathbb{N}^n$, one associates the monomial $x_1^{a_1} \cdots x_n^{a_n}$ with its exponent vector $a$. The relation $b = (b_1, \ldots, b_n)$ properly divides $c = (c_1, \ldots, c_n)$ (denoted $b < c$) if for every $i$, either $b_i = c_i = 0$ or $b_i < c_i$. The poset $P(a_1, \ldots, a_n)$ consists of all $b \in \mathbb{N}^n$ such that $b \leq a$ under the closure of this order, with unique minimal element $0$ and maximal element $a$.

The Hasse diagram of $P(a_1, \ldots, a_n)$—the Monomial Divisibility Diagram (MDD)—has as vertices the exponent vectors $b \in P(a)$, and directed edges for each covering relation under proper divisibility. Moving up in the MDD corresponds to increasing a single positive coordinate or transitioning from zero to one in a coordinate. This structure captures intricate refinement of monomial divisibility and appears naturally in the context of monomial ideals, Buchberger’s algorithm, and the combinatorics of free resolutions (Bolognini et al., 2015).

2. Order Complexes and Shellability Properties

The order complex $\Delta(P(a))$ is the simplicial complex whose $k$-simplices are chains $b^0 < b^1 < \cdots < b^k$ in $P(a) \setminus \{0,a\}$. In general, $P(a)$ is not pure, and thus $\Delta(P(a))$ may be a non-pure complex in the sense of poset topology. The dual poset $P(a)^*$ admits a recursive atom-ordering for every $a$, making it CL-shellable. Consequently, $\Delta(P(a))$ is (non-pure) shellable and vertex-decomposable. Shellability provides strong topological information: $\Delta(P(a))$ is homotopy equivalent to a wedge of spheres in various dimensions, and all reduced homology groups are torsion-free (Bolognini et al., 2015).

An explicit counterexample, $P(4,4)$, demonstrates that the poset $P(a)$ itself need not be CL-shellable even if its order complex is shellable, but its dual always is. This resolves questions about implications between shellability and CL-shellability for this class of posets.

3. Canonical Tree and DAG Representations for Monomial Ideals

MDDs have a dual computational interpretation: as the canonical tree or directed acyclic graph (DAG) representation of monomial ideals as subsets of $\mathbb{N}^n$. Each monomial $x_1^{\alpha_1} \cdots x_n^{\alpha_n}$ is identified with its exponents $\alpha=(\alpha_n,\ldots,\alpha_1)$. An ideal tree is a rooted $n$-level prefix tree with nodes corresponding to partial exponent vectors and edges labeled by integer exponents. The tree encodes inclusion-minimal monomials and the strictly monotonic structure induced by the ideal property $I + \mathbb{N}^n = I$ (Lairez et al., 8 Jan 2026).

Maximal sharing of identical subtrees across the ideal tree yields the MDD as a DAG: structurally equal subtrees are identified via hash-consing, producing an efficient and compact representation.

Structure	Representation	Typical Use
Poset MDD	Hasse diagram	Divisibility topology, homology
Ideal-tree MDD	Canonical n-tree/DAG	Membership, insertion in algorithms

4. Algorithms: Membership, Insertion, and Complexity

Membership testing in an MDD exploits the prefix tree structure. Given $t$ representing the current monomial ideal and $\alpha=(\alpha_1,\ldots,\alpha_n)$, a recursive search descends along paths determined by the exponents, performing at each step a binary search over strictly sorted child lists per node. With maximal out-degree $\delta$, the complexity is $O(n \log \delta)$, reduced to $O(n)$ if dense array storage is deployed.

Insertion of a new generator $\alpha$ modifies only the relevant paths and their strict-fibre decompositions, via recursive calls and pruning of redundant subtrees, yielding an overall $O(N)$ complexity per generator, where $N$ is the current number of edges in the MDD. The width of the MDD after $r$ insertions satisfies $w \leq 2^r - 1$ in the worst case.

Empirically, unoptimized ideal trees can become infeasibly large, while the DAG MDD representation achieves orders-of-magnitude memory savings for structured ideals, particularly those arising in Gröbner basis computations.

5. Homology and Topological Invariants in the Two-Variable Case

For $n=2$, denoting $P(a,b)=P(a_1,a_2)$ with $a \leq b$, one obtains precise closed formulas for the reduced homology groups of $\Delta(P(a,b))$, as well as the Euler characteristic $\chi$:

• For $2 < a < b$ and $0 \leq i \leq a-2$:

$$\mathrm{rank}\; \tilde H_i(\Delta(P(a,b));\mathbb{Z}) = \sum_{t=0}^i \binom{a-3-i}{t}\binom{b-2-i}{i-t} + \sum_{t=0}^{i+1}\binom{a-2-i}{t-1}\binom{b-3-i}{i+1-t}$$

with the convention $\binom{n}{-1}=0$, $\binom{0}{0}=1$.

• All non-reduced homology groups above $i=a-2$ vanish. The degree of top non-vanishing homology, $t(a,b)$, is given by $t(a,b) = a-2-k$ whenever $2a-3k-2 \leq b \leq 2a-3k$.
• The reduced Euler characteristic is

$$\chi(\Delta(P(a,b))) = (-1)^a 2 \sum_{h=0}^{\lfloor (a-2)/2 \rfloor} (-1)^h \binom{b-a}{a-2-2h}$$

(Bolognini et al., 2015).

For example, $P(2,3)$ consists of $12$ points and its order complex is homotopy equivalent to three points, with $\tilde H_0$ of rank $3$ and higher homology vanishing.

6. Applications in Gröbner Basis Computation

MDDs are well-suited for signature-based Gröbner basis (sGB) algorithms, which require repeated, rapid membership checks in evolving monomial ideals. In sGB, signatures $s(p) \in \mathbb{N}^n$ track polynomial origins and guide avoidance of redundant reductions via the syzygy and Koszul criteria. These criteria reduce to monomial-ideal membership tests, which are executed at very large scale—for example, $\sim 1.2 \times 10^8$ tests in eco-14.

Implementations such as the Julia package AlgebraicSolving.jl maintain two concurrent MDDs—one for the leading monomials of the current Gröbner basis, one for observed signatures. Each new critical pair triggers membership checks and, if new, insertions, as captured by formal pseudocode in (Lairez et al., 8 Jan 2026). Empirical results show symbolic (monomial) computation speedups ranging from $2\times$ to $8.4\times$ (e.g., $4.3\times$ symbolic speedup on eco-14) and sizable decreases in memory footprint for typical algebraic geometry benchmarks.

7. Strengths, Limitations, and Future Directions

MDDs offer dynamic insertion in $O(N)$ time and membership queries in $O(n \log \delta)$, independent of the generator count. Subtree sharing produces substantial memory savings for highly structured monomial ideals. However, for unstructured collections of monomials, the worst-case width can still be exponential in the number of generators, and the overhead from hash-consing and memoization can dominate for small ideals. MDDs efficiently address only the membership and insertion tasks for monomial ideals; other symbolic operations in Gröbner basis methods may remain computational bottlenecks after MDD optimization (Lairez et al., 8 Jan 2026).

From a topological perspective, the shellability of posets $P(a_1,\ldots,a_n)$ and their order complexes establishes a robust connection to the persistence and structure of syzygies and free resolutions. The study of shellability beyond $n=2$ and the investigation of the proper-division product of posets reveal further research directions, with preliminary results indicating rich combinatorial phenomena in Boolean and related lattices (Bolognini et al., 2015).