Preprint
Article

This version is not peer-reviewed.

Antiprism Triangulations Have Real-Rooted \( h \)-Polynomials

Submitted:

14 November 2026

Posted:

17 November 2026

You are already at the latest version

Abstract
We prove that the \( h \)-polynomial of the antiprism triangulation of every simplicial complex with nonnegative \( h \)-vector has only real roots, as Athanasiadis, Brunink and Juhnke-Kubitzke conjectured. The key step is their conjecture that the \( h \)-polynomial of the antiprism triangulation of a simplex interlaces the theta polynomial of the next simplex. For the local \( h \)-polynomial we prove a convolution identity and reduce its conjectured real-rootedness to one interlacing relation of the same shape, which holds for \( n\le40 \).
Keywords: 
;  ;  

MSC:  Primary 05E45; 26C10; Secondary 05A15; 52B05

1. Introduction

Let Δ be a finite simplicial complex of dimension n − 1 with f i − 1 ( Δ ) faces of dimension i − 1 . Its h-polynomial h ( Δ , x ) = ∑ i = 0 n h i ( Δ ) x i is defined by
∑ i = 0 n f i − 1 ( Δ ) x i = ∑ i = 0 n h i ( Δ ) x i ( 1 + x ) n − i .
The h-vector is nonnegative for every Cohen–Macaulay complex, in particular for every triangulation of a ball or a sphere [16]. A recurring question in geometric combinatorics asks which subdivisions turn a nonnegative h-polynomial into a polynomial with only real roots. Brenti and Welker [11] proved that this happens for the barycentric subdivision sd ( Δ ) : the polynomial h ( sd ( Δ ) , x ) is real-rooted whenever h ( Δ , x ) has nonnegative coefficients. Real-rootedness implies the log-concavity and unimodality of the coefficients.
The antiprism triangulation  sd A ( Δ ) was introduced by Izmestiev and Joswig [13] in their study of branched coverings of manifolds, and appears in theoretical distributed computing as the chromatic subdivision1 [14]. Its vertices are the pointed faces ( F , v ) of Δ , with v ∈ F , and its faces are the sets { ( F 1 , v 1 ) , … , ( F m , v m ) } with F 1 ⊆ … ⊆ F m and v j ∈ F j ∖ F i whenever F i ⊊ F j and i < j [1,3]. Figure 1 shows the antiprism triangulation of a triangle. Athanasiadis, Brunink and Juhnke-Kubitzke [1] developed the combinatorics of sd A ( Δ ) and showed that it shares many properties of the barycentric subdivision, often in a harder form. Their main motivation was the following analogue of the theorem of Brenti and Welker.
Conjecture 1
([1] [Conjecture 1.1]). The polynomial h ( sd A ( Δ ) , x ) is real-rooted for every simplicial complex Δ with nonnegative h-vector.
Write σ n for the ( n − 1 ) -dimensional simplex and ∂ σ n for its boundary, and set
h n ( x ) : = h ( sd A ( σ n ) , x ) , θ n ( x ) : = h n ( x ) − h ( sd A ( ∂ σ n ) , x ) .
Athanasiadis, Brunink and Juhnke-Kubitzke proved that h n is real-rooted and interlaces h n + 1 [1] [Theorem 5.2]. Using the theory of uniform triangulations of Athanasiadis [2] [Theorem 1.2], they reduced Conjecture 1 to a single interlacing relation between two explicit sequences of polynomials.
Conjecture 2
([1] [Conjecture 5.3]). The polynomial h n − 1 interlaces θ n for every n ≥ 2 .
They verified Conjecture 2 by computer for n ≤ 20 . Athanasiadis and Tzanaki [4] [Section 3] observed that it is the only missing part of the strong interlacing property of the antiprism triangulation, and Athanasiadis recorded Conjecture 1 again as open in [3] [Section 6]. We prove both conjectures.
Theorem 3.
Let Δ be an ( n − 1 ) -dimensional simplicial complex with nonnegative h-vector. Then h ( sd A ( Δ ) , x ) is real-rooted. Moreover, h n ( x ) interlaces h ( sd A ( Δ ) , x ) , which in turn interlaces x n h n ( 1 / x ) .
Theorem 4.
For every n ≥ 2 the polynomial h n − 1 interlaces θ n . Equivalently, the polynomial
b n ( x ) = h n ( x ) − x n − 1 h n ( 1 / x ) x − 1
interlaces h n − 1 . Moreover b n has degree n − 2 , leading coefficient n − 1 and only negative roots, so all its coefficients are positive.
Theorem 4 is Conjecture 2, and Theorem 3, which we derive from it, settles Conjecture 1. Together with the results of Athanasiadis and Tzanaki [4], which build on the work of Brändén and Solus [10], it gives real-rooted symmetric decompositions for antiprism triangulations of skeleta, and it implies that the γ -polynomial of the antiprism triangulation of any homology sphere is real-rooted (Section 6).
Corollary 1.
Let Δ be an ( n − 1 ) -dimensional simplicial complex.
(a)
If Δ is Cohen–Macaulay, then the h-polynomial and the f-polynomial of sd A ( Δ ) are real-rooted; their coefficients are log-concave and unimodal.
(b)
If Δ is a homology sphere, then h ( sd A ( Δ ) , x ) = ∑ j γ j x j ( 1 + x ) n − 2 j with ∑ j γ j u j real-rooted; in particular all γ j are nonnegative and form a log-concave sequence.
(c)
If Δ is the ( n − 1 ) -skeleton of an n-dimensional complex Γ with nonnegative h-vector, then h ( sd A ( Δ ) , x ) has a nonnegative, real-rooted and interlacing symmetric decomposition with respect to n, and it interlaces h ( sd A ( Γ ) , x ) .
Part (b) strengthens the γ -positivity proved by Athanasiadis [3] [Corollary 6.1], which confirmed Gal’s conjecture [12] for antiprism triangulations of homology spheres.2
The local h-polynomial ℓ A ( σ n , x ) is conjectured to be real-rooted as well [1] [Conjecture 5.8]. In Section 8 we express it through h 0 , … , h n by a convolution in which every factor is real-rooted (Theorem 7), and we show that the part of the conjecture concerning ℓ A ( σ n , x ) , in a strong form that includes the interlacing of a whole local analogue of the sequence of [1], is equivalent to the relations ℓ A ( σ n , x ) ≺ ℓ A ( σ n + 1 , x ) and h n ≺ ℓ A ( σ n + 1 , x ) for all n (Theorem 8). The second relation has the shape of Theorem 4. Both relations hold for n ≤ 40 ; we do not know a proof.

The Idea of the Proof

Athanasiadis, Brunink and Juhnke-Kubitzke proved that h n is real-rooted by means of an interlacing sequence Q n − 1 = ( f 0 , f 1 , … , f n ) of polynomials whose partial sums are P i = f 0 + … + f i − 1 , with P n = h n and f n = x n h n ( 1 / x ) . We show that the symmetric part b n + 1 of h n + 1 splits along the same partial sums,
b n + 1 = ∑ i = 1 n a ( P i ) , a ( P ) = x rev n − 1 P − P x − 1 ,
where rev n − 1 P = x n − 1 P ( 1 / x ) . Each summand a ( P i ) is interlaced against h n by a sign count at the roots of h n , because both P i and its reverse rev n − 1 P i interlace h n . The second fact is the heart of the matter: it follows from the observation that the reverse of h n itself sits at the start of the interlacing chain P 1 ≺ P 2 ≺ … ≺ P n . Figure 3 shows the structure of the argument. The passage from simplices to all complexes follows the method of Athanasiadis [2]; we give a complete proof, with a short “rotation lemma” in place of the general results on interlacing-preserving matrices that are used there.

Organization

Section 2 collects the facts about interlacing that we use, with proofs. Section 3 recalls the recurrence for h n and the interlacing sequence of [1]. Section 4 proves Theorem 4, and Section 5 proves Theorem 3. Section 6 derives the consequences, Section 7 gives examples and tables, Section 8 treats the local h-polynomial, and Section 9 describes the computer checks. Section 10 lists open problems.

2. Interlacing

All polynomials have real coefficients. A polynomial is real-rooted if all its complex roots are real. In this section every polynomial is nonzero, real-rooted and has a positive leading coefficient, unless stated otherwise. For such a polynomial f with roots α 1 ≥ α 2 ≥ … ≥ α e , listed with multiplicity, put
N f ( t ) : = # { i : α i ≥ t } ( t ∈ R ) .
For d ≥ 0 and a polynomial p of degree at most d we write rev d p ( x ) : = x d p ( 1 / x ) .
Definition 1.
Let f have roots α 1 ≥ … ≥ α e and g have roots β 1 ≥ … ≥ β d . We say that finterlacesg, and write f ≺ g , if d ∈ { e , e + 1 } and
… ≤ α 2 ≤ β 2 ≤ α 1 ≤ β 1 .
A sequence ( g 0 , … , g m ) isinterlacingif g i ≺ g j for all 0 ≤ i < j ≤ m .
Thus the largest root belongs to g, and a constant interlaces every polynomial of degree at most one. In particular f ≺ f for every f.3
Lemma 1.
f ≺ g if and only if 0 ≤ N g ( t ) − N f ( t ) ≤ 1 for all t ∈ R .
Proof. 
If f ≺ g , then on running t downwards from + ∞ the roots are met in the order β 1 , α 1 , β 2 , α 2 , … , so N g − N f alternates between 1 and 0 (ties only shorten the steps). Conversely, assume the inequalities. Since N g ( β k ) ≥ k , we get N f ( β k ) ≥ k − 1 , that is, α k − 1 ≥ β k . Since N f ( α k ) ≥ k , we get N g ( α k ) ≥ k , that is, β k ≥ α k . Letting t → − ∞ gives d − e ∈ { 0 , 1 } .    □
Lemma 2.
(a)
If c is real-rooted with positive leading coefficient, then f ≺ g if and only if c f ≺ c g .
(b)
If f ≺ g or g ≺ f , and β is a root of g of multiplicity m, then β is a root of f of multiplicity at least m − 1 .
(c)
(Weak transitivity.)If g 0 ≺ g 1 ≺ … ≺ g m and g 0 ≺ g m , then ( g 0 , … , g m ) is interlacing.
Proof. (a) N c f = N c + N f and N c g = N c + N g , so the differences in Lemma 1 agree. (b) Put D = N g − N f and choose t > β with no root of f or g in ( β , t ] . Then D ( β ) − D ( t ) = m − mult f ( β ) . Both values of D lie in [ 0 , 1 ] , or both lie in [ − 1 , 0 ] , so m − mult f ( β ) ≤ 1 . (c) Put D i = N g i . For i < j the difference D j − D i is a sum of consecutive differences D l + 1 − D l ≥ 0 , hence it is nonnegative and bounded by the full sum D m − D 0 ≤ 1 .    □
For g with distinct roots β of multiplicities m β put c g : = ∏ β ( x − β ) m β − 1 and g ∘ : = g / c g , so that g ∘ has only simple roots. By Lemma 2, if f ≺ g (or g ≺ f ), then c g divides f, and f ≺ g ⇔ f / c g ≺ g ∘ (respectively g ≺ f ⇔ g ∘ ≺ f / c g ). This reduces most questions to the case of simple roots, where the following sign test applies.
Lemma 3
(Sign test). Let g have degree d and simple roots β 1 > β 2 > … > β d , and let f be a nonzero polynomial with positive leading coefficient.
(i)
f ≺ g if and only if deg f ∈ { d − 1 , d } and ( − 1 ) k − 1 f ( β k ) ≥ 0 for k = 1 , … , d .
(ii)
g ≺ f if and only if deg f ∈ { d , d + 1 } and ( − 1 ) k f ( β k ) ≥ 0 for k = 1 , … , d .
In both cases f is real-rooted.
Proof. 
If f ≺ g (respectively g ≺ f ), then f has exactly k − 1 (respectively k) roots larger than β k , unless f ( β k ) = 0 ; this gives the signs. For the converse let K = { k : f ( β k ) ≠ 0 } and ψ = ∏ k ∉ K ( x − β k ) , so that ψ divides f, and put φ = f / ψ . If k ∈ K is the r-th element of K in increasing order, then ψ ( β k ) has sign ( − 1 ) k − r , so φ ( β k ) is nonzero with sign ( − 1 ) r − 1 in case (i) and ( − 1 ) r in case (ii). Hence φ has a root in each of the | K | − 1 open intervals between consecutive points β k , k ∈ K . In case (i), deg φ ∈ { | K | − 1 , | K | } , and when deg φ = | K | the signs at the smallest point β k , k ∈ K , and at − ∞ differ, which places the last root of φ below all these points. In case (ii) the sign at the largest point is negative and the sign at + ∞ is positive, which places one root of φ above all points, and a possible further root lies below all points. So all roots of φ are real and located, and N g − N f = N g / ψ − N φ takes values in { 0 , 1 } in case (i) and in { − 1 , 0 } in case (ii). Lemma 1 finishes the proof. (If K = ∅ , then f is a positive multiple of g or of ( x − γ ) g with γ real, and both cases follow from Lemma 1.)    □
Lemma 4
(Sums). Let λ 1 , … , λ r > 0 .
(a)
If f i ≺ g for all i, then ∑ i λ i f i ≺ g .
(b)
If g ≺ f i for all i, then g ≺ ∑ i λ i f i .
(c)
If f ≺ g , then f ≺ f + g ≺ g .
Proof. 
For (a) and (b), divide all polynomials by c g and apply Lemma 3 to g ∘ : the sign conditions and the conditions on the degree are preserved by positive linear combinations, since leading coefficients are positive and cannot cancel. (c) follows from (a) and (b) with f ≺ f and g ≺ g .    □
Lemma 5
(Multiplication by x and reversal). Suppose that all roots of f and g are nonpositive.
(a)
f ≺ g if and only if g ≺ x f .
Suppose moreover that all roots are negative.
(b)
If deg f = deg g = d , then f ≺ g if and only if rev d g ≺ rev d f .
(c)
If deg g = deg f + 1 = d , then f ≺ g if and only if rev d − 1 f ≺ rev d g .
Proof. (a) N x f ( t ) = N f ( t ) + [ t ≤ 0 ] . For t > 0 all counting functions vanish, and for t ≤ 0 the two conditions of Lemma 1 coincide. (b), (c) The roots of rev d f are the reciprocals of the roots of f, and t ↦ 1 / t reverses the order of ( − ∞ , 0 ) . In (b) the relation f ≺ g reads α d ≤ β d ≤ α d − 1 ≤ … ≤ α 1 ≤ β 1 < 0 , and taking reciprocals gives 1 / β 1 ≤ 1 / α 1 ≤ 1 / β 2 ≤ … ≤ 1 / β d ≤ 1 / α d , which says rev d g ≺ rev d f . In (c) the chain β d ≤ α d − 1 ≤ β d − 1 ≤ … ≤ α 1 ≤ β 1 becomes 1 / β 1 ≤ 1 / α 1 ≤ … ≤ 1 / α d − 1 ≤ 1 / β d , which says rev d − 1 f ≺ rev d g . The converses follow by applying the same argument to the reversed polynomials.    □
Lemma 6
(Sequences [1] [Lemma 2.2], [8,17]). Let ( p 1 , … , p m ) be an interlacing sequence.
(a)
The sequence ( p 1 + p 2 + … + p m , p 2 + … + p m , … , p m − 1 + p m , p m ) is interlacing.
(b)
For λ 1 , … , λ m ≥ 0 , not all zero, ∑ i λ i p i is real-rooted, is interlaced by p 1 and interlaces p m .
(c)
p 1 + … + p m − 1 ≺ p 1 + 2 p 2 + … + m p m .
Proof. (a) Let s i = p i + … + p m and i < j . Every p l with l < j interlaces every p k with k ≥ j , so u = p i + … + p j − 1 ≺ s j by Lemma 4(a),(b), and then s i = u + s j ≺ s j by Lemma 4(c). (b) p 1 ≺ p i ≺ p m for all i; apply Lemma 4(a),(b). (c) We first show, by induction on m, that s = p 1 + … + p m ≺ c 1 p 1 + … + c m p m for 0 < c 1 ≤ … ≤ c m . For m = 1 this is p 1 ≺ c 1 p 1 . For m ≥ 2 the sequence ( p 1 , … , p m − 2 , p m − 1 + p m ) is interlacing, by Lemma 4, and has sum s; by induction s ≺ c 1 p 1 + … + c m − 2 p m − 2 + c m − 1 ( p m − 1 + p m ) . Also s ≺ p m , since every term does. Adding ( c m − c m − 1 ) p m preserves the relation by Lemma 4(b). Now p 1 + … + p m − 1 ≺ p 1 + 2 p 2 + … + ( m − 1 ) p m − 1 by the claim, and p 1 + … + p m − 1 ≺ m p m ; add the two.    □

3. The Antiprism Triangulation of a Simplex

3.1. The Recurrence

The polynomials h n = h ( sd A ( σ n ) , x ) satisfy h 0 = 1 and
h n ( x ) = ∑ k = 0 n − 1 n k x k h k ( 1 / x ) ( n ≥ 1 )
[1] [Proposition 5.1]. The first values are
h 1 = 1 , h 2 = 1 + 2 x , h 3 = 1 + 9 x + 3 x 2 , h 4 = 1 + 28 x + 42 x 2 + 4 x 3 .
By induction on n, (1) shows that h n has nonnegative integer coefficients, h n ( 0 ) = 1 , degree n − 1 and leading coefficient n for n ≥ 1 : the term k = 0 contributes the constant 1, the terms k ≥ 1 contribute x k h k ( 1 / x ) = x rev k − 1 h k , of degree k and with zero constant term, and only k = n − 1 reaches degree n − 1 , with coefficient n h n − 1 ( 0 ) = n . Since h n ( 0 ) ≠ 0 , the reverse rev n − 1 h n has degree n − 1 , constant term n and leading coefficient 1.4

3.2. The Symmetric Part

Every polynomial p of degree at most n − 1 can be written uniquely as p = a + x b with rev n − 1 a = a and rev n − 2 b = b , namely
b = p − rev n − 1 p x − 1 , a = p − x b = p − x rev n − 1 p 1 − x .
For p = h n we write a n and b n . By the previous paragraph, h n − rev n − 1 h n has degree n − 1 and leading coefficient n − 1 , so b n has degree n − 2 and leading coefficient n − 1 for n ≥ 2 , while b 1 = 0 . In Section 5 we shall see that a n = h ( sd A ( ∂ σ n ) , x ) and hence θ n = x b n ; this is also shown in [1] [Section 5]. The polynomials b n for n ≤ 9 are listed in Table 1.

3.3. The Interlacing Sequence of Athanasiadis, Brunink and Juhnke-Kubitzke

For n , r ≥ 0 put
q n , r ( x ) : = ∑ k = 0 n n k x k + r h k + r ( 1 / x ) .
Then q n , r has nonnegative coefficients and degree n + r with positive leading coefficient, and q n , r ( 0 ) equals 1 if r = 0 and 0 if r ≥ 1 . By (1) and Pascal’s rule,
q n , 0 = h n + x n h n ( 1 / x ) , q 0 , r = x r h r ( 1 / x ) ,
q n , r = q n − 1 , r + q n − 1 , r + 1 ( n ≥ 1 ) .
Iterating (4) along the antidiagonal gives, for 0 ≤ r ≤ n ,
q n − r , r = q n − r − 1 , r + q n − r − 2 , r + 1 + … + q 0 , n − 1 + q 0 , n ,
and taking r = 0 and comparing with (3) and (1),
q n , 0 + q n − 1 , 1 + … + q 0 , n = h n + 1 .
Indeed, the left side of (6) is ∑ r ∑ k n − r k x k + r h k + r ( 1 / x ) = ∑ m = 0 n n + 1 m x m h m ( 1 / x ) by Vandermonde’s identity ∑ r n − r m − r = n + 1 m , and this is h n + 1 by (1).
Theorem 5
([1] [Theorem 5.2]). For every n ≥ 0 the sequence
Q n : = ( q n , 0 , q n − 1 , 1 , … , q 1 , n − 1 , q 0 , n , q 0 , n + 1 )
is interlacing. In particular h n is real-rooted with negative roots, h n ≺ h n + 1 , and h n ≺ x n h n ( 1 / x ) .
Proof. 
We include the short proof of [1]. Q 0 = ( 1 , x ) is interlacing. Let n ≥ 1 and assume that Q n − 1 is interlacing. By (5), the first n + 1 terms of Q n are the partial sums of the reverse of Q n − 1 , so they form an interlacing sequence by Lemma 6(a). By Lemma 2(c) it remains to show that q 0 , n ≺ q 0 , n + 1 and q n , 0 ≺ q 0 , n + 1 . By (6) the polynomial h n + 1 is the sum of the first n + 1 terms of Q n , so q n , 0 ≺ h n + 1 by Lemma 6(b). Applying Lemma 6(c) to Q n − 1 = ( p 1 , … , p n + 1 ) , the sum p 1 + … + p n is h n by (6) with n − 1 in place of n, and ∑ i i p i = ∑ r q n − r , r = h n + 1 by (5) and (6); hence h n ≺ h n + 1 . All polynomials involved have nonnegative coefficients, so their roots are nonpositive, and h n , h n + 1 , q n , 0 have nonzero constant terms, so their roots are negative. Now q 0 , n = x rev n − 1 h n and q 0 , n + 1 = x rev n h n + 1 , so q 0 , n ≺ q 0 , n + 1 is equivalent to h n ≺ h n + 1 by Lemma 2(a) and Lemma 5(c). Since rev n q n , 0 = q n , 0 , the relation q n , 0 ≺ x rev n h n + 1 is equivalent to rev n h n + 1 ≺ q n , 0 by Lemma 5(a), and to q n , 0 ≺ h n + 1 by Lemma 5(b). This completes the induction. The last two terms of Q n give x rev n − 1 h n ≺ x rev n h n + 1 , that is, h n ≺ h n + 1 , and the sum of the first n terms of Q n − 1 interlaces its last term, that is, h n ≺ x n h n ( 1 / x ) .    □

4. Proof of Theorem 4

We prove that b n + 1 ≺ h n for every n ≥ 1 . For n = 1 both polynomials equal 1, so let n ≥ 2 be fixed. Write Q n − 1 = ( f 0 , f 1 , … , f n ) , that is,
f j = q n − 1 − j , j ( 0 ≤ j ≤ n − 1 ) , f n = q 0 , n = x R , R : = rev n − 1 h n ,
and let P i = f 0 + f 1 + … + f i − 1 for 1 ≤ i ≤ n . We record the properties we need.
  • ( f 0 , … , f n ) is interlacing (Theorem 5), all f j have nonnegative coefficients, deg f j = n − 1 for j < n , and f 0 ( 0 ) = 1 .
  • P n = h n , by (6) with n − 1 in place of n.
  • h n + 1 = ∑ j = 0 n ( j + 1 ) f j , by (5) and (6).
  • R has degree n − 1 , leading coefficient 1 and negative roots, and rev n − 1 R = h n .
Each P i has degree n − 1 , nonnegative coefficients and P i ( 0 ) = f 0 ( 0 ) = 1 . Hence rev n − 1 P i has degree n − 1 and leading coefficient 1. For a polynomial P of degree at most n − 1 put
a ( P ) : = x rev n − 1 P − P x − 1 ,
which is a polynomial because the numerator vanishes at x = 1 .

Step 1: A Decomposition of b n + 1

Lemma 7.
b n + 1 = ∑ i = 1 n a ( P i ) .
Proof. 
Let S ( p ) = x rev n − 1 p , a linear map on polynomials of degree at most n − 1 , and H = h n + 1 = ∑ j = 0 n ( j + 1 ) f j by (P3). For j < n we have deg f j ≤ n − 1 and therefore rev n f j = S ( f j ) . Moreover rev n f n = rev n ( x R ) = rev n − 1 R = h n = P n and f n = x R = S ( P n ) . Hence
rev n H = ∑ j < n ( j + 1 ) S ( f j ) + ( n + 1 ) P n .
On the other hand ∑ i = 1 n P i = ∑ j < n ( n − j ) f j , so ∑ i = 1 n S ( P i ) − P i = ∑ j < n ( n − j ) S ( f j ) − f j . Subtracting,
H − rev n H − ∑ i = 1 n S ( P i ) − P i = ( n + 1 ) ∑ j < n f j + f n − ∑ j < n S ( f j ) − P n = ( n + 1 ) f n − S ( P n ) = 0 .
Dividing H − rev n H = ∑ i ( S ( P i ) − P i ) by x − 1 and using (2) with n + 1 in place of n gives the claim.    □

Step 2: The Partial Sums

Lemma 8.
P i ≺ P i + 1 for 1 ≤ i < n , and P i ≺ h n for 1 ≤ i ≤ n .
Proof. 
For j < i ≤ k we have f j ≺ f k . By Lemma 4(a), P i ≺ f k for every k ≥ i . Taking k = i and applying Lemma 4(c) gives P i ≺ P i + f i = P i + 1 . Summing over i ≤ k ≤ n − 1 with Lemma 4(b) gives P i ≺ f i + … + f n − 1 = : T i when i < n , and then P i ≺ P i + T i = P n = h n by Lemma 4(c). For i = n the claim is h n ≺ h n .    □

Step 3: The Reverse of h n Starts the Chain

Lemma 9.
R ≺ P i for 1 ≤ i ≤ n , and consequently rev n − 1 P i ≺ h n for 1 ≤ i ≤ n .
Proof. 
First, R ≺ P 1 = f 0 . Indeed, f 0 ≺ f n = x R by (P1), and this is equivalent to R ≺ f 0 by Lemma 5(a), since both R and f 0 have nonnegative coefficients and real roots. Second, R ≺ P n = h n : by Lemma 4(a) the sum P n of f 0 , … , f n − 1 interlaces f n = x R , and h n ≺ x R is equivalent to R ≺ h n by Lemma 5(a). By Lemma 8, R ≺ P 1 ≺ P 2 ≺ … ≺ P n and R ≺ P n , so the sequence ( R , P 1 , … , P n ) is interlacing by Lemma 2(c). In particular R ≺ P i . Both polynomials have degree n − 1 and negative roots, so Lemma 5(b) turns R ≺ P i into rev n − 1 P i ≺ rev n − 1 R = h n .    □

Step 4: The Sign Count

Lemma 10.
a ( P i ) ≺ h n for 1 ≤ i ≤ n .
Proof. 
Let c = c h n and g = h n ∘ = h n / c , with simple roots β 1 > … > β d , all negative. By Lemmas 8 and 9, P i ≺ h n and rev n − 1 P i ≺ h n , so c divides P i and rev n − 1 P i , and the quotients p = P i / c and r = rev n − 1 P i / c satisfy p ≺ g and r ≺ g . By Lemma 3(i),
( − 1 ) k − 1 p ( β k ) ≥ 0 , ( − 1 ) k − 1 r ( β k ) ≥ 0 ( 1 ≤ k ≤ d ) .
Since c ( 1 ) ≠ 0 , the polynomial c divides a ( P i ) and
( x − 1 ) a ( P i ) c = x r − p .
At x = β k < 0 this gives
( − 1 ) k − 1 a ( P i ) c ( β k ) = β k ( − 1 ) k − 1 r ( β k ) − ( − 1 ) k − 1 p ( β k ) β k − 1 ≥ 0 ,
because the numerator is a nonpositive number minus a nonnegative number and the denominator is negative. Finally, x rev n − 1 P i − P i has degree n and leading coefficient P i ( 0 ) = 1 , so a ( P i ) has degree n − 1 = deg h n and leading coefficient 1, and a ( P i ) / c has degree d. Lemma 3(i) gives a ( P i ) / c ≺ g , hence a ( P i ) ≺ h n .    □
Figure 2. The partial sums P 1 , … , P 5 = h 5 for n = 5 , scaled by ( 1 + | t | ) − 4 on the negative axis, against the roots β 1 > … > β 4 of h 5 . All P i have the sign ( − 1 ) k − 1 at β k , which is the sign condition of Lemma 3(i) behind P i ≺ h 5 .
Figure 2. The partial sums P 1 , … , P 5 = h 5 for n = 5 , scaled by ( 1 + | t | ) − 4 on the negative axis, against the roots β 1 > … > β 4 of h 5 . All P i have the sign ( − 1 ) k − 1 at β k , which is the sign condition of Lemma 3(i) behind P i ≺ h 5 .
Preprints 244175 g002

Step 5: Conclusion

By Lemmas 7 and 10 and Lemma 4(a),
b n + 1 = ∑ i = 1 n a ( P i ) ≺ h n .
Both polynomials have degree n − 1 , so the largest root of b n + 1 is at most the largest root of h n , which is negative; thus all roots of b n + 1 are negative, and since its leading coefficient n is positive, all its coefficients are positive. By Lemma 5(a), b n + 1 ≺ h n is equivalent to h n ≺ x b n + 1 . Since θ n + 1 = x b n + 1 (Lemma 11(e) below), this is Conjecture 2 for n + 1 , and Theorem 4 is proved.□
Example 1
(The case n = 3 ). Here Q 2 = ( f 0 , f 1 , f 2 , f 3 ) = ( 1 + 4 x + x 2 , 3 x + x 2 , 2 x + x 2 , 3 x + 9 x 2 + x 3 ) [1] [Table 1], so
P 1 = 1 + 4 x + x 2 , P 2 = 1 + 7 x + 2 x 2 , P 3 = 1 + 9 x + 3 x 2 = h 3 ,
and R = rev 2 h 3 = 3 + 9 x + x 2 , with f 3 = x R . The reversed partial sums are 1 + 4 x + x 2 , 2 + 7 x + x 2 and 3 + 9 x + x 2 , and
a ( P 1 ) = 1 + 4 x + x 2 , a ( P 2 ) = 1 + 6 x + x 2 , a ( P 3 ) = 1 + 7 x + x 2 .
Their sum is 3 + 17 x + 3 x 2 = b 4 , as Lemma 7 asserts. The roots of h 3 are β 1 ≈ − 0.1156 and β 2 ≈ − 2.8844 . Each a ( P i ) is positive at β 1 and negative at β 2 ; for instance a ( P 2 ) ( β 1 ) ≈ 0.320 and a ( P 2 ) ( β 2 ) ≈ − 7.987 . So each a ( P i ) interlaces h 3 , and so does b 4 , whose roots − 5.4843 and − 0.1823 satisfy − 5.4843 < β 2 < − 0.1823 < β 1 . Multiplying by x, h 3 = 1 + 9 x + 3 x 2 interlaces θ 4 = 3 x + 17 x 2 + 3 x 3 .
Remark 1.
The proof uses only Theorem 5 and the recurrence (1); it never refers to the combinatorics of sd A . The polynomial a ( P ) is the part b of the symmetric decomposition, with respect to n, of the polynomial x rev n − 1 P , so Lemma 7 says that the symmetric part of h n + 1 is assembled from the symmetric parts of the reversed partial sums x rev n − 1 P i . Figure 3 summarizes the argument.
Figure 3. Structure of the proof of Theorems 4 and 3. Here f ≺ g means that f interlaces g.
Figure 3. Structure of the proof of Theorems 4 and 3. Here f ≺ g means that f interlaces g.
Preprints 244175 g003

5. All Simplicial Complexes

5.1. The Transformation of the H-Polynomial

For m ≥ 0 let f m ∘ ( x ) = ∑ j f m , j ∘ x j , where f m , j ∘ is the number of ( j − 1 ) -dimensional faces of sd A ( σ m ) that do not lie in sd A ( ∂ σ m ) , the interior faces; thus f 0 ∘ = 1 counts the empty face. The antiprism triangulation is compatible with restriction: for every face F of Δ the faces of sd A ( Δ ) with carrier F are the interior faces of sd A ( 2 F ) ≅ sd A ( σ | F | ) [1] [Section 4], [3] [Section 6]. Sorting the faces of sd A ( Δ ) by their carriers gives
f ( sd A ( Δ ) , x ) = ∑ F ∈ Δ f | F | ∘ ( x ) = E f ( Δ , x ) ,
where f ( Δ , x ) = ∑ i f i − 1 ( Δ ) x i and E is the linear map with E ( x m ) = f m ∘ . This is the uniform structure studied by Athanasiadis [2]. For a polynomial φ of degree at most n put T n ( φ ) ( x ) : = ( 1 − x ) n φ x / ( 1 − x ) , so that h ( Δ , x ) = T n f ( Δ , x ) for an ( n − 1 ) -dimensional complex Δ . For 0 ≤ k ≤ n define
p n , k : = T n E x k ( 1 + x ) n − k .
Since f ( Δ , x ) = ∑ k h k ( Δ ) x k ( 1 + x ) n − k and sd A ( Δ ) has dimension n − 1 , (7) and linearity give
h ( sd A ( Δ ) , x ) = ∑ k = 0 n h k ( Δ ) p n , k ( x ) = : D n h ( Δ , x ) .
The polynomials p n , k coincide with those of [2] [Theorem 4.1]. We use the reciprocity theorem for balls: if Γ triangulates an ( n − 1 ) -dimensional ball and f ( int Γ , x ) = f ( Γ , x ) − f ( ∂ Γ , x ) counts its interior faces, then
T n f ( int Γ , x ) = x n h ( Γ , 1 / x )
[16] [Section II.7], [15] [Lemma 6.2], [2] [Proposition 2.1].
Lemma 11.
For n ≥ 1 :
(Lemma 11.)
p n , 0 = h n ;
(b)
p n , k = p n , k − 1 + ( x − 1 ) p n − 1 , k − 1 for 1 ≤ k ≤ n ;
(c)
p n , n = x n h n ( 1 / x ) ;
(d)
∑ k = 0 n − 1 p n − 1 , k = h ( sd A ( ∂ σ n ) , x ) ;
(e)
h ( sd A ( ∂ σ n ) , x ) = a n , and hence θ n = x b n .
Proof. (a) E ( ( 1 + x ) n ) = E ( f ( σ n , x ) ) = f ( sd A ( σ n ) , x ) . (b) From x k − 1 ( 1 + x ) n − k + 1 = x k ( 1 + x ) n − k + x k − 1 ( 1 + x ) n − k and T n ( φ ) = ( 1 − x ) T n − 1 ( φ ) for deg φ ≤ n − 1 we get p n , k − 1 = p n , k + ( 1 − x ) p n − 1 , k − 1 . (c) E ( x n ) = f n ∘ = f ( int sd A ( σ n ) , x ) ; apply (9). (d) ∂ σ n has dimension n − 2 and f ( ∂ σ n , x ) = ( 1 + x ) n − x n = ∑ k = 0 n − 1 x k ( 1 + x ) n − 1 − k , so its h-vector consists of ones; apply (8). (e) Let Γ = sd A ( σ n ) , a triangulated ball with boundary sd A ( ∂ σ n ) of dimension n − 2 . By (9), x n h n ( 1 / x ) = T n f ( Γ ) − T n f ( ∂ Γ ) = h n − ( 1 − x ) h ( ∂ Γ , x ) . Since x n h n ( 1 / x ) = x rev n − 1 h n , this says h ( ∂ Γ , x ) = ( h n − x rev n − 1 h n ) / ( 1 − x ) = a n by (2). Then θ n = h n − a n = x b n .    □
Put p n − 1 , n : = θ n = x b n . Summing (b) and using (a), (d), (e) gives the recurrence of [2] [Lemma 6.3],
p n , k = x ∑ i = 0 k − 1 p n − 1 , i + ∑ i = k n p n − 1 , i ( 0 ≤ k ≤ n ) .
Indeed, p n , k = h n + ( x − 1 ) ∑ i < k p n − 1 , i and h n = a n + θ n = ∑ i = 0 n p n − 1 , i .

5.2. The Rotation Lemma

Lemma 12.
Let m ≥ 1 and let ( g 0 , … , g m ) be an interlacing sequence of nonzero polynomials with nonnegative coefficients. For 0 ≤ k ≤ m + 1 put
G k = x ∑ i < k g i + ∑ i ≥ k g i .
Then ( G 0 , G 1 , … , G m + 1 ) is an interlacing sequence of polynomials with nonnegative coefficients.
Proof. 
Note G m + 1 = x G 0 . Since g 0 interlaces every g i , it interlaces G 0 = ∑ g i , so G 0 is real-rooted with nonpositive roots, and G 0 ≺ x G 0 by Lemma 5(a) applied to G 0 ≺ G 0 . By Lemma 2(c) it suffices to show G k ≺ G k + 1 for 0 ≤ k ≤ m . Fix k and put U = x ∑ i < k g i + ∑ i > k g i , which is nonzero because m ≥ 1 . Then
G k = U + g k , G k + 1 = U + x g k = ( 1 − x ) U + x G k .
For i > k we have g k ≺ g i , and also g i ≺ x g k by Lemma 5(a). For i < k we have g i ≺ g k , hence g k ≺ x g i by Lemma 5(a) and x g i ≺ x g k by Lemma 2(a). Thus every summand of U is interlaced by g k and interlaces x g k , so g k ≺ U ≺ x g k by Lemma 4. By Lemma 4(c), g k ≺ G k ≺ U ; in particular G k is real-rooted, and its roots are nonpositive. Let c = c G k . As G k ≺ U , c divides U, hence also G k + 1 = ( 1 − x ) U + x G k . If φ 1 > … > φ d are the roots of G k / c , then
G k + 1 c ( φ j ) = ( 1 − φ j ) U c ( φ j ) , 1 − φ j > 0 ,
and ( − 1 ) j ( U / c ) ( φ j ) ≥ 0 by Lemma 3(ii) applied to G k / c ≺ U / c . For the degrees: g k ≺ U gives deg G k = deg U , and U ≺ x g k gives deg G k + 1 = deg ( x g k ) ∈ { deg U , deg U + 1 } . Lemma 3(ii) yields G k / c ≺ G k + 1 / c , that is, G k ≺ G k + 1 .    □

5.3. Proof of Theorem  3

For n ≥ 1 let Q n = ( p n , 0 , p n , 1 , … , p n , n ) . We show by induction on n that Q n is an interlacing sequence of nonzero polynomials with nonnegative coefficients. For n = 1 , Lemma 11(a),(c) give Q 1 = ( 1 , x ) . Let n ≥ 2 and consider
P n = ( p n − 1 , 0 , … , p n − 1 , n − 1 , θ n ) .
Its first n terms form Q n − 1 , which is interlacing with nonnegative coefficients by induction; θ n = x b n has positive coefficients by Theorem 4. We have p n − 1 , 0 = h n − 1 ≺ θ n by Theorem 4. Further, p n − 1 , n − 1 = x rev n − 2 h n − 1 by Lemma 11(c), and b n ≺ h n − 1 (Theorem 4) gives rev n − 2 h n − 1 ≺ rev n − 2 b n = b n by Lemma 5(b), hence p n − 1 , n − 1 ≺ θ n by Lemma 2(a). By Lemma 2(c), P n is interlacing. Lemma 12 applied to P n produces G 0 , … , G n + 1 with G k = p n , k for 0 ≤ k ≤ n by (10), so Q n is interlacing with nonnegative coefficients.
Now let Δ be ( n − 1 ) -dimensional with nonnegative h-vector. By (8), h ( sd A ( Δ ) , x ) is a nonnegative combination of Q n , not identically zero because h 0 ( Δ ) = 1 . Lemma 6(b) shows that it is real-rooted, interlaced by p n , 0 = h n , and interlaces p n , n = x n h n ( 1 / x ) . This proves Theorem 3.□
The same argument proves that D n ( h ) is real-rooted for every polynomial h of degree at most n with nonnegative coefficients, whether or not it is the h-polynomial of a complex.

6. Consequences

6.1. Cohen–Macaulay Complexes

If Δ is Cohen–Macaulay over some field, its h-vector is nonnegative [16] [Chapter II], so h ( sd A ( Δ ) , x ) is real-rooted with nonpositive roots. The f-polynomial f ( sd A ( Δ ) , x ) = ( 1 + x ) n h ( sd A ( Δ ) , x / ( 1 + x ) ) is then real-rooted as well: each linear factor x − s of h with s ≤ 0 becomes ( 1 − s ) x − s . By Newton’s inequalities the coefficients of both polynomials are log-concave with no internal zeros, hence unimodal. This proves Corollary 1(a) and strengthens the unimodality of the h-vector of sd A ( Δ ) proved in [1] [Theorem 1.3].

6.2. Homology Spheres

If Δ is a homology sphere, then so is sd A ( Δ ) , and h ( sd A ( Δ ) , x ) is symmetric of degree n by the Dehn–Sommerville equations. A symmetric real-rooted polynomial h with nonnegative coefficients has its roots in pairs { s , 1 / s } with s < 0 , together with the root − 1 ; each pair contributes
( x − s ) ( x − 1 / s ) = ( 1 + x ) 2 + c x , c = − s + 1 s − 2 ≥ 0 .
Hence h = λ ( 1 + x ) n − 2 r ∏ i = 1 r ( 1 + x ) 2 + c i x with λ , c i ≥ 0 , and substituting u = x / ( 1 + x ) 2 gives
∑ j γ j u j = λ ∏ i = 1 r ( 1 + c i u ) ,
which is real-rooted with nonpositive roots. This proves Corollary 1(b).

6.3. The Strong Interlacing Property

In the language of Athanasiadis and Tzanaki [4] [Definition 3.2], a uniform triangulation has the strong interlacing property if h m is real-rooted for 2 ≤ m < n and θ m is zero or a real-rooted polynomial of degree m − 1 with nonnegative coefficients interlaced by h m − 1 for 2 ≤ m ≤ n . For the antiprism triangulation, the first condition is Theorem 5, and the second is Theorem 4 with Lemma 11(e). Thus the antiprism triangulation has the strong interlacing property for every n, which was the open part of [1] [Section 5] recorded in [4] [Section 3]. The main results of [4] [Theorems 4.1 and 6.1] therefore apply. In particular, if Γ is an n-dimensional complex with nonnegative h-vector and Δ is its ( n − 1 ) -skeleton, then h ( sd A ( Δ ) , x ) has a nonnegative, real-rooted and interlacing symmetric decomposition with respect to n, and it interlaces h ( sd A ( Γ ) , x ) [4] [Theorem 6.1]. This is Corollary 1(c).

6.4. The Boundary of the Simplex

The sequence P n of Section 5 is interlacing, and its first term is h n − 1 . By Lemma 6(b), h n − 1 interlaces the sum of its first n terms, which is a n = h ( sd A ( ∂ σ n ) , x ) by Lemma 11(d),(e), and the sum of all its terms, which is h n .
Corollary 2.
For every n ≥ 2 the polynomial h n − 1 interlaces a n = h ( sd A ( ∂ σ n ) , x ) , θ n and h n , and b n interlaces h n − 1 .
The relation h n − 1 ≺ h n is part of [1] [Theorem 5.2]; the relations with a n and θ n are new.

7. Examples and Tables

Example 2
(The octahedron). Let Δ be the boundary of the octahedron, a 2-sphere with h ( Δ , x ) = 1 + 3 x + 3 x 2 + x 3 . From the values p 3 , k in Table 2,
h ( sd A ( Δ ) , x ) = p 3 , 0 + 3 p 3 , 1 + 3 p 3 , 2 + p 3 , 3 = 1 + 51 x + 51 x 2 + x 3 = ( 1 + x ) ( 1 + 50 x + x 2 ) ,
with roots − 1 and − 25 ± 624 , that is, approximately − 49.980 , − 1 and − 0.020 . Here γ ( u ) = 1 + 48 u . We checked this value independently by counting the faces of sd A ( Δ ) from (7).
Example 3
(Two triangles). Let Δ consist of two triangles sharing an edge, a 2-ball with f ( Δ , x ) = 1 + 4 x + 5 x 2 + 2 x 3 and h ( Δ , x ) = 1 + x . Then h ( sd A ( Δ ) , x ) = p 3 , 0 + p 3 , 1 = 1 + 17 x + 8 x 2 , with roots ( − 17 ± 257 ) / 16 .
Example 4
(Three-spheres). For the boundary of the 4-simplex, h ( Δ , x ) = 1 + x + x 2 + x 3 + x 4 and h ( sd A ( Δ ) , x ) = a 5 = 1 + 71 x + 231 x 2 + 71 x 3 + x 4 , with γ ( u ) = 1 + 67 u + 91 u 2 . For the boundary of the 4-dimensional cross-polytope, h ( Δ , x ) = ( 1 + x ) 4 and h ( sd A ( Δ ) , x ) = 1 + 212 x + 774 x 2 + 212 x 3 + x 4 , with γ ( u ) = 1 + 208 u + 352 u 2 . Both γ-polynomials have positive discriminant, as Corollary 1(b) predicts.
Figure 4. Roots of h n and nonzero roots of θ n + 1 on a logarithmic scale. The alternation in each row is the interlacing h n ≺ θ n + 1 of Theorem 4.
Figure 4. Roots of h n and nonzero roots of θ n + 1 on a logarithmic scale. The alternation in each row is the interlacing h n ≺ θ n + 1 of Theorem 4.
Preprints 244175 g004
Two features of Table 3 deserve comment. Since b n is symmetric, its roots come in reciprocal pairs, and for odd n it has odd degree and therefore the root − 1 . In the table the largest root of h n approaches 0 and the smallest one decreases quickly; by Theorem 3 the roots of every h ( sd A ( Δ ) , x ) with dim Δ = n − 1 interlace those of h n .

8. The Local H-Polynomial of the Simplex

The restriction of sd A ( σ n ) to a face of σ n with i vertices is the antiprism triangulation of that face. Hence the local h-polynomial ℓ n ( x ) : = ℓ A ( σ n , x ) of Stanley [15] satisfies
h n = ∑ i = 0 n n i ℓ i , ℓ n = ∑ i = 0 n ( − 1 ) n − i n i h i .
The first values are ℓ 0 = 1 , ℓ 1 = 0 , ℓ 2 = 2 x , ℓ 3 = 3 x + 3 x 2 and ℓ 4 = 4 x + 30 x 2 + 4 x 3 . By [15] [Theorem 3.3] the polynomial ℓ n is symmetric, x n ℓ n ( 1 / x ) = ℓ n , and for n ≥ 2 it has degree n − 1 , zero constant term and leading coefficient n.5 Write ℓ n = ∑ k n k ℓ ¯ ( n , k ) x k and ℓ ¯ n = ∑ k ℓ ¯ ( n , k ) x k .
Conjecture 6
([1] [Conjecture 5.8]). For every n the polynomial ℓ n is real-rooted and interlaces ℓ n + 1 , and ℓ ¯ n is real-rooted and interlaces ℓ ¯ n + 1 .
Theorem 4 says that h n − 1 interlaces θ n . This section shows that the first half of Conjecture 6 is equivalent to an interlacing relation of the same shape, h n ≺ ℓ n + 1 , taken together with ℓ n ≺ ℓ n + 1 , and gives two identities that express ℓ n + 1 through h 0 , … , h n .

8.1. Generating Functions

Put
E ℓ = ∑ n ≥ 0 ℓ n t n n ! , E h = ∑ n ≥ 0 h n t n n ! , E ˜ h = ∑ n ≥ 0 x n h n ( 1 / x ) t n n ! .
Proposition 1.
E h = e t E ℓ , E ˜ h = e x t E ℓ and
E ℓ = 1 1 − ( e t − 1 ) ( e x t − 1 ) .
Consequently ℓ ¯ ( n , k ) = ∑ j ≥ 0 ( j ! ) 2 S ( k , j ) S ( n − k , j ) , where S ( m , j ) denotes a Stirling number of the second kind.
Proof. 
The first identity is (11). By symmetry, x n h n ( 1 / x ) = ∑ i n i x n ℓ i ( 1 / x ) = ∑ i n i x n − i ℓ i , which is the second identity. The recurrence (1) says that E h = 1 + ( e t − 1 ) E ˜ h . Substituting the first two identities gives E ℓ ( e t − e ( 1 + x ) t + e x t ) = 1 , and e t − e ( 1 + x ) t + e x t = 1 − ( e t − 1 ) ( e x t − 1 ) . Finally, E ℓ = ∑ j ( e t − 1 ) j ( e x t − 1 ) j and ( e t − 1 ) j = j ! ∑ m S ( m , j ) t m / m ! ; comparing coefficients of x k t n / n ! gives the formula for ℓ ¯ ( n , k ) .    □
The recurrence [1] [(18)], ℓ n = ∑ k < n n k ℓ k ( 1 + x ) n − k − 1 − x n − k , is the coefficientwise form of Proposition 1. Its kernels ( 1 + x ) m − 1 − x m have non-real roots for m ≥ 4 . Differentiating instead gives a convolution in which every factor is real-rooted. For j ≥ 1 put
S j : = h j + x j − 1 h j ( 1 / x ) = h j + rev j − 1 h j .
By Theorem 5 and Lemma 5(a) we have rev j − 1 h j ≺ h j , so S j is real-rooted by Lemma 4(c); it is symmetric with center of symmetry ( j − 1 ) / 2 .
Theorem 7.
For every n ≥ 1 ,
ℓ n + 1 = x ∑ j = 1 n n j ℓ n − j S j , h n + 1 = h n + x ∑ j = 1 n n j h n − j S j .
Proof. 
Differentiating 1 / E ℓ = 1 − ( e t − 1 ) ( e x t − 1 ) gives
E ℓ ′ = E ℓ · e t ( e x t − 1 ) E ℓ + x e x t ( e t − 1 ) E ℓ .
By Proposition 1, x e x t ( e t − 1 ) E ℓ = x ( e t − 1 ) E ˜ h = x ( E h − 1 ) . Also e t ( e x t − 1 ) E ℓ = E ˜ h − 1 , since the difference of the two sides is E ℓ ( e ( 1 + x ) t − e t − e x t ) + 1 = 0 . Hence E ℓ ′ = E ℓ G with
G = x ( E h − 1 ) + ( E ˜ h − 1 ) = ∑ j ≥ 1 x h j + x j h j ( 1 / x ) t j j ! = ∑ j ≥ 1 x S j t j j ! .
Comparing coefficients of t n / n ! in E ℓ ′ = E ℓ G gives the first identity. Since E h = e t E ℓ , we get E h ′ = E h + e t E ℓ ′ = E h ( 1 + G ) , which gives the second.    □
In other words, log E ℓ has the exponential coefficients 0 , x S 1 , x S 2 , … : the local h-polynomials are the moments whose cumulants are x S m − 1 . For example, ℓ 4 = x 3 ℓ 2 S 1 + 3 ℓ 1 S 2 + ℓ 0 S 3 = x 12 x + ( 4 + 18 x + 4 x 2 ) , with S 1 = 2 and S 3 = 4 + 18 x + 4 x 2 . Every summand of Theorem 7 is real-rooted whenever ℓ n − j is, but for 3 ≤ n ≤ 13 none of the nonzero summands with j < n interlaces h n , so the identity does not by itself give real-rootedness.

8.2. A Decomposition Along an Interlacing Chain

For m , r ≥ 0 put
s m , r : = ∑ i = 0 m m i ℓ i + r ,
the local analogue of the polynomials q m , r of Section 3. Then s m , 0 = h m by (11), s 0 , r = ℓ r , and Pascal’s rule gives s m , r = s m − 1 , r + s m − 1 , r + 1 for m ≥ 1 . Fix n ≥ 2 and put
f j : = s n − 1 − j , j ( 0 ≤ j ≤ n − 1 ) , f n : = ℓ n , T i : = f i + f i + 1 + … + f n .
All f j have degree at most n − 1 . As in Section 4, a ( P ) = ( x rev n − 1 P − P ) / ( x − 1 ) for a polynomial P of degree at most n − 1 .
Proposition 2.
For n ≥ 2 we have f 0 + … + f n = h n , h n + 1 − ℓ n + 1 = ∑ j = 0 n ( j + 1 ) f j , and
ℓ n + 1 x = b n + 1 + a ( h n ) + ∑ i = 1 n a ( T i ) .
Proof. 
Iterating Pascal’s rule along the antidiagonal gives s n − r , r = ∑ j = r n f j for 0 ≤ r ≤ n ; the case r = 0 is the first identity. Summing over r gives ∑ r = 0 n s n − r , r = ∑ j ( j + 1 ) f j , and the same iteration at level n + 1 gives h n + 1 = s n + 1 , 0 = ∑ r = 0 n s n − r , r + s 0 , n + 1 , which is the second identity.
Let S ( p ) = x rev n − 1 p and L = ℓ n + 1 / x , a symmetric polynomial of degree n − 1 , so that rev n ℓ n + 1 = L and ℓ n + 1 − rev n ℓ n + 1 = ( x − 1 ) L . With h n + 1 − rev n h n + 1 = ( x − 1 ) b n + 1 from (2), the polynomial G = h n + 1 − ℓ n + 1 satisfies G − rev n G = ( x − 1 ) ( b n + 1 − L ) . On the other hand rev n f j = S ( f j ) for every j, so
G − rev n G = ∑ j = 0 n ( j + 1 ) f j − S ( f j ) = h n − S ( h n ) + ∑ i = 1 n T i − S ( T i ) ,
because ∑ i = 1 n T i = ∑ j j f j . Since p − S ( p ) = − ( x − 1 ) a ( p ) , dividing by x − 1 gives b n + 1 − L = − a ( h n ) − ∑ i a ( T i ) .    □
By Theorem 4 the first summand b n + 1 interlaces h n , and since P n = h n , Lemma 10 shows that a ( h n ) interlaces h n . Thus L ≺ h n , which by Lemma 5(a) is the relation h n ≺ ℓ n + 1 , would follow from ∑ i a ( T i ) ≺ h n . This is where the method of Section 4 stops. The tail sums T i vanish at 0; for 3 ≤ n ≤ 12 they are interlaced by h n rather than interlacing it, and none of the real-rooted polynomials a ( T 1 ) , … , a ( T n − 1 ) interlaces h n , while a ( T n ) = a ( ℓ n ) = 0 .

8.3. A Reduction of Conjecture  ??

For n ≥ 1 consider the sequence
A n : = ( s n , 0 , s n − 1 , 1 , … , s 1 , n − 1 , s 0 , n , s 0 , n + 1 ) = ( h n , s n − 1 , 1 , … , s 1 , n − 1 , ℓ n , ℓ n + 1 ) ,
the local analogue of the sequence Q n of Theorem 5. For n ≥ 2 all its terms are nonzero.
Theorem 8.
The sequence A 2 = ( 1 + 2 x , 2 x , 2 x , 3 x + 3 x 2 ) is interlacing. For n ≥ 3 , if A n − 1 is interlacing, then A n is interlacing if and only if
ℓ n ≺ ℓ n + 1 and h n ≺ ℓ n + 1 .
Consequently A n is interlacing for every n ≥ 2 if and only if ℓ n ≺ ℓ n + 1 and h n ≺ ℓ n + 1 for every n ≥ 2 . In that case the first half of Conjecture 6 holds, since for n ≤ 1 there is nothing to prove, and every polynomial s m , r is real-rooted.
Proof. 
The roots of the terms of A 2 are − 1 / 2 , 0, 0 and { 0 , − 1 } , and the relations required by the definition are immediate. Let n ≥ 3 and assume that A n − 1 = ( f 0 , … , f n ) , in the notation of Proposition 2, is interlacing. By the proof of that proposition the first n + 1 terms of A n are s n − r , r = f r + … + f n , the partial sums of A n − 1 taken from the end, so they form an interlacing sequence by Lemma 6(a). The two displayed relations are relations between terms of A n , so they are necessary. Conversely, they provide the last consecutive relation s 0 , n ≺ s 0 , n + 1 and the relation between the first and the last term, and Lemma 2(c) shows that A n is interlacing. The remaining statements follow by induction on n; every s m , r with m + r = n is a term of A n .    □
The proof of Theorem 5 obtained the analogous two relations for Q n from the recurrence (1). For A n they carry the local information, and they are the whole difficulty. The second relation has the shape of Theorem 4, with ℓ n + 1 in place of θ n + 1 ; both θ n + 1 and ℓ n + 1 are x times a symmetric polynomial of degree n − 1 .
Proposition 3.
For 2 ≤ n ≤ 40 ,
ℓ n ≺ ℓ n + 1 , h n ≺ ℓ n + 1 , ℓ ¯ n ≺ ℓ ¯ n + 1 , p ¯ n ≺ p ¯ n + 1 ,
where p ¯ n = ∑ k p ¯ ( n , k ) x k with p ¯ ( n , k ) = ∑ i = 0 n − k n − k i ℓ ¯ ( n − i , k ) . Consequently A n is interlacing for 2 ≤ n ≤ 40 , and the statements of Conjecture 6 and of [1] [Conjecture 5.6] hold for 2 ≤ n ≤ 40 .
Proof. 
The four relations are finite statements about explicit integer polynomials. They were decided by exact Sturm sequences, as described in Section 9. The second assertion follows from Theorem 8.    □

9. Computer Verification

Every statement of the paper is proved in the text. We nevertheless checked the finite content by two independent programs, archived in the public repository https://github.com/creelie/d-dome-enrique.

Lean 4

A Lean 4 project with Mathlib contains a kernel-checked proof of the general identity behind Lemma 7, stated for an arbitrary additive group with two additive maps, and of the sign transfer inequality of Lemma 10: if β < 0 , ( − 1 ) k p ≥ 0 , ( − 1 ) k r ≥ 0 and ( β − 1 ) a = β r − p , then ( − 1 ) k a ≥ 0 . It also verifies, by evaluation in the kernel, the recurrence (1) against the table of [1], and the identities (P2), (P3), f n = x R and Lemma 7 with exact integer arithmetic for 2 ≤ n ≤ 11 . The only axioms used are the standard ones of Lean (propext, Classical.choice, Quot.sound).

Julia

A Julia program computes h n , q n , r , P i and b n + 1 in exact integer arithmetic and decides each interlacing relation by Sturm sequences after removing common factors, without floating point. For every n ≤ 18 it confirms (6), Lemma 7, the degrees and positivity of b n + 1 , all interlacing relations of Q n − 1 , and the relations of Lemmas 8–10 and Theorem 4, 2127 checks in total. Deliberately wrong inputs are rejected.
Two further programs, with the same exact Sturm test, check Section 8. For every 2 ≤ n ≤ 40 the first confirms that ℓ n is symmetric with zero constant term, its expansion in the numbers ℓ ¯ ( n , k ) , the two convolution identities of Theorem 7, the decomposition of ℓ n + 1 / x in Proposition 2, and the relations ℓ n ≺ ℓ n + 1 , h n ≺ ℓ n + 1 and ℓ ¯ n ≺ ℓ ¯ n + 1 . For n ≤ 19 it also checks directly that A n is interlacing, and for n ≤ 12 it checks every pair in A n ; these direct checks are redundant by Theorem 8. It makes 1066 checks. The second program confirms p ¯ n ≺ p ¯ n + 1 for 2 ≤ n ≤ 40 and, for 3 ≤ n ≤ 12 , that no a ( T i ) with i < n interlaces h n while a ( T n ) = 0 , which is the failure of termwise interlacing discussed after Proposition 2; it makes 114 checks.

10. Open Problems

By Theorem 8, the first half of Conjecture 6, together with the interlacing of all sequences A n , comes down to one question.
Question 9.
Does h n interlace ℓ A ( σ n + 1 , x ) , and does ℓ A ( σ n , x ) interlace ℓ A ( σ n + 1 , x ) , for every n ≥ 2 ?
Proposition 2 shows that the method of Section 4 needs one more ingredient here: a chain that controls the tail sums T i and their reversals, in the way that rev n − 1 h n controls the partial sums P i in Lemma 9. Another possible route starts from the identity ℓ A ( σ n , x ) = ∑ k n k θ k d n − k of [1] [Proposition 5.9], where d m is the mth derangement polynomial: by Theorem 4 every θ k = x b k is real-rooted, so each summand is real-rooted, and the question is whether the summands have a common interlacer. More generally, Athanasiadis asked whether ℓ V ( sd A ( Γ ) , x ) is real-rooted for every triangulation Γ of a simplex [3] [Question 6.3]; the case Γ = σ n is Conjecture 6.
A second question is whether the interlacing in Theorem 4 is strict, that is, whether h n − 1 and θ n / x have no common root; in all computed cases they have none.
Simplicial complexes built from covers also appear in the author’s work on iterated nerve complexes [7], and earlier work of the author on group actions and on hypercomplex structures is in [5,6].

Author Contributions

Deep Bhattacharjee is the sole author. He chose the problem, proved the results, wrote the paper, drew the figures, and wrote the Lean 4 and Julia programs that check the finite statements.

Funding

No funding of any kind was received for this work.

Ethics approval

Not applicable. The work involves no human participants, animals or personal data.

Data and code availability

No data were generated. The Lean 4 project and the Julia programs that recompute every numerical statement of the paper are available at https://github.com/creelie/d-dome-enrique. The Lean files are checked by the kernel, with no unproved statement and no axiom beyond the standard ones.

Competing interests

The author declares no competing interests.

Open Gluon and Anaxia

Two tools built by the author organised this project. Open Gluon is a workspace that keeps pinned copies of the mathematical libraries, the source record of the project and its checks in one reproducible place. Anaxia, the engine of the eBLLM project, stores each construction as a route with its input, its output, the obligations that must be discharged and a test that makes the route fail when an obligation is skipped, and it certifies finite statements in exact integer and rational arithmetic. It was used here to keep the polynomial data of Section 3, Section 4, Section 5, Section 6 and Section 7 and their checks in one record. Every statement of this paper is proved in the text.

References

  1. Athanasiadis, Christos A.; Brunink, Jan-Marten; Juhnke-Kubitzke, Martina. Combinatorics of antiprism triangulations. Discret. Comput. Geom. 2022, 68(no. 1), 72–106. [Google Scholar] [CrossRef]
  2. Athanasiadis, Christos A. Face numbers of uniform triangulations of simplicial complexes. Int. Math. Res. Not. 2022, no. 20, 15756–15787. [Google Scholar] [CrossRef]
  3. Athanasiadis, Christos A. Triangulations of simplicial complexes and theta polynomials. Tohoku Math. J. 78 (2026)(no. 3), 423–441. [CrossRef]
  4. Athanasiadis, Christos A.; Tzanaki, Eleni. Symmetric decompositions, triangulations and real-rootedness. Mathematika 2021, 67(no. 4), 840–859. [Google Scholar] [CrossRef]
  5. Bhattacharjee, Deep. Establishing equivariant class [O] for hyperbolic groups. Asian Res. J. Math. 2022, 18(no. 11), 362–369. [Google Scholar] [CrossRef]
  6. Bhattacharjee, Deep. Non-commutativity over canonical suspension η for genus g≥1 in hypercomplex structures for potential ρϕ. Asian Res. J. Math. 2022, 18(no. 11), 332–341. [Google Scholar] [CrossRef]
  7. Bhattacharjee, Deep; Bhattacharya, Ushashi. Iterated nerve complexes for k-fold sensor coverage. Preprints.org 2026. [Google Scholar] [CrossRef]
  8. Brändén, Petter. On linear transformations preserving the Pólya frequency property. Trans. Am. Math. Soc. 2006, 358(no. 8), 3697–3716. [Google Scholar] [CrossRef]
  9. Brändén, Petter. Unimodality, log-concavity, real-rootedness and beyond. In Handbook of Enumerative Combinatorics; CRC Press: Boca Raton, 2015; pp. 461–508. [Google Scholar] [CrossRef]
  10. Brändén, Petter; Solus, Liam. Symmetric decompositions and real-rootedness. Int. Math. Res. Not. 2021, no. 10, 7764–7798. [Google Scholar] [CrossRef]
  11. Brenti, Francesco; Welker, Volkmar. f-vectors of barycentric subdivisions. Math. Z. 2008, 259(no. 4), 849–865. [Google Scholar] [CrossRef]
  12. Gal, Światosław R. Real root conjecture fails for five- and higher-dimensional spheres. Discret. Comput. Geom. 2005, 34(no. 2), 269–284. [Google Scholar] [CrossRef]
  13. Izmestiev, Ivan; Joswig, Michael. Branched coverings, triangulations, and 3-manifolds. Adv. Geom. 2003, 3(no. 2), 191–225. [Google Scholar] [CrossRef]
  14. Kozlov, Dmitry N. Chromatic subdivision of a simplicial complex. Homol. Homotopy Appl. 2012, 14(no. 2), 197–209. [Google Scholar] [CrossRef]
  15. Stanley, Richard P. Subdivisions and local h-vectors. J. Am. Math. Soc. 1992, 5(no. 4), 805–851. [Google Scholar] [CrossRef]
  16. Richard P. Stanley, Combinatorics and Commutative Algebra. In Progress in Mathematics, second ed.; Birkhäuser, Boston, 1996; vol. 41.
  17. Wagner, David G. Total positivity of Hadamard products. J. Math. Anal. Appl. 1992, 163(no. 2), 459–483. [Google Scholar] [CrossRef]
1
The chromatic subdivision of a simplex describes the protocol complex of one round of an immediate snapshot in a shared-memory system; Kozlov [14] gives a combinatorial treatment and shows that it is a subdivision.
2
Gal [12] showed that the h-polynomial of a flag sphere need not be real-rooted in dimension five and higher, so real-rootedness is a special feature of the triangulations considered here and not a general property of flag spheres.
3
We follow the conventions of [1,2]; Wagner [17] and Brändén [9] use the same order of roots.
4
The coefficients of h n have several combinatorial interpretations; for instance, the coefficient of x k is n k times the number of permutations of [ n ] with excedance set [ k ] [1] [Proposition 5.1]. We shall not need them.
5
Athanasiadis, Brunink and Juhnke-Kubitzke proved that ℓ n is γ -positive [1] [Proposition 5.9] and gave several combinatorial interpretations of its coefficients [1] [Proposition 5.7]; for instance, the coefficient of x k counts ordered set partitions ( B 1 , … , B 2 j ) of [ n ] with an even number of blocks such that B 1 ∪ B 3 ∪ … ∪ B 2 j − 1 has k elements.
Figure 1. The antiprism triangulation of the triangle with vertices 1 , 2 , 3 . The three interior vertices u 1 , u 2 , u 3 are the three pointed faces of the whole triangle, each edge carries two vertices, the pointed edges, and the complex has 12 vertices and 13 triangles, the number of ordered set partitions of { 1 , 2 , 3 } . Its h-polynomial is h 3 ( x ) = 1 + 9 x + 3 x 2 .
Figure 1. The antiprism triangulation of the triangle with vertices 1 , 2 , 3 . The three interior vertices u 1 , u 2 , u 3 are the three pointed faces of the whole triangle, each edge carries two vertices, the pointed edges, and the complex has 12 vertices and 13 triangles, the number of ordered set partitions of { 1 , 2 , 3 } . Its h-polynomial is h 3 ( x ) = 1 + 9 x + 3 x 2 .
Preprints 244175 g001
Table 1. The polynomials h n and b n , listed by their coefficients in increasing degree.
Table 1. The polynomials h n and b n , listed by their coefficients in increasing degree.
n h n b n
2 1, 2 1
3 1, 9, 3 2, 2
4 1, 28, 42, 4 3, 17, 3
5 1, 75, 310, 150, 5 4, 79, 79, 4
6 1, 186, 1725, 2300, 465, 6 5, 284, 859, 284, 5
7 1, 441, 8211, 23625, 13685, 1323, 7 6, 888, 6362, 6362, 888, 6
8 1, 1016, 35476, 193256, 241570, 70952, 3556, 8 7, 2547, 38023, 86337, 38023, 2547, 7
9 1, 2295, 143676, 1367100, 3179106, 2050650, 335244, 9180, 9 8, 6893, 198461, 882011, 882011, 198461, 6893, 8
Table 2. The polynomials p n , k for n ≤ 4 , by their coefficients in increasing degree. Each row is an interlacing sequence; the row sums are a n + 1 = h ( sd A ( ∂ σ n + 1 ) , x ) .
Table 2. The polynomials p n , k for n ≤ 4 , by their coefficients in increasing degree. Each row is an interlacing sequence; the row sums are a n + 1 = h ( sd A ( ∂ σ n + 1 ) , x ) .
n p n , 0 , p n , 1 , … , p n , n
1 ( 1 ) , ( 0 , 1 )
2 ( 1 , 2 ) , ( 0 , 3 ) , ( 0 , 2 , 1 )
3 ( 1 , 9 , 3 ) , ( 0 , 8 , 5 ) , ( 0 , 5 , 8 ) , ( 0 , 3 , 9 , 1 )
4 ( 1 , 28 , 42 , 4 ) , ( 0 , 20 , 48 , 7 ) , ( 0 , 12 , 51 , 12 ) , ( 0 , 7 , 48 , 20 ) , ( 0 , 4 , 42 , 28 , 1 )
Table 3. Roots of h n and of b n + 1 (six significant digits). The roots of the two polynomials alternate, the smallest belonging to b n + 1 , as Theorem 4 asserts.
Table 3. Roots of h n and of b n + 1 (six significant digits). The roots of the two polynomials alternate, the smallest belonging to b n + 1 , as Theorem 4 asserts.
n roots of h n roots of b n + 1
2 − 0.5 − 1
3 − 2.88444 , − 0.115563 − 5.48433 , − 0.182338
4 − 9.7874 , − 0.674739 , − 0.0378562 − 18.6965 , − 1 , − 0.0534859
5 − 27.7883 , − 1.93481 , − 0.26278 , − 0.0141559 − 53.6155 , − 2.81003 , − 0.355868 , − 0.0186513
6 − 72.2493 , − 4.36241 , − 0.759957 , − 0.122666 , − 0.00567251 − 140.507 , − 6.32766 , − 1 , − 0.158036 , − 0.00711707
7 − 178.131 , − 8.80588 , − 1.61953 , − 0.378476 , − 0.0626506 , − 0.00237158 − 348.366 , − 12.8288 , − 2.10655 , − 0.47471 , − 0.0779497 , − 0.00287054
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.