Chapter 44: Advanced Algebraic Combinatorics

Why do geometric ideas produce log-concave combinatorial sequences?

A sequence of nonnegative numbers $a_0,\ldots,a_r$ is log-concave if $a_i^2\ge a_{i-1}a_{i+1}$ at each interior index. Positive coefficients alone do not imply this: $1,1,4$ is positive but fails the inequality at its middle term. Geometry can supply something stronger than positivity—a bilinear form whose signs constrain neighboring numbers.

Start with the uniform matroid $U_{3,4}$ on $E=\{1,2,3,4\}$. Its rank function is $\operatorname{rk}(S)=\min(3,\lvert S\rvert)$: every set of at most three elements is independent. Define its characteristic polynomial directly by

$$\chi_M(t)=\sum_{S\subseteq E}(-1)^{\lvert S\rvert}t^{3-\operatorname{rk}(S)}.$$

Grouping subsets by size gives $\chi_M(t)=t^3-4t^2+6t-3$. Since $\chi_M(1)=0$, divide by $t-1$ to get the reduced characteristic polynomial

$$\overline\chi_M(t)=t^2-3t+3.$$

Its alternating coefficient magnitudes are $1,3,3$, and log-concavity says $3^2\ge1\cdot3$. A direct numerical check establishes this instance. To understand the mechanism, consider the symmetric matrix

$$G=\begin{pmatrix}1&3\\ 3&3\end{pmatrix}.$$

It has negative determinant $-6$, so its two eigenvalues have opposite signs. A form with one positive direction behaves differently from a Euclidean inner product: suitable positive vectors satisfy a reversed Cauchy–Schwarz inequality. Here the negative determinant is exactly the desired inequality in disguise. The advanced sections construct an intersection ring whose pairing on two particular classes really is this matrix; it is not merely an arbitrarily chosen numerical analogy.

For the core investigation, it suffices to know the subset formula above, elementary polynomial division, and the supplied two-by-two form. If $Q$ denotes that form, let $\alpha=(1,0)$ and $\beta=(0,1)$. The vector $v=\beta-3\alpha$ is orthogonal to $\alpha$ and has $Q(v,v)=-6$. More generally, if $Q(\alpha,\alpha)>0$ and $Q$ is nonpositive on its orthogonal complement, subtracting the projection of $\beta$ gives

$$Q(\alpha,\beta)^2\ge Q(\alpha,\alpha)Q(\beta,\beta).$$

This is the complete linear-algebra bridge from an intersection signature to a neighboring-coefficient inequality. It explains what needs proving in larger examples: the coefficient interpretation and the sign theorem, not simply the presence of a polynomial.

Review symmetric functions and tableaux, representation theory, matroids, and polyhedral and Ehrhart combinatorics for earlier models. All numbered sections here are optional research extensions. The core investigation can be completed from this foundation and its explicitly supplied data.

44.1 Quasisymmetric and noncommutative symmetric functions

Optional extension; not required to complete the chapter foundation.

A composition $\gamma=(\gamma_1,\ldots,\gamma_k)$ of $n$ is an ordered list of positive integers summing to $n$. In commuting variables $x_1,x_2,\ldots$, define the monomial quasisymmetric function

$$M_\gamma=\sum_{i_1\lt\cdots\lt i_k}x_{i_1}^{\gamma_1}\cdots x_{i_k}^{\gamma_k}.$$

A bounded-degree formal series is quasisymmetric if the coefficient of a monomial with increasing indices depends only on its ordered positive exponents. Symmetric functions forget still more information: they are invariant under all finite permutations of variables. For example, $M_{(1,2)}$ is quasisymmetric but not symmetric, since swapping indices can reverse the exponent order.

Solved example. Multiplying $M_{(1)}$ by itself splits index pairs into equal indices and the two possible orders of distinct indices:

$$M_{(1)}^2=M_{(2)}+2M_{(1,1)}.$$

The coefficient two counts two assignments to distinct ordered indices. Nonnegative coefficients in this particular basis have a combinatorial meaning, but positivity must always specify the basis.

The graded algebra $\mathrm{NSym}$ of noncommutative symmetric functions is the free associative algebra on symbols $H_1,H_2,\ldots$, with $\deg H_i=i$ and $H_0=1$. For a composition $\gamma$, write $H_\gamma=H_{\gamma_1}\cdots H_{\gamma_k}$. In particular, $H_{(1,2)}$ and $H_{(2,1)}$ are different basis elements. Define

$$\Delta H_n=\sum_{i=0}^n H_i\otimes H_{n-i}$$

and extend multiplicatively. This gives a graded Hopf structure, whose graded dual is $\mathrm{QSym}$ under $\langle H_\gamma,M_\delta\rangle=\delta_{\gamma,\delta}$. In the monomial basis, its dual coproduct is deconcatenation:

$$\Delta M_{(a,b)}=1\otimes M_{(a,b)}+M_{(a)}\otimes M_{(b)}+M_{(a,b)}\otimes1.$$

These are established constructions; see Gelfand and collaborators and the Hopf framework of Aguiar, Bergeron, and Sottile. Proof scope: the displayed product and coproduct are checked directly; the complete graded-duality theorem is cited rather than proved.

Exercise. Compute $M_{(1)}M_{(2)}$. Hint: split equal and unequal indices. Solution: the product is $M_{(3)}+M_{(1,2)}+M_{(2,1)}$. This illustrates how ordered combinatorial data survives even though the variables themselves commute.

44.2 Hopf algebras of combinatorial structures

Optional extension; not required to complete the chapter foundation.

A Hopf algebra combines multiplication, a unit, a coproduct $\Delta$, a counit $\varepsilon$, and an antipode $S$. The coproduct separates an object into parts; the antipode is a convolution inverse, often expressed by alternating decompositions. A connected graded bialgebra has a unique antipode, recursively determined in positive degree by

$$S(h)=-h-\sum S(h')h''$$

when $\Delta h=1\otimes h+h\otimes1+\sum h'\otimes h''$ is its reduced decomposition. Established theorem, proof scope: connectedness makes the recursion decrease degree; the general bialgebra-to-Hopf proof is not reproduced.

For a concrete model, take a rational vector space with basis the isomorphism classes of finite simple graphs. Multiply by disjoint union. If $G$ has vertex set $V$, set

$$\Delta G=\sum_{A\subseteq V}G[A]\otimes G[V\setminus A].$$

The empty graph is the unit, and the counit is one on it and zero on positive-degree graphs. Coassociativity follows because either iterated coproduct indexes ordered partitions of $V$ into three parts. Compatibility with multiplication follows by independently splitting the vertices of each disjoint component. Multiplicities remain even when the resulting graphs are isomorphic.

Aguiar, Bergeron, and Sottile establish the general canonical map from a connected graded Hopf algebra equipped with a character to $\mathrm{QSym}$. The graph instance above is fully specified and checked; no claim that every positive Hopf basis has positive antipode coefficients is made.

Exercise. Compute the proper-coloring series of the three-vertex path. Hint: the only independent two-vertex set is the pair of endpoints. Solution: it is $M_{(1,2)}+M_{(2,1)}+6M_{(1,1,1)}$. The first two terms place the endpoints in one color class and the middle vertex in the other; six orders account for three distinct colors.

44.3 Schubert polynomials and Schubert calculus

Optional extension; not required to complete the chapter foundation.

A permutation $w\in S_n$ indexes a Schubert class on the complete flag variety of $\mathbb C^n$, whose points are chains of subspaces of dimensions $1,\ldots,n$. Write $s_i$ for swapping positions $i$ and $i+1$, and let $\ell(w)$ be the inversion count. On polynomials define the divided difference

$$\partial_i f=\frac{f-s_i f}{x_i-x_{i+1}},$$

where $s_i f$ swaps variables. The numerator is divisible by the denominator. Starting from $\mathfrak S_{w_0}=x_1^{n-1}x_2^{n-2}\cdots x_{n-1}$ for the reverse permutation, define $\mathfrak S_{ws_i}=\partial_i\mathfrak S_w$ when $\ell(ws_i)=\ell(w)-1$. The divided-difference relations ensure consistency. This is an established construction originating with Lascoux and Schützenberger; omitted prerequisite: the general braid-relation proof and the identification with flag-variety cohomology.

Solved example. For $S_3$, starting from $x_1^2x_2$ yields:

For instance, $\partial_1(x_1^2x_2)=x_1x_2$ gives $231$, while $\partial_2(x_1^2x_2)=x_1^2$ gives $312$. Applying $\partial_1$ to $x_1^2$ then gives $x_1+x_2$, the polynomial for $132$.

In Grassmannian Schubert calculus, partitions label classes. For $\operatorname{Gr}(2,4)$, the allowed shapes fit a two-by-two rectangle. Multiplication by the divisor class $\sigma_{(1)}$ adds one box, discarding shapes outside the rectangle:

$$\sigma_{(1)}^2=\sigma_{(2)}+\sigma_{(1,1)},\qquad \sigma_{(1)}^3=2\sigma_{(2,1)},\qquad \sigma_{(1)}^4=2\sigma_{(2,2)}.$$

The top class has degree one, so the last product has degree two. The diagram verifies every one-box path: there are exactly two paths from the empty shape to the rectangle. Proof scope: the paths prove the displayed calculation assuming the established Pieri rule; its geometric intersection proof requires transversality and the Schubert-class basis and is not reproduced. A primary treatment of Schubert formulas is Buch and Fulton.

Exercise. Compute $\partial_1(x_1x_2)$ and $\partial_2(x_1x_2)$. Hint: swap the indicated variables before subtracting. Solution: the first is zero because $x_1x_2$ is symmetric in those variables. The second is $x_1$, since $x_1(x_2-x_3)/(x_2-x_3)=x_1$. A divided difference lowers degree when it is nonzero; it does not always produce a nonzero polynomial.

44.4 Cluster algebras and combinatorial models

Optional extension; not required to complete the chapter foundation.

A seed consists of algebraically independent cluster variables and an exchange matrix controlling mutations. Mutation replaces one variable by a rational expression whose numerator is the sum of two monomials. Iterating generates a cluster algebra inside the ambient rational function field. Fomin and Zelevinsky's foundational paper proves the Laurent phenomenon: cluster variables are Laurent polynomials in an initial cluster. This does not follow merely from looking at a rational exchange rule.

Consider the coefficient-free rank-two seed with variables $a,b$ and matrix $\begin{pmatrix}0&1\\ -1&0\end{pmatrix}$. Alternating mutations give the type-$A_2$ recurrence $x_{i-1}x_{i+1}=1+x_i$, starting with $x_1=a,x_2=b$. These variables are formal indeterminates; division takes place in $\mathbb Q(a,b)$.

Solved example. Direct simplification gives

$$x_3=\frac{1+b}{a},\qquad x_4=\frac{1+a+b}{ab},\qquad x_5=\frac{1+a}{b},\qquad x_6=a,\qquad x_7=b.$$

For example, $x_4=(1+x_3)/b$. Then $(1+x_4)/x_3$ simplifies to $(1+a)/b$, since the numerator factors as $(a+1)(b+1)$. All five distinct variables have positive Laurent coefficients in $a,b$. This example proves positivity directly rather than inferring it from Laurentness. Type $A_2$ also has the pentagon model: clusters correspond to the five triangulations of a pentagon, each with two diagonals, and mutation flips one diagonal. It is a small established finite-type model, not a classification proof for all exchange matrices.

Exercise. Specialize $a=b=1$ and list the first seven terms. Hint: retain the exchange recurrence. Solution: $1,1,2,3,2,1,1$. Checking numerical periodicity supports the example, but the rational identities above prove the five-step return for all initial indeterminates. Numerical substitutions alone do not prove a Laurent identity.

44.5 Crystal graphs and tableaux models

Optional extension; not required to complete the chapter foundation.

A crystal graph records the combinatorics of a representation after a quantum deformation is reduced to its crystal limit. Vertices carry weights, and colored operators $e_i,f_i$ move weight by opposite simple roots. The arrows are partially defined and labeled by $i$; they are not arbitrary edges between tableaux of similar shape. The established theory originates in Kashiwara's crystal-basis paper. Omitted prerequisites: quantum groups, crystal lattices, and the theorem identifying general tableau crystals with representations.

A complete elementary model avoids general reading-word conventions. Take one-row semistandard tableaux of length two with entries in $\{1,2,3\}$. These are weakly increasing words $11,12,13,22,23,33$. Their weight is the triple of multiplicities of the three letters. Define $f_i$ by replacing the rightmost occurrence of $i$ with $i+1$ if it exists, and define $e_i$ by replacing the leftmost occurrence of $i+1$ with $i$. Otherwise the operator is undefined. These rules are valid for this single-row model; general shapes require a signature rule, not unrestricted replacement.

Solved example. The full colored arrow list is:

Each $e_i$ reverses its corresponding arrow. The highest vertex is $11$: neither raising operator is defined there. The weight generating function is $x_1^2+x_1x_2+x_1x_3+x_2^2+x_2x_3+x_3^2$, the complete homogeneous symmetric polynomial $h_2(x_1,x_2,x_3)$. Thus counting these vertices by weight gives a character with nonnegative multiplicities. This positivity concerns weight coefficients, not a statement that the character coefficients form a log-concave sequence in every ordering.

Exercise. Apply $f_1$ and $f_2$ to $13$, and compute $e_1(23)$. Hint: use the exact replacement convention. Solution: $f_1(13)=23$, $f_2(13)$ is undefined because there is no $2$, and $e_1(23)=13$. The weight change under the first arrow is $(-1,1,0)$.

44.6 Tropical combinatorics and tropical geometry

Optional extension; not required to complete the chapter foundation.

Use the min-plus convention: $a\oplus b=\min(a,b)$ and $a\odot b=a+b$. A tropical polynomial is a minimum of affine-linear functions. Its tropical hypersurface is the set where the minimum is attained by at least two distinct terms. This is a piecewise-linear corner locus, not the ordinary zero set of the minimum function.

Solved example. The tropical line defined by $\min(0,x,y)$ consists of three rays:

$$\{(0,t):t\ge0\},\qquad\{(t,0):t\ge0\},\qquad\{(-t,-t):t\ge0\}.$$

On the first ray, the constant term and $x$ tie; on the second, the constant and $y$ tie; on the third, $x=y$ is no larger than zero. The outgoing primitive directions are $(0,1),(1,0),(-1,-1)$, each of weight one, and their sum is zero. This verifies balancing at the vertex. At $(1,2)$ the minimum is uniquely the constant term, so that point is not on the tropical line even though the minimum equals zero.

For a matroid $M$ on $E$, its Bergman fan in $\mathbb R^E/\mathbb R\mathbf1$ consists, in the min convention, of weights whose minimum on every circuit occurs at least twice. Adding a constant to every coordinate does not change that condition. For $U_{3,4}$ the only circuit is $E$, so the condition simply says the minimum of the four coordinates is attained at least twice. Thus $(0,0,1,2)$ belongs and $(0,1,2,3)$ does not.

The connection with linear spaces comes from valuations: a non-Archimedean valuation satisfies $\nu(a+b)\ge\min(\nu(a),\nu(b))$, and cancellation in a linear relation forces a minimal valuation to occur at least twice. Speyer and Sturmfels give the primary tropical Grassmannian framework. Proof scope: the ray and circuit examples are verified directly; the general correspondence between valued linear spaces, Plücker relations, and tropical linear spaces requires valued-field algebra and is not proved here. Circuit conditions for an abstract matroid do not by themselves furnish a field representation.

Exercise. Decide whether $(2,2,2,5)$ and $(2,3,4,5)$ lie in the Bergman fan of $U_{3,4}$. Hint: check the unique circuit, then subtract a common constant if desired. Solution: the first has three minimal coordinates and belongs; the second has a unique minimum and does not. Translation by $-2\mathbf1$ leaves both conclusions unchanged.

44.7 Positivity and log-concavity in algebraic combinatorics

Optional extension; not required to complete the chapter foundation.

Three statements must be distinguished. Basis positivity says that coefficients in a named algebraic basis are nonnegative. Log-concavity compares neighboring numbers. Hodge-type positivity says that a signed bilinear form is positive on a particular primitive subspace. These conditions can interact, but none should be substituted for another without a theorem.

Let $M$ be a loopless matroid of rank $r+1$. Write

$$\overline\chi_M(t)=\sum_{i=0}^r(-1)^i\mu_i t^{r-i}.$$

The established Adiprasito–Huh–Katz theorem implies that the nonnegative sequence $\mu_0,\ldots,\mu_r$ is log-concave. It applies to arbitrary loopless matroids, not only field-representable ones. The classical characteristic-coefficient log-concavity conjectures are thus theorems; this chapter does not treat their historical formulation as still open.

Proof mechanism, explicitly a sketch. In the Chow ring, canonical nef classes $\alpha,\beta$ satisfy $\mu_i=\deg(\alpha^{r-i}\beta^i)$. Mixed Hodge-type inequalities imply $\mu_i^2\ge\mu_{i-1}\mu_{i+1}$. Nef means a limit of classes supplied by strictly submodular functions, so the boundary case is obtained by continuity from the appropriate positive cone. Omitted prerequisites: the general coefficient-degree identity and the mixed/limiting intersection inequalities; the ordinary Euclidean Cauchy–Schwarz inequality would have the wrong direction. Section 44.9 verifies the rank-three model directly.

Solved example. For $U_{3,4}$, reduced magnitudes $1,3,3$ satisfy $9\ge3$. Its full characteristic magnitudes are $1,4,6,3$, with checks $16\ge6$ and $36\ge12$. These are two related sequences, not interchangeable labels for the same coefficients. The full sequence is obtained from the reduced one by multiplying by $t-1$ with alternating signs.

A separate established theorem is the strongest Mason inequality for independent-set counts $I_k$ of a matroid on $n$ elements:

$$\left(\frac{I_k}{\binom nk}\right)^2\ge\frac{I_{k-1}}{\binom n{k-1}}\frac{I_{k+1}}{\binom n{k+1}}\quad(1\le k\lt n).$$

Counts beyond the rank are zero. Brändén and Huh's Lorentzian-polynomial paper proves this normalized log-concavity. It is stronger than unnormalized log-concavity and is an established result, not an unresolved conjecture in this treatment.

Exercise. List $I_k$ and the normalized counts for $U_{3,4}$. Hint: all subsets of size at most three are independent. Solution: the counts are $1,4,6,4,0$, and division by $\binom4k$ gives $1,1,1,1,0$. The normalized inequalities hold, with equality at the first two interior indices. They describe independent sets rather than characteristic coefficients.

44.8 Lorentzian and stable polynomials

Optional extension; not required to complete the chapter foundation.

A real polynomial is real stable if it is nonzero whenever every variable has positive imaginary part. This is a multivariate nonvanishing condition, not merely positivity of real coefficients. A homogeneous polynomial with nonnegative coefficients is Lorentzian if its support has the appropriate exchange property and all its quadratic derivatives have Hessians with at most one positive eigenvalue. The following precise criterion is an established characterization from Brändén and Huh.

For degree $d\ge2$, the support consists of exponent vectors with nonzero coefficients. It must be $M$-convex: whenever $u_i\gt v_i$ for two support vectors, there is $j$ with $u_j\lt v_j$ such that both $u-e_i+e_j$ and $v+e_i-e_j$ belong to the support. Here $e_i$ is the coordinate vector. For every sequence of $d-2$ coordinate derivatives, the resulting quadratic has at most one positive Hessian eigenvalue; zero derivatives are allowed. For degree zero or one, all nonnegative-coefficient homogeneous polynomials are admitted. Proof scope: the equivalence with the recursive/closure definition is cited, not proved. In degree two, the nonnegative-coefficient Hessian criterion alone characterizes the class.

Solved example. The basis polynomial of $U_{2,3}$ is

$$f=x_1x_2+x_1x_3+x_2x_3.$$

Its Hessian is $J_3-I_3$, with eigenvalues $2,-1,-1$, so it is Lorentzian. It is also stable, with an independent short proof: when all $z_i$ have positive imaginary part, none is zero, and

$$f(z_1,z_2,z_3)=z_1z_2z_3\left(\frac1{z_1}+\frac1{z_2}+\frac1{z_3}\right).$$

Each reciprocal has strictly negative imaginary part, so their sum cannot vanish. This proves the required nonvanishing over the entire upper-half-plane domain, rather than at a few sample points.

For a matroid $M$, its basis generating polynomial $B_M(x)=\sum_{B\text{ basis}}\prod_{i\in B}x_i$ is Lorentzian. Homogeneous stable polynomials with nonnegative coefficients are Lorentzian, but these classes are not equal in general; one must not infer stability for every matroid basis polynomial from the Lorentzian theorem. The general theorem and the implication are established, with their proofs omitted here.

Exercise. For $q(x,y)=ax^2+2bxy+cy^2$ with $a,b,c\gt0$, translate the quadratic Hessian condition into an inequality. Hint: the trace is positive. Solution: the Hessian is $2\begin{pmatrix}a&b\\ b&c\end{pmatrix}$. It has at most one positive eigenvalue exactly when its determinant is nonpositive, namely $b^2\ge ac$. The normalized coefficient list is $(a,b,c)$, not $(a,2b,c)$; the normalization is part of the log-concavity interpretation.

44.9 Combinatorial Hodge theory

Optional extension; not required to complete the chapter foundation.

Let $M$ be loopless of rank $r+1$. A flat is a set equal to its matroid closure. Define the graded real Chow ring $A^{\ast}(M)$ with degree-one generators $x_F$ for nonempty proper flats. Impose $x_Fx_G=0$ for incomparable flats and the linear relations

$$\sum_{F\ni i}x_F=\sum_{F\ni j}x_F\quad(i,j\in E).$$

The degree map $\deg:A^r(M)\to\mathbb R$ sends the product along every maximal chain of $r$ nonempty proper flats to one. Its existence and compatibility with the relations are established theorems. For a strictly submodular function $c$ on subsets, with $c(\varnothing)=c(E)=0$, put $\ell=\sum_F c(F)x_F$. Strict submodularity requires $c(I)+c(J)\gt c(I\cap J)+c(I\cup J)$ for incomparable subsets.

The established Adiprasito–Huh–Katz Hodge theorem gives hard Lefschetz: multiplication by $\ell^{r-2q}$ is an isomorphism $A^q\to A^{r-q}$ for $0\le q\le r/2$. On the primitive kernel of multiplication by $\ell^{r-2q+1}$, the form $(-1)^q\deg(u\ell^{r-2q}v)$ is positive definite. Omitted prerequisites: the inductive proof, Poincaré duality, and the fan-theoretic construction of the positive cone. In rank three, the assertion means one positive intersection direction and a negative-definite orthogonal complement.

Solved example. For $U_{3,4}$ the proper flats are its four singletons and six pairs. Write $x_i$ for a singleton generator and $x_{ij}$ for a pair generator. The top-degree rules are

$$\deg(x_i x_{ij})=1,\quad\deg(x_{ij}^2)=-1,\quad\deg(x_i^2)=-2.$$

All products of distinct incomparable generators have degree zero. To derive the pair square, multiply the linear relation for an element in $\{i,j\}$ and an element outside it by $x_{ij}$: the surviving terms are $x_i x_{ij}+x_{ij}^2=0$. To derive the singleton square, multiply the relation for $i$ and $j$ by $x_i$: two of the three containing pairs remain after cancellation, giving degree $-2$.

Set $\alpha=\sum_{F\ni1}x_F=x_1+x_{12}+x_{13}+x_{14}$ and $\beta=\sum_{F\not\ni1}x_F=x_2+x_3+x_4+x_{23}+x_{24}+x_{34}$. Expansion now gives

$$\deg(\alpha^2)=-2-3+6=1,\qquad\deg(\alpha\beta)=3,\qquad\deg(\beta^2)=-6-3+12=3.$$

These are exactly the reduced characteristic magnitudes. The restricted pairing is the foundation's matrix $G$, and $\deg((\beta-3\alpha)^2)=-6$. This exact calculation proves the small inequality. It does not prove the full Hodge theorem or claim that the two classes are strictly ample; they lie on the nef boundary.

Exercise. Verify strict submodularity of $c(S)=\lvert S\rvert(4-\lvert S\rvert)$ and compute $\deg(\ell^2)$. Hint: put $a=\lvert I\setminus J\rvert$, $b=\lvert J\setminus I\rvert$. Solution: the submodularity difference is $2ab\gt0$ for incomparable sets. Thus $\ell=3\sum_i x_i+4\sum_{i\lt j}x_{ij}$. The squares contribute $-72-96$, and the twelve incident singleton-pair products contribute $2\cdot12\cdot3\cdot4=288$, giving $\deg(\ell^2)=120\gt0$.

44.10 Connections to matroid representability and intersection theory

Optional extension; not required to complete the chapter foundation.

A matroid is representable over a field $F$ if its ground-set elements can be assigned columns of a matrix over $F$ so that its independent sets are exactly the linearly independent column sets. Representability is an additional condition on an abstract matroid and depends on the field.

Solved example. The matrix

$$A=\begin{pmatrix}1&0&0&1\\ 0&1&0&1\\ 0&0&1&1\end{pmatrix}$$

represents $U_{3,4}$ over every field. The first three columns are a basis. Replacing any one coordinate column by the final column gives determinant $1$ or $-1$, nonzero even in characteristic two. Thus every triple is independent and the full four-element set is dependent. This verifies the represented matroid directly, rather than assuming a genericity condition.

For a complex representation, the associated hyperplane arrangement has a complement that admits a wonderful compactification. Boundary divisors indexed by flats lead to an intersection ring. The combinatorial Chow presentation in section 44.9 recovers this geometric structure in the suitable realization setting; for arbitrary matroids it still defines a ring and satisfies the Hodge theorem without requiring a complex arrangement. Primary sources are De Concini and Procesi's wonderful-model construction and Adiprasito, Huh, and Katz. Proof sketch and omitted prerequisites: the identification requires algebraic varieties, blowups, Chow groups, and the arrangement model; none is proved by merely displaying a matrix. Hodge-type log-concavity does not certify field representability.

The same flats provide another, purely combinatorial encoding. They form a ranked poset: $U_{3,4}$ has rank layers $1,4,6,1$. A maximal chain from empty set to $E$ chooses a singleton and then a containing pair; there are $4\cdot3=12$. Recording rank jumps along chains produces quasisymmetric series, while intersection theory assigns degrees to monomials along maximal flat chains. These use shared incidence data, but their products and relations are different. A coefficient in a chain enumerator is not automatically an intersection number.

Exercise. How many maximal chains pass through the flat $\{1,2\}$, and what is the degree of each corresponding top monomial? Hint: choose its singleton subflat. Solution: two chains, using $\{1\}$ or $\{2\}$. Their monomials $x_1x_{12}$ and $x_2x_{12}$ each have degree one, so their sum has degree two. This agrees with the explicit incidence count, without conflating the two encodings.

Reconstruct a positivity proof in a small algebraic or matroid model

The core investigation reconstructs the coefficient and signature calculation for $U_{3,4}$. Its optional branches examine the full intersection presentation and a quasisymmetric flat-chain enumerator. Together they show how the same finite matroid can support ordered enumeration, representation by vectors, and intersection-theoretic inequalities. Neither positive chain counts nor a representable matrix alone proves the general Hodge theorem.