Some Lower Bounds in the B. and M. Shapiro Conjecture for Flag Varieties
Discrete Comput Geom
Some Lower Bounds in the B. and M. Shapiro Conjecture for Flag Varieties
Monique Azar 0 1
Andrei Gabrielov 0 1
0 A. Gabrielov ( ) Department of Mathematics, Purdue University , West Lafayette, IN 47907 , USA
1 M. Azar Department of Mathematics, American University of Beirut , Beirut , Lebanon
The B. and M. Shapiro conjecture stated that all solutions of the Schubert Calculus problems associated with real points on the rational normal curve should be real. For Grassmannians, it was proved by Mukhin, Tarasov, and Varchenko. For flag varieties, Sottile found a counterexample and suggested that all solutions should be real under certain monotonicity conditions. In this paper, we compute lower bounds on the number of real solutions for some special cases of the B. and M. Shapiro conjecture for flag varieties, when Sottile's monotonicity conditions are not satisfied.
Schubert calculus; Wronski map; Real enumerative geometry; Rational functions

1 Introduction
Schubert Calculus is a recipe for counting geometric objects subject to certain
incidence relations [
17
]. For example:
Problem 1.1 Given 2d − 2 lines in general position in CPd , how many codimension 2
subspaces of CPd meet all 2d − 2 lines?
The answer is ud = d1 2dd−−12 , the d th Catalan number. Schubert Calculus was
based on the semiempirical principle of conservation of number. Its rigorous
foundation, subject of Hilbert’s 15th problem, was established through the development
The second author was supported by the NSF grant DMS0801050.
of intersection theory. From the point of view of intersection theory, enumerative
problems such as Problem 1.1 are solved by counting intersection multiplicities of
Schubert varieties in the Grassmannian (see, e.g., [
1
]).
Over the reals, the principle of conservation of number fails, and solving
enumerative problems becomes considerably more complicated. In [
7
], W. Fulton says “The
question of how many solutions of real equations can be real is still very much open,
particularly for enumerative problems.”
For example, suppose that all 2d − 2 lines in Problem 1.1 are real. How many
codimension 2 subspaces that meet all these lines are real?
Sottile [
23
] proved that all of them can be real for some choice of the 2d − 2 lines.
Boris and Michael Shapiro conjectured that all these subspaces are real whenever
the given lines are tangent to the rational normal curve γ (t ) = [1 : t : t 2 : · · · : t d ] at
distinct real points. Eremenko and Gabrielov [
4
] (see also [
5
]) proved the following
equivalent theorem.
Theorem 1.2 If the critical points of a rational function f are all real, then f is
equivalent to a real rational function, i.e., there exists a fractional linear
transformation l such that l ◦ f is a real rational function.
A more general version of the B. and M. Shapiro conjecture, proved in [
12
],1
claims a similar result for higherdimensional subspaces.
Theorem 1.3 Let P1, P2, . . . , Pk(d−k+1) be (k − 1)dimensional planes in CPd
osculating the rational normal curve at distinct real points. Any (d − k)dimensional
plane that meets all Pi , 1 ≤ i ≤ k(d − k + 1), must be real.
B. and M. Shapiro suggested an extension of their conjecture to flag varieties,
replacing osculating planes by osculating flags. A special case of that conjecture would
imply that all solutions to the following problem are real.
Problem 1.4 Let y1, y2, . . . , y2d−3, r, s be distinct points in R. For 1 ≤ i ≤ 2d − 3,
let Ti be the line tangent to γ at γ (yi ), and let T2d−2 be the line through the points
γ (r) and γ (s). Among all the codimension 2 subspaces of CPd that meet all the lines
T1, T2, . . . , T2d−2, how many are real?
Sottile [
20
] found examples when the B. and M. Shapiro conjecture for flag
varieties fails, in particular, Problem 1.4 does have nonreal solutions. Computer
experiments suggested that the conjecture might hold whenever a certain monotonicity
condition is met [
16
]. For the case of Problem 1.4, Sottile’s monotonicity condition
simply means that the interval I with endpoints r and s contains either all or none
of the points y1, y2, . . . , y2d−3. A proof of this special case was given in [
6
]. For a
complete survey of the B. and M. Shapiro conjecture and its modifications, see [
21
]
or [
22
].
1This result dates back to 2005.
In this paper we study Problem 1.4 when the monotonicity condition does not
hold, i.e., when the interval I contains k of the points y1, y2, . . . , y2d−3 for some k
with 1 ≤ k < 2d − 3. We give an algorithm to compute a (nonstrict) lower bound2 for
the number of real subspaces for any pair of integers d and k with 1 ≤ k < 2d − 3.
We do this by giving a lower bound for the number of solutions to the following
equivalent problem.
Problem 1.5 Let y1, y2, . . . , y2d−3, r, s be distinct points in R and suppose that the
interval I having endpoints r and s contains k of the points y1, y2, . . . , y2d−3 for
some k ∈ {1, 2, . . . , 2d − 4}. How many equivalence classes of real rational functions
f of degree d having critical points at y1, y2, . . . , y2d−3 and satisfying f (r) = f (s)
are there?
Remark 1.6 The answer to Problem 1.5 does not change if we replace k by 2d − 3 − k.
Lower bounds have been computed for several other real enumerative problems.
Eremenko and Gabrielov [
3
] showed that the number of real solutions to certain
systems of real polynomial equations is bounded below by the topological degree of the
corresponding Wronski map and computed this degree. Soprunova and Sottile [
19
]
extended the definition of the real Wronski map to certain sparse polynomial
systems associated with partially ordered sets, computed its degree, and derived the
corresponding lower bounds on the number of real solutions. Degtyarev and
Kharlamov [
2
] showed that there are at least 8 real rational cubics passing through 8
general real points in the plane. Welschinger [
26
] discovered an invariant that gives
a lower bound for the number of real rational curves of degree d passing through
3d − 1 real points. Itenberg, Kharlamov, and Shustin [
10
] showed that this invariant
is at least d!/2 using Mikhalkin’s tropical correspondence theorem [
11
]. Welschinger
discovered other invariants for similar enumerative problems (see [
25
] and the
references there), some of which have been computed by Pandharipande, Solomon, and
Walcher [
13
]. Solomon [
18
] showed how the Welschinger invariant can be interpreted
as the degree of a map. An excellent review of these results can be found in [
21, 22
].
While the previous results are based on topological degree invariants, our method
uses a different line of reasoning that does not seem to be associated with a
topological degree. The lower bound for the number of real solutions of Problem 1.5 is
obtained by studying a oneparameter family of real rational functions with critical
points y1, . . . , y2d−3, θ and the dependence on θ of the increments on [r, s] of the
functions in that family.
2 The Wronski Map and OneParameter Families of Rational Functions
In this section, we represent the points of the Grassmannian G of twodimensional
planes in the space of polynomials of degree d by pencils (p, q) of linearly
independent polynomials, define the Wronski map W : G −→ CP2d−2, and prove that W
2The lower bound is strict for d = 4, 5, and 6.
is not ramified over the space Q of polynomials with all real roots of multiplicity
at most 2. We study the properties of a oneparameter family ft = pt /qt of rational
functions obtained by lifting to G a path t → wθ(t) in Q. Here θ (t ) is a path in RP1,
and wθ is a polynomial with roots y1, . . . , y2d−3, θ . Our main tool is a net of a real
rational function with real critical points (see [
4, 5
]).
Let G be the Grassmannian of twodimensional planes in the space of complex
polynomials of degree at most d . An element of G can be defined by a pencil
L = L(p, q) = {αp + βq}, where p and q are two linearly independent
polynomials of degree at most d , and [α : β] ∈ CP1. If L ∈ GR, the Grassmannian of real
twodimensional planes, we always assume that p and q are real and call L a real
pencil. The Wronskian of (p, q), given by W (p, q) = p q − q p, is a nonzero
polynomial of degree at most 2d − 2. If the degree of W (p, q) is 2d − 2 − k for some
k > 0, we say that W (p, q) has a root at infinity of multiplicity k. If (p1, q1) is
another basis for L, then W (p1, q1) differs from W (p, q) by a nonzero
multiplicative constant. For a nonzero polynomial r = ai zi of degree at most 2d − 2, let
h(r) = [a0 : a1 : . . . : a2d−2].
Definition 2.1 The Wronski map W : G −→
h(W (p, q)), where (p, q) is a basis for L.
CP2d−2 is defined by W(L) =
The map W is well defined and finite. Its degree is u(d), the d th Catalan number.
This was originally computed by Schubert in 1886. A proof can be found in [
8
] or [
9
].
Definition 2.2 Let Q = h(Q0), where Q0 is the set of all polynomials r satisfying
(i) 2d − 4 ≤ deg(r) ≤ 2d − 2,
(ii) all roots of r belong to RP1, and
(iii) all roots of r have multiplicity at most 2.
In Theorem 2.13, we shall show that W is unramified over Q and hence,
WW−1(Q) is a covering map. Thus, for any path ρ in Q with initial point ρ0, given
L0 ∈ G with W(L0) = ρ0, ρ can be lifted in a unique way to a path in G starting
at L0.
Definition 2.3 Let D = {y1, y2, . . . , y2d−3} be a set of points in R. For y ∈ R, let
ωy = (x − y1)(x − y2) . . . (x − y2d−3)(x − y), and let ω∞ = (x − y1)(x − y2) . . . (x −
y2d−3).
The map RP1 −→ CP2d−2 given by y −→ h(wy ) is continuous. In fact, it extends
to an algebraic map CP1 −→ CP2d−2. Let I be an interval in R, and let t0 ∈ I.
A path θ : I −→ RP1 induces a path ρ : I −→ Q given by ρ(t ) = h(ωθ(t)). Given
L0 ∈ G with W(L0) = h(ωθ(t0)), ρ can be lifted to a unique path χ in G satisfying
χ (t0) = L0. It was shown in [
4
] that W−1(RP2d−2) ⊂ GR. This implies that χ (I) is
contained in GR.
Lemma 2.4 Let Rd be the space of real rational functions of degree at most d all of
whose critical points are real. If θ is analytic, all the lifts defined above are analytic,
and polynomials pt and qt with coefficients analytic in t can be selected to form the
bases of χ (t ), t ∈ I. Let ft be the real rational function pt /qt viewed as a
holomorphic function from CP1 to CP1. The path η : I −→ Rd given by t −→ ft has the
following properties.
(i) The Wronskian W (pt , qt ) is equal to ρ(t ).
(ii) If x ∈ CP1 \ D, then the map t −φ→x ft (x) is continuous.
φx
(iii) For each x ∈ D, the map t −→ ft (x) is continuous at all values t for which pt
and qt do not have a common root at x.
Proof The first property follows from the construction of ft . To prove (ii) and (iii),
first observe that if x ∈ CP1 \ D, then, for any t ∈ I, pt (x) and qt (x) cannot both be
zero. Let (t0, x) ∈ I × CP1 be such that at least one of pt0 (x) and qt0 (x) is not equal
to 0. This implies that there exists a neighborhood N of t0 such that either ft (x) or
1/ft (x) is a continuous complexvalued function on N and hence, ft (x) : N −→ CP1
is continuous.
For pairs (t, x) such that pt (x) = qt (x) = 0, we shall see in the following lemma
that for small > 0, ft+ is monotone on (x, x + ), and the image of (x, x + )
covers RP1 except for an interval of length O( ).
Lemma 2.5 Suppose that y1 = 0 and that {(p , q ) : ≥ 0} is a family of ordered
pairs of linearly independent real polynomials of degree at most d satisfying:
(i) the coefficients of p and q are analytic in ,
(ii) the roots of w = W (p , q ) are y1 = 0, y2, . . . , y2d−3, ,
(iii) p0 and q0 have a common root at x = 0, and
(iv) f0(0) = 0.
Then for small > 0, f (0) = O( ), f ( ) = O( ), and f has a pole in (0, ).
Moreover, f : [0, ] −→ RP1 is onetoone, and its image is RP1 \ (α, β), where
(α, β) is an open interval of size O( ) with endpoints at f (0) and f ( ).
Proof By our assumption, p0 and q0 have a common root at x = 0, and f0(0) = 0,
so there exist k1 = 0 and k2 = 0 such that p0 = k1x2 + O(x3) and q0 = k2x + O(x2).
The requirement that f has a fixed critical point at x = 0 for all > 0 implies
that p (x) = k1(x2 + a x + ab 2) + O(x3, x2, 2x, 3) and q (x) = k2(x + b ) +
O(x2, x, 2). Therefore, w (x) = k1k2(x2 + 2b x) + O(x3, x2, 2x, 3). In
particular, w ( ) = (1 + 2b)k1k2 2 + O( 3).
Since for all > 0, f has a critical point at x = , we should have b = − 21 .
Hence, q has a root at x = 21 + O( 2) between the critical points 0 and of f ,
and p (x ) = 0. Since p (0) = O( 2) and p ( ) = O( 2), we have f (0) = O( )
and f ( ) = O( ).
For small enough > 0, the interval (0, ) does not contain any points of D. If
f < 0 on (0, ), then when x moves from 0 to , the value of f (x) decreases from
O( ), tends to −∞ as x approaches x , and returns from ∞ to O( ) at x = . If
f > 0 on (0, ), then when x moves from 0 to , the value of f (x) increases from
O( ), tends to ∞ as x approaches x , and returns from −∞ to O( ) at x = .
Suppose t −→ gt = pt∗/qt∗ is another path satisfying L(pt∗, qt∗) = χ (t ) =
L(pt , qt ). There exists a real fractional linear transformation ht = (at x + bt )/(ct x +
dt ) such that gt = ht ◦ ft . In particular, for each t , ft and gt have the same critical
points. Observe that the continuity of pt , qt , pt∗, qt∗ implies that the sign of at dt − bt ct
does not change as t varies in I. Postcomposition with a real fractional linear
transformation defines an equivalence relation on Rd . This proves the following.
Theorem 2.6 With the above notation, let ft0 ∈ Rd be a function whose Wronskian
is equal to ρ(t0). Let Rd0 be the family of equivalence classes of functions in Rd of
degree exactly d , and let [ft0 ] ∈ Rd0 be the equivalence class of ft0 . There exists a
unique map η : I −→ Rd0 with η(t0) = [ft0 ] that projects to ρ.
The set Rd0 is in onetoone correspondence with the set of nondegenerate real
pencils. A pencil L(p, q) ∈ G is nondegenerate if deg(p/q) = d , in other words, if
max(deg p, deg q) = d and p and q have no common zeros. Otherwise, L is said to
be degenerate.
Notation 2.7 We shall identify RP1 with the unit circle S1 via a fractional linear
transformation ϑ : CP1 −→ CP1 mapping RP1 to S1, 0 to 1, and preserving
orientation. A real rational function f will be identified with the function f = ϑ ◦ f ◦ ϑ −1.
The critical points of f are the images of the critical points of f under ϑ . In
particular, all critical points of f belong to the unit circle. Moreover, for any two points a
and b in RP1, f (a) = f (b) if and only if f (ϑ (a)) = f (ϑ (b)).
Let Sd = {f = ϑ ◦ f ◦ ϑ −1 : f ∈ Rd }. We shall write f instead of f and specify
whether we are considering f as a function in Rd or Sd . With this identification, any
construction or proof given for Rd also applies to Sd and vice versa.
Definition 2.8 Let L = L(p, q) be a real pencil whose Wronskian has 2d − 2 real
roots counted with multiplicity, and let f = p/q. The set Γ = f −1(RP1) is called
the net of f [
4
]. Alternatively, let f = ϑ ◦ f ◦ ϑ −1. Then Γ = f −1(S1).
Observe that Γ is independent of the choice of basis (p, q), so we can also refer
to Γ as the net of L. Unless explicitly stated otherwise, we shall consider Γ to be
defined as f −1(S1) rather than f −1(RP1). Clearly, S1 ⊂ Γ , and Γ is invariant under
the reflection s(z) = 1/z with respect to the unit circle.
The net Γ defines a cell decomposition of CP1 as follows:
(1) the vertices (0cells) of the cell decomposition correspond to the zeros of
W (p, q),
(2) the edges (1cells) are the components of Γ \ V , where V is the set of vertices,
(3) the faces (2cells) are the components of C \ Γ .
This cell decomposition has the following properties. The closure of each cell is
homeomorphic to a closed ball of the corresponding dimension. If W (p, q) has
distinct real roots, then each vertex of Γ has degree 4. If W (p, q) has multiple roots, we
call Γ a degenerate net.
Remark 2.9 Let L(p, q) be a real pencil whose net Γ is degenerate and whose
Wronskian has only simple or double roots, all real. A vertex of Γ of degree 2 corresponds
to a point x such that p(x) = q(x) = 0, while a vertex of degree 6 corresponds to a
double root x of W (p, q) such that at least one of p(x) and q(x) is not 0.
Definition 2.10 The edges of Γ that lie inside the unit disc are called chords or
interior edges of Γ . Since Γ is symmetric with respect to S1, it suffices to consider
its chords. Given a vertex v, the pair γ = (Γ , v) is called the net of f with respect to
v, or simply the net of f if v is clear. The vertex v is called the distinguished vertex
of γ .
The standard orientation of S1 induces a cyclic order ≺ on the vertices of Γ . We
shall label these vertices v1, v2, . . . , v2d−2 so that v1 ≺ v2 ≺ · · · ≺ v2d−2 ≺ v1, and we
shall set v2d−1 = v1. For a net γ = (Γ , v1), we define a linear order on the vertices
by v1 < v2 < · · · < v2d−2.
Two nets (Γ , v1) and (Γ , v1) are said to be equivalent if there exists a
homeomorphism CP1 −→ CP1 mapping Γ to Γ , v1 to v1, preserving orientation of both CP1
and S1, and commuting with the reflection s. By abuse of notation, we shall often
refer to an equivalence class of nets by one of its representatives.
Definition 2.11 For a net γ = (Γ , v), let Shift(γ ) = (Γ , v ) where v is the
predecessor of v under ≺.
Remark 2.12 Without loss of generality, we may assume that the vertices of Γ are
equally spaced on the circle. The net Shift(γ ) is equivalent to the net (Γ , v) where
the ordered vertex set of Γ coincides with that of Γ , and the chords of Γ are
obtained by rotating the chords of Γ by π/(d − 1) counterclockwise.
Theorem 2.13 The Wronski map W : G −→ CP2d−2 is unramified over the space
Q of all polynomials with real roots of multiplicity at most 2.
Proof Let π0 ∈ Q, L(p0, q0) ∈ W−1(π0), and f0 = p0/q0. The function f0 can be
chosen real [
4
]. From the definition of Q, the common roots r1, . . . , rm of p0 and
q0 (if any) must be simple. Suppose that W is ramified at L(p0, q0), with
ramification index ν. Let πj be a sequence of polynomials with distinct real roots converging
to π0 as j → ∞. Since the polynomials with distinct real roots form an open set
in the space of all real polynomials, we can assume that W is not ramified over πj
for all j ≥ 1. Then W−1(πj ) contains ν distinct points L(pj,k, qj,k) of the
Grassmannian converging to L(p0, q0) as j → ∞. Theorem 1.2 implies that all pencils
L(pj,k, qj,k) are real. With the proper normalization, we can assume that pj,k and
qj,k are real polynomials and that (pj,k, qj,k) → (p0, q0) as j → ∞. Then the
rational functions fj,k = pj,k/qj,k are real, with all real critical points, converge to f0
as j → ∞ uniformly on every compact set not containing the points r1, . . . , rm, and
their nets converge to the (degenerate) net of f0. Since the roots of π0 are at most
double, the net of f0 may have vertices of degree either 6 or 2. In both cases, the net
of fj,k is uniquely determined by the net of f0 for large enough j , independent of k.
It follows from [
4
] that, for large enough j , pencils L(pj,k, qj,k) do not depend on k.
Hence, ν = 1, and W is not ramified over π0.
3 Properties of the Nets of Functions ft
In this section, we study dependence on the parameter t of the net γt of the rational
function ft defined in Sect. 2. In particular, we show that the net γt is shifted
counterclockwise when θ (t ) passes its first vertex y1. We describe also degenerate nets
corresponding to polynomials wθ with double roots.
Definition 3.1 A 2 × (d − 1) Young tableau is the distribution of the integers
1, 2, . . . , 2d − 2 on a 2 × (d − 1) rectangular array such that every integer is greater
than the one above it and the one to its left, if any.
Equivalence classes of nets with 2d − 2 vertices can be identified with 2 × (d − 1)
Young tableaux: to a net γ = (Γ , v1) with the ordered vertex set {v1, v2, . . . , v2d−2},
we associate the tableau with an integer i belonging to the first row if and only if
vi is connected by a chord to a vertex vj with j > i. In particular, the number of
equivalence classes of nets having 2d − 2 vertices is the d th Catalan number ud =
1 2d−2 (see [
24
]).
d d−1
Example 3.2 For d = 4, there are u4 = 5 equivalence classes of nets with 6 vertices.
These equivalence classes (with the distinguished vertex v1 = 1) and their respective
Young tableaux are shown in Fig. 1.
Let D = {y1, y2, . . . , y2d−3} be an ordered set of distinct points in S1 all different
from 1. Let θ : [0, 2d − 2] −→ S1 be the path given by θ (t ) = e2πit . By Lemma 2.4
and Notation 2.7, there exists a family {ft ∈ Sd : t ∈ [0, 2d − 2]} with the following
properties:
(i) for each t , the critical points of ft counted with multiplicity are y1, y2, . . . , y2d−3,
and θ (t );
(ii) given (t0, z0) ∈ [0, 2d − 2] × CP1, ft (z0) as a function of t is continuous at
t0 whenever deg ft0 = d . In particular, for any z0 ∈/ D, ft (z0) is a continuous
function of t .
For t ∈ [0, 2d − 2], let γt = (Γt , y1) be the net of ft with respect to y1. Let
θ −1(D) = {t1, t2, . . . , tκ−1} with 0 < t1 < t2 < · · · < tκ−1 < 2d − 2, and let t0 = 0
and tκ = 2d − 2. Here κ − 1 = (2d − 2)(2d − 3) since θ makes 2d − 2 turns around
S1 as t goes from 0 to 2d − 2. The sequence {s1, s2, . . . , s2d−2} = θ −1(y1) is a
subsequence of {ti }. Let s0 = 0 and s2d−1 = 2d − 2.
Lemma 3.3 Let l ∈ {0, 1, . . . , 2d − 2}. For t ∈ (sl , sl+1) \ θ −1(D), the nets γt are
equivalent.
Proof Let a and b be two points in (sl , sl+1) \ θ −1(D) with a < b. If θ ([a, b]) ∩ D =
∅, then for each z ∈ CP1, ft (z) is a continuous function of t on [a, b], and, for all
t ∈ [a, b], the nets γt are nondegenerate. Therefore, vi vj is a chord in γa if and only
if it is also a chord in γb, and hence the two nets are equivalent.
If θ ([a, b]) ∩ D = {yi }, i = 1, then θ (a) = vi and yi = vi+1 in the ordered vertex
set of γa , while yi = vi and θ (a) = vi+1 in that of γb. For each z ∈ CP1 \ {yi }, ft (z)
is a continuous function of t on [a, b], and, for all t ∈ [a, b] \ θ −1(yi ), the nets γt are
nondegenerate. Thus, for all m, n ∈/ {i, i + 1}, vmvn is a chord in γa if and only if it is
also a chord in γb. We have to consider the following two cases.
Case 1: vi vi+1 is a chord in γa . In this case, every other chord vmvn in γa satisfies
m, n ∈/ {i, i + 1} and hence must also be a chord in γb. Therefore, vi vi+1 must be a
chord in γb. An example of this case is given in Fig. 2a–c.
Case 2: vi vi+1 is not a chord in γa . Let yj be the endpoint of the chord from vi ,
and yk the endpoint of that from vi+1. The continuity of ft (z) on [a, b] for any z = yi
implies that yj yk is not a chord in γb. Determining Γb is thus reduced to finding a
net with the vertex set {yj ≺ vi ≺ vi+1 ≺ yk} such that yj and yk are not endpoints
of the same chord. There are only two nets having four vertices, and exactly one of
them does not have a chord joining yj to yk . For an example, see Fig. 2c–e.
Remark 3.4 A similar argument can be used to determine the degenerate net Γti . If,
for t close to ti , there is a chord of Γt connecting θ (t ) and θ (ti ) (see, for example,
Fig. 2a, c), then the vertex θ (ti ) is of degree 2 in Γti (Fig. 2b). Otherwise, θ (ti ) is of
degree 6 in Γti (Fig. 2d).
Lemma 3.5 Let [γ ]l = ([Γ ]l , y1) denote the class {γt : t ∈ (sl , sl+1) \ θ −1(D)}. Then
[γ ]l+1 = Shift([γ ]l ).
Proof Choose representatives γt and γt of [γ ]l and [γ ]l+1, respectively, with
y2d−2 ≺ θ (t ) ≺ y1 ≺ θ (t ) ≺ y2. As t goes from t to t , θ (t ) crosses the
distinguished vertex y1, so Lemma 3.3 cannot be applied to γt and γt . Instead we
shall consider γ = (Γt , y2) and γ = (Γt , y2) which are related to γt and γt
by γt = Shift(γ ) and γt = Shift2(γ ). Since θ (t ) does not cross y2 as t goes
from t to t , we can apply Lemma 3.3 to γ and γ to get that γ = γ . Thus,
γt = Shift2(γ ) = Shift2(γ ) = Shift(γt ).
Example 3.6 The nets a, c, e, g in Fig. 2 all belong to [γ ]0, while the net k belongs
to [γ ]1 and is the shift of [γ ]0.
An anonymous referee pointed out the correspondence between the (inverse of)
the Shift operator and the Schutzenberger promotion p. If γ = Shift(γ ) and Y and
Y are the Young tableaux corresponding to γ and γ , respectively, then p(Y ) = Y .
This correspondence is described in [
14
] where it is attributed to White. Furthermore,
this correspondence, as well as Lemma 3.3, follow from Remark 3.7 in [
15
].
Corollary 3.7 If t ∈ (t0, t1) and t ∈ (tκ−1, tκ ), then γt = γt . Equivalently, [γ ]0 =
Proof As t goes from t to t , θ (t ) makes 2d − 2 turns around the circle. By Lemma
3.5, each complete turn of θ corresponds to the shift operator (i.e., rotation by
π/(d − 1)) applied to the net. Applying the shift operator 2d − 2 times results in
a complete rotation of the net, resulting in the original net.
Remark 3.8 It could happen that [γ ]0 = [γ ]l for some positive integer l < 2d − 2.
This occurs when [γ ]0 has rotational symmetry of order n for some integer n > 1, in
which case l = (2d − 2)/n, a factor of 2d − 2. For example, the net [γ ]0 in Fig. 2a
has rotational symmetry of order 3, so [γ ]0 = [γ ]2.
Example 3.9 Figure 2 describes how a given net is modified as θ (t ) makes a full turn
around the circle. As θ (t ) moves in (y1, y2) (Fig. 2a), the net remains unchanged.
When θ (t ) reaches y2 (Fig. 2b), the net becomes degenerate with deg(y2) = 2. As
θ (t ) moves away from y2 (Fig. 2c), we recover the original net until θ (t ) reaches y3
(Fig. 2d). At this point the net becomes degenerate again with deg(y3) = 6. As θ (t )
moves away from y3 (Fig. 2e), we recover the original net, and the process continues
in a similar manner until θ (t ) crosses y1. The net obtained at this step (Fig. 2k) is the
shift of the net in Fig. 2a.
4 Lower Bounds for k = 1
In this section, we derive lower bounds on the number of real solutions to Problem
1.5 in the special case where the interval (r, s), or its complement, contains only one
fixed critical point of f . To simplify notation, we shall assume that y1 = ∞ is the
only fixed vertex not contained in the interval (r, s). Throughout this section, f is a
real rational function of degree d, and its net is given by Γ = f −1(RP1).
Theorem 4.1 Let r < y2 < y3 < · · · < y2d−3 < s be points in R, and let y1 = ∞.
Then there are at least ud − 2ud−1 classes of real rational functions f of degree d
having critical points at yj , 1 ≤ j ≤ 2d − 3, and satisfying f (r) = f (s).
Proof Let C be an equivalence class of the nets with 2d − 2 vertices and no interior
edges connecting the distinguished vertex y1 to any of its two neighboring vertices.
The number of such classes is ud − 2ud−1, since there are ud−1 classes of nets having
two given neighboring vertices connected by an edge. It remains to show that for each
such class C, there exist y ∈ R and a real rational function fy having the net γy ∈ C
with the vertex set {y1, y2, . . . , y2d−3, y}, satisfying fy (r) = fy (s).
Let y ∈ (s, ∞). There exists a unique class Fy of real rational functions with
the net belonging to C and critical points y1, y2, . . . , y2d−3, y [
4
]. Choose fy ∈ Fy
so that fy has a double pole at ∞ and fy (x) < 0 for large x. Let ρ : R −→ Q
be given by ρ(y) = (x − y2)(x − y3) . . . (x − y2d−3)(x − y). Note that ρ(y) is the
Wronskian of fy . By Lemma 2.4, ρ can be lifted to a path η : R −→ Rd with η(y) =
fy satisfying properties (i)–(iii) of Lemma 2.4. By Lemma 3.3, γy ∈ C for all y ∈
The map φx : R −→ RP1 given by φx (y) = fy (x) is continuous for all x in R −
{yj }j2d=−23. The continuity of φx (y) on (s, ∞) implies that for all y ∈ (s, ∞), fy (x) < 0
for large x. The following lemma completes the proof.
Lemma 4.2 There exists y ∈ R such that fy (r) = fy (s).
Proof Assume that fy (r) = fy (s) for all y ∈ [r, s]. If y ∈ [r, s], then fy cannot have a
pole at r nor at s since each of [s, ∞) and (−∞, r] belongs to the boundary of a face
of Γy and fy has a pole at ∞. Since both φr and φs are continuous, this implies that
either −∞ < fy (r) < fy (s) < ∞ for all y ∈ [r, s] or −∞ < fy (s) < fy (r) < ∞ for
all y ∈ [r, s]. Without loss of generality, we may assume that −∞ < fy (r) < fy (s) <
∞ for all y ∈ [r, s]. In particular, −∞ < fs (r) < fs (s) < ∞.
For y ∈ (s, ∞), fy (y) is finite since fy has a pole at ∞ and (y, ∞) belongs to
the boundary of a face of Γy . Moreover, fy (s) cannot exceed fy (y) without first
decreasing to −∞ since φs is continuous and fy has a local maximum at y and no
critical points between s and y. On the other hand, since Γy has no interior edges
connecting ∞ to y for all y ∈ [s, ∞), limy→∞ fy (y) = −∞, and it follows that
fy (s) must also decrease to −∞ on a subinterval of (s, ∞). In particular, there exists
y ∈ (s, ∞) such that fy (r) = fy (s).
5 Upper and Lower Bounds for the Arc Length of f ([r, s])
In this section, we assume that f maps S1 to S1. We derive lower and upper bounds
for the arc length of f ([r, s]), in terms of the net of f and the position of the points r
and s relative to the vertices of the net.
Consider S1 with the standard orientation. Given any two points r and s in S1, we
define [r, s] and (r, s), respectively, to be the closed and open positively oriented arcs
in S1 starting at r and ending at s. The standard orientation on S1 induces an order
on [r, s] given by a ≤ b if and only if a ∈ [r, b].
Definition 5.1 Let Γ be a net such that all its vertices y1 < y2 < · · · < yn inside (r, s)
are simple. The Young tableau of Γ corresponding to (r, s), denoted by YΓ (r, s), has
2 rows and n entries defined as follows. The integer i is placed in the first row of
YΓ (r, s) if and only if yj is not connected by a chord of Γ to yi , i > j .
Note that YΓ (r, s) is part of the Young tableau of (Γ , y1). See examples in Fig. 3
and Fig. 4.
Definition 5.2 Given a rational function f and an oriented segment or a closed loop
c in CP1 with f (c) ⊂ S1, let L(c) = Lf (c) be the argument increment of f on c.
Let Γ be a net with simple vertices y1 < · · · < yn inside (r, s). Let f be a rational
function with the net Γ . Assume that f is orientation preserving on (r, y1). Let Y =
YΓ (r, s).
Definition 5.3 Let E and O be the numbers of even and odd entries in the second
row of Y . Let m be the number of vertices of Γ inside (r, s) connected by a chord to
vertices outside (r, s).
Lemma 5.4 The number m = n − 2(O + E) is the difference between the length of
the first row of Y and the length of its second row.
Consider first the case m = 0.
Lemma 5.5 The following are equivalent:
(i) m = 0;
(ii) Y is rectangular;
(iii) no vertex in (r, s) is connected by a chord of Γ to a vertex outside (r, s).
Definition 5.6 Assuming that m = 0, let V = {y1, . . . , yn} be the set of vertices of
Γ inside (r, s). Let C be the set of chords of Γ having both endpoints in V , and let
F = F (r, s) be the set of faces F of Γ satisfying ∂F ∩ S1 ⊂ [r, s]. A chord in C
will be denoted by yi yj where yi and yj are its endpoints, with i < j . Given a chord
yi yj ∈ C, let Fj be the face in F having yj as its largest vertex. Let F0 be the face of
Γ whose boundary contains [r, y1].
There is a onetoone correspondence between the sets C and F . Condition (iii)
of Lemma 5.5 implies that ∂F0 must also contain [yn, s].
Example 5.7 For the net Γ and arc (r, s) in Fig. 3, n = 8, E = 2, O = 2, m = 0,
C = {y1y2, y3y8, y4y5, y6y7}, and F = {F2, F5, F7, F8}.
Definition 5.8 Assuming that m = 0, a chord yi yj in C is even (resp., odd) if j is
even (resp., odd).
By the construction of YΓ (r, s) we have the following.
Lemma 5.9 The numbers E and O in Definition 5.3 are, respectively, the numbers
of even and odd chords in C.
Definition 5.10 A face F of Γ is positive if f is orientation preserving on ∂F ,
otherwise F is negative. Let P and N be the numbers of positive and negative faces
in F . The face F0 is always positive since f is orientation preserving on (r, y1).
Since f maps the boundary of each face F ∈ F bijectively onto S1, it follows that
L(∂F ) = 2π if F is positive and L(∂F ) = −2π if F is negative.
Lemma 5.11 Assuming that m = 0, a chord yi yj in C is odd if and only if Fj is
positive. In particular, O = P and E = N .
Proof For yi yj ∈ C, the arc [yj−1, yj ] of S1 belongs to ∂Fj , and its orientation
induced by Fj agrees with the standard orientation on S1. Thus, since f preserves
orientation on [r, y1] and has only simple critical points in (r, s), it preserves orientation
on [yj−1, yj ] if and only if j is odd.
Lemma 5.12 Let m = 0. Then L([r, s]) ∈ (2π(O − E), 2π(O − E + 1)).
Proof Let Ω be the region bounded by the arc [y1, yn] ⊂ S1 and α, the positively
oriented curve on ∂F0 from y1 to yn (see Fig. 3). Since Ω \ Γ is the union of the
faces in F ,
L(∂Ω) =
L(∂F ) = 2π(P − N ) = 2π(O − E).
On the other hand, L([r, y1]) + L(α) + L([yn, s]) ∈ (0, 2π ) since [r, y1] ∪ α ∪ [yn, s]
is part of the boundary of the positive face F0 having positive orientation. Therefore,
F ∈F
L([r, s]) = L [r, y1] + L [y1, yn] + L [yn, s]
= L [r, y1] + L(α) + L(∂Ω) + L [yn, s]
∈ 2π(O − E), 2π(O − E + 1) .
Remark 5.13 Assume that m = 0. If f is orientation reversing on (r, y1), then g =
1/f is orientation preserving on (r, y1), and Lg([r, s]) = −Lf ([r, s]). Replacing f
by g, we obtain from Lemma 5.12 that Lf ([r, s]) ∈ (2π(E − O − 1), 2π(E − O)).
For the arc (y0, yj1 ), f is orientation preserving on (y0, y1), so by Lemma
5.12, L([y0, yj1 ]) ∈ (2π(O0 − E0), 2π(O0 − E0 + 1)). For (yj1 , yj2 ), f is
orientation reversing on (yj1 , yj1+1). Since j1 is odd, the vertices that are even with
respect to (yj1 , yj2 ) belong to O1, while the vertices that are odd with respect to
(yj1 , yj2 ) belong to E1. Accordingly, by Remark 5.13, Lf ([yj1 , yj2 ]) ∈ (2π(O1 −
E1 − 1), 2π(O1 − E1)). Applying the same arguments to (yji , yji+1 ), we get the
following.
Lemma 5.15 Let i ∈ {0, 1, . . . , m}. If i is even, then L([ji , ji+1]) ∈ (2π(Oi −
Ei ), 2π(Oi − Ei + 1)). If i is odd, then L([ji , ji+1]) ∈ (2π(Oi − Ei − 1), 2π(Oi −
Ei )).
Summing up over i = 0, . . . , m, we obtain the following.
Theorem 5.16 Let I = IΓ (r, s) =
Then L([r, s]) ∈ I .
mod 2 − n + 4O − 1 π, (n + 1)
mod 2 + n − 4E + 1 π .
Now we consider the case where W (p, q) has a double root y and the net Γ is
degenerate. If y ∈/ (r, s), this does not affect L([r, s]). If y ∈ (r, s), then in labeling
the vertices y1, . . . , yn in (r, s), we assign to y two consecutive indices and define
YΓ (r, s) to be YΓ (r, s) where Γ is a nondegenerate net with the vertices and chords
close to those of Γ . With this agreement, all the arguments above can be applied to
Γ , and the statement of Theorem 5.16 remains true except for the case where the
degree of y is 6 and both chords of Γ with the ends at y have other ends outside
(r, s). This means that y represents a segment (yji , yji+1 ) of length zero.
Theorem 5.17 Let y be a double root of W (p, q) inside (r, s) which is a vertex of
Γ of degree 6 such that both chords of Γ with the ends at y have other ends outside
(r, s). Let I = IΓ (r, s) be
mod 2 − n + 4O − 1 π, (n + 1)
mod 2 + n − 4E − 1 π
if the number of vertices between r and y is odd, and
mod 2 − n + 4O + 1 π, (n + 1)
mod 2 + n − 4E + 1 π
otherwise. Then L([r, s]) ∈ I .
We can similarly compute IΓ (s, r) and use it to improve IΓ (r, s) in some cases.
Let f be a function in Sd all of whose critical points are simple except for possibly
one double critical point. Let the critical points of f be y1, . . . , y2d−2 (counted with
multiplicity) with y1 ≺ r ≺ y2 y3 · · · yk+1 ≺ s ≺ yk+2 · · · y2d−2 y1.
Assume that f is orientation preserving on (y1, y2). Let (Γ , y1) be the net of f with
respect to y1, and let Y be the corresponding Young tableau. Let F be the set of all
faces of Γ . Let E and O be the numbers of even and odd entries in the second row
of Y , respectively. Then
L [r, s] + L [s, r] =
∂F = 2π(E − O − 1).
F ∈F
So if IΓ (r, s) = (a, b), IΓ (s, r) = (a , b ) and L([r, s]) + L([s, r]) = c, then IΓ (r, s)
can be improved to (a, b) ∩ (c − b , c − a ).
6 Lower Bounds in the General Case
In this section, we apply the results of Sect. 5 to derive lower bounds on the number
of real solutions to Problem 1.5. Without loss of generality, we may set r = −1 and
s = 1. Let D = {y1, y2, . . . , y2d−3} be a set of distinct points in S1 all different from
±1. Assume that y1 does not belong to the arc [r, s]. Let f0 be a function in Sd
having critical set D ∪ {1}. Let θ : R −→ S1 be the path given by θ (t ) = e2πit . By
Lemma 2.4 and Notation 2.7, there exists a path η : R −→ Sd given by η(t ) = ft ,
where ft = pt /qt is a rational function having critical points y1, y2, . . . , y2d−3, θ (t )
counted with multiplicity, satisfying properties (i)–(iii) of Lemma 2.4.
Lft∗ ([r, s]) − limt→t∗ Lft ([r, s]) = 2π .
Lemma 6.1 The map R −→ R given by t → Lft ([r, s]) is continuous everywhere
except on a subset D of θ −1(D) given by D = {t ∈ θ −1(D) : pt (x) = qt (x) =
0 for some x ∈ D}. At any point of discontinuity t ∗, limt→t∗ Lft ([r, s]) exists, and
Proof The first statement follows from Lemma 2.4. The second statement follows
from Lemmas 2.4 and 2.5 and from the fact that ft∗ (x) is continuous at θ (t ∗) as a
function of x.
Let L : R −→ R be given by
L(t ) =
Lft ([r, s]) if t ∈/ D,
limx→t Lfx ([r, s]) if t ∈ D.
Corollary 6.2 The map L is continuous on R. For any t ∈ R, L(t ) ∈ 2π Z if and only
if ft (r) = ft (s).
For t ∈ R, let γt = (Γt , y1) be the net corresponding to ft . The net Γt is
nondegenerate if and only if θ (t ) ∈/ D. If t ∈ D, then θ (t ) is a vertex of degree 2 in Γt , and
if t ∈ θ −1(D) \ D, then θ (t ) is of degree 6 in Γt . This follows from Lemma 2.4 and
Remark 2.9.
Lemma 6.3 The map t → γt is periodic. The period T is a factor of 2d − 2. It is
equal to the number of distinct nondegenerate nets in the image.
Proof This follows from Corollaries 3.3 and 3.7 and Remark 3.4.
1 ) and Wn = (n + 21 , n + 1) so that θ (Vn) = (s, r)
For n ∈ Z, let Vn = (n, n + 2
and θ (Wn) = (r, s). For t ∈ R, let 2π l1(t ) and 2π u1(t ) be the lower and upper
endpoints of the interval IΓt (r, s) defined in Theorems 5.16 and 5.17, and let 2π l2(t ) and
2π u2(t ) be the lower and upper endpoints of the interval IΓt (s, r) defined similarly.
Let 2π c = L([r, s]) + L([s, r]). Then L(t )/(2π ) ∈ (l1(t ), u1(t )) ∩ (c − u2(t ), c −
l2(t )).
Lemma 6.4 The functions l1(t ) and u1(t ) are constant on each set Vn.
Proof The interval IΓt (r, s) depends only on the Young tableau of Γt corresponding
to (r, s) which is independent of any changes outside (r, s).
Lemma 6.5 The functions l1(t ) and u1(t ) are constant on each set Wn \ (θ −1(D) \ D)
and assume only finitely many values on each set Wn.
Proof The arc (r, s) does not contain the distinguished vertex y1, so by Corollary
3.3, the nets γt , t ∈ Wn \ θ −1(D) are equivalent. Since θ (t ) ∈ (r, s) \ D for all t ∈
Wn \ θ −1(D), Definition 5.1 implies that the Young tableaux YΓt (r, s) are identical
for all t ∈ Wn \ θ −1(D). For t ∈ Wn ∩ D, the Young tableaux YΓt (r, s) are identical to
those for t ∈ Wn \ θ −1(D), as defined after Theorem 5.16. This, along with the fact
that Wn ∩ (θ −1(D) \ D) is finite, proves the second part of the statement.
If we let γt = (Γt , y) for some fixed vertex y ∈ (r, s), we get the following.
Corollary 6.6 The functions l2(t ) and u2(t ) are constant on each set Wn and assume
only finitely many values on each set Vn.
Let E = E(f0) be the set of equivalence classes of functions ft , t ∈ R. The map
t −→ [ft ] is periodic with period T . The value of T depends on the choice of f0, but
it is always a factor of 2d − 2. Therefore, instead of calculating T for each f0, we
shall consider the interval [0, 2d − 2] and deal with the issue of having counted some
equivalence classes more than once at the end of this section.
Let U and L be two integervalued functions defined by
1
U n + 2
U (n) = inf u1(t ), c − l2(t ) : t ∈ Vn ,
= inf u1(t ), c − l2(t ) : t ∈ Wn ,
L(n) = sup l1(t ), c − u2(t ) : t ∈ Vn ,
= sup l1(t ), c − u2(t ) : t ∈ Wn .
The functions U and L can be easily computed since l1, u1, l2, u2 assume only
finitely many values on each Vn and Wn. In addition, the continuity of L(t )
implies the existence of tn ∈ Vn and tn+1/2 ∈ Wn such that 21π L(tn) ∈ (L(n), U (n)) and
21π L(tn+1/2) ∈ (L(n + 1/2), U (n + 1/2)). Let S = {k/2 : 0 ≤ k < 4d − 4}. Let ≺ be
the cyclic order on S given by 0 ≺ 1/2 ≺ 1 ≺ · · · ≺ (4d − 5)/2 ≺ 0. For i, j ∈ S, let
(i, j ) = {k ∈ Si ≺ k ≺ j }.
Definition 6.7 A point i ∈ S is called a max point if
(i) ∃k ∈ S such that L(i) ≥ U (k) and L(i) > L(j ), ∀j ∈ (k, i), and
(ii) ∃k ∈ S such that L(i) ≥ U (k) and L(i) ≥ L(j ), ∀j ∈ (i, k).
A point i ∈ S is called a min point if
(i) ∃k ∈ S such that U (i) ≤ L(k) and U (i) < U (j ), ∀j ∈ (k, i), and
(ii) ∃k ∈ S such that U (i) ≤ L(k) and U (i) ≤ U (j ), ∀j ∈ (i, k).
Lemma 6.8 Between any two max (resp. min) points of S, there is a min (resp. max)
point.
Proof Let m and m be two max points in S. Assume that L(m) ≥ L(m ). There
exists k ∈ (m, m ) such that L(m) ≥ L(m ) ≥ U (k). Choose k to be the first value
where U attains a minimum on (m, m ) (the order here is the order induced by the
cyclic order). Then k is a min point.
Let m1 < m2 < · · · < ml be the max and min points of S, and let ml+1 = m1.
Lemma 6.9 Let i ∈ {1, 2, . . . , l}. If mi is a max point, then U (mi+1) ≤ L(mi ). If mi
is a min point, then U (mi ) ≤ L(mi+1).
Proof Let mi be a max point. Assume that U (mi+1) > L(mi ). Since mi+1 is a min
point, ∃k ∈ (mi , mi+1) such that U (mi+1) ≤ L(k). So L(mi ) < L(k) < U (k). Since
mi is a max point, ∃k ∈ (mi , k) such that U (k ) ≤ L(mi ) < L(k). Choose k to be the
first point where U attains a minimum on (mi , k). Then k is a min point between mi
and mi+1, which is impossible. The proof of the second statement is similar.
To simplify notation, fix i ∈ {1, 2, . . . , l} and let j = mi and k = mi+1. If j is a
max point, then L(tk) < 2π U (k) ≤ 2π L(j ) < L(tj ), so the interval (L(tk), L(tj ))
contains L(j ) − U (k) + 1 multiples of 2π . The continuity of L implies that each of
these multiples of 2π is attained by L at some value t ∈ (tj , tk). Similarly, if j is a min
point, then each of the L(k) − U (j ) + 1 multiples of 2π in the interval (L(tj ), L(tk))
is attained by L at some value t ∈ (tj , tk). Summing up over i, we get a value V (f0)
which is a lower bound for the number of times L crosses a multiple of 2π over the
interval [0, 2d − 2]. At each point t where this happens, the corresponding function
ft satisfies ft (r) = ft (s).
Algorithm 6.10 The algorithm described above to compute V (f0) depends on the
net of f0 and not on f0 itself, so we may think of it as accepting a net γ and label its
output V (γ ).
Example 6.11 Figure 5 shows some steps of Algorithm 6.10 applied to a net γ with
6 vertices and c = 0. The only min point of γ corresponds to W2 where U = 0, and
the only max point corresponds to W5 where L = 0. So V (γ ) = 2.
Let f01, f02, . . . , f0ud be nonequivalent functions in Sd having critical points at
1 and at y1, y2, . . . , y2d−3. For each i ∈ {1, 2, . . . , ud }, there exists a family fti ,
t ∈ [0, 2d − 2], satisfying properties (i)–(iii) of Lemma 2.4. By Theorem 2.6 and
Notation 2.7, the map t −→ [fti ] is unique. Let γti = (Γti , y1) be the net
corresponding to fti , and let Ti be the period of the map t −→ γti . The uniqueness of the map
t −→ [fti ] implies that the map t −→ γti is unique and that Ti is well defined. Let
Ci = {γji : j = 0, 1, . . . , Ti − 1}. The sets Ci are well defined and partition the set
C of nets with 2d − 2 vertices. This follows from the fact that the map t −→ γti is
unique and its period is a factor of 2d − 2. Let k be the number of distinct sets Ci .
For i ∈ {1, 2, . . . , k}, let NCi be the number of equivalence classes of functions f in
Sd whose net belongs to Ci and which satisfy f (r) = f (s). For γ ∈ C, let V (γ ) be
the output of Algorithm 6.10 when the input net is γ .
Theorem 6.12 The number N of classes [f ] ∈ Sd having critical points at the points
y1, y2, . . . , y2d−3 and satisfying f (r) = f (s) is bounded below by γ ∈C 2Vd(−γ 2) .
Proof Since Ci , i = 1, 2, . . . , k, partition C, N =
. . . , k},
k
i=1 NCi . For each i ∈ {1, 2,
Dividing both sides by Ti and taking the sum over all classes Ci , we get
k
i=1
NCi =
k Ti −1 V (γji )
i=1 j=0
2d − 2 =
γ ∈C
V (γ )
2d − 2
.
The lower bounds computed by the algorithm for 4 ≤ d ≤ 14 and the
corresponding values of k appear in Table 1. The first row and column give the values of d and
k, respectively. Computations in www.math.tamu.edu/~sottile/research/pages/Flags/
Data/F125/We2Xe5.5.html, www.math.tamu.edu/~sottile/research/pages/Flags/Data/
F126/Ve2We7.14.html, and
www.math.tamu.edu/~secant/monotone/monotoneindivproblemview.php?DB=monotoneSecant&result_id=957 show that the lower bounds
for d = 4, 5, and 6 are sharp.
Since t −→ γt is periodic, V (γji ) = NCi for all j ∈ {0, 1, . . . , Ti − 1}, hence,
m
o
S
5
.
g
i
12
7 Combinatorial Interpretation for k = 1, 2
In this section, we give a combinatorial interpretation of the algorithm in the cases
where (r, s) contains exactly one or exactly two points of D. The first case agrees
with the result obtained in Sect. 4. For t ∈ Wn, (l1(t ), u1(t )) ⊆ (c − u2(t ), c − l2(t )).
For t ∈ Vn, (l1(t ), u1(t )) ⊆ (c − u2(t ), c − l2(t )) except when θ (t ) coincides with a
fixed vertex y and the two edges with endpoint y have their other endpoint outside
(s, r ). When k = 1 and t ∈ Vn, there is only one fixed vertex outside (s, r ). So for
k = 1, it is enough to consider (l1, u1).
Case 1: (r, s) contains one point.
Suppose that there is only one fixed vertex y in (r, s). The Young tableaux
corresponding to a net with one fixed vertex in (r, s) can have one or two squares. The
only one having one square gives rise to a unique (L, U )interval, namely (−1, 1).
Each of the two tableaux with two squares gives rise to two intervals. These intervals
are (−1, 0), (0, 1), (−2, 0), and (0, 2). All of these 5 intervals appear in Fig. 6.
The three sequences (0, 1)(−1, 1)(−1, 0), (0, 1)(−1, 1)(−2, 0), (0, 2)(−1, 1)(−1,
0), and their reversals cannot occur in the sequence of (L, U )intervals of any net with
d > 2. So (−1, 0) and (0, 1) do not yield any max or min points. The (L, U )interval
(0, 2) gives a max point if and only if it is preceded by (−2, 0)(−1, 1). In this case
the net corresponding to (−1, 1) has the property that the moving vertex is outside
(r, s) and the fixed vertex in (r, s) is not connected to any of its neighboring vertices.
The same holds when (−2, 0) is a min point. Also, if the net has the property that the
moving vertex is outside (r, s) and the fixed vertex in (r, s) is not connected to any
of its neighboring vertices, then its (L, U )interval must be preceded by (0, 2) and
followed by (−2, 0) or vice versa.
So given a net Γ with 2d − 2 vertices, V (Γ ) is precisely the number of vertices y
of Γ not connected to any of their neighboring vertices. The lower bound for a given
d and k = 1 is the number of distinct nets (Γ , y) such that Γ has 2d − 2 vertices
and y is not connected to any of its neighboring vertices. This agrees with the result
obtained in Sect. 4.
Example 7.1 Figure 6 shows some steps of Algorithm 6.10 applied to a net γ with
6 vertices and k = 1. The degenerate nets with degenerate vertex outside (r, s) are
not shown. Below each net is its Young tableau corresponding to (r, s). The last row
gives the (L, U )interval for each Vi and Wi , i = 0, 1, . . . , 5. The only min point of
γ corresponds to W2 where U = 0, and the only max point corresponds to W5 where
L = 0. So V (γ ) = 2, which is the number of vertices of Γ not connected to any of
the neighboring vertices.
Case 2: (r, s) contains two points.
Suppose that there are two fixed vertices y1 and y2 in (r, s). All possible (L, U
)intervals appear in Fig. 5. Any max point must correspond to an (L, U )interval with
L = 0. More specifically, any max point must correspond to an (L, U )interval equal
to (0, 2) since the interval (0, 1) must always be preceded by (0, 2). Similarly, a min
point must correspond to an (L, U )interval equal to (−2, 0).
Any interval giving rise to a max (resp. min) point must be immediately followed
by the interval (0, 1) (resp. (−1, 0)). The interval (0, 1) (resp. (−1, 0)) is obtained
when the moving critical point is outside (r, s), the two critical points in (r, s) are
connected by a chord, and the function is orientation reversing (resp. preserving) in a
neighborhood of r .
So the min/max points correspond to chords in the net connecting adjacent critical
points and separated by an odd number of critical points. The lower bound is N2d−2,
the number of nets γ = (Γ , y1) with 2d − 2 vertices satisfying y1y2 is a chord of Γ
and the next chord connecting two consecutive vertices is of the form y2k y2k+1 with
1 < k < d − 1.
Counting the number of such nets for a fixed k is equivalent to finding the number
Nk of nets γ = (Γ , v1) having 2d − 6 vertices v1, v2, . . . , v2d−6 such that if vi vi+1
is a chord of Γ , then i ≥ 2k − 3. Let S be the set of all nondegenerate nets γ having
2d − 6 vertices. Given a net γ , let Cγ be the set of chords of γ .
N2 = ud−2 since, when k = 2, there is no condition on the chords of γ .
N3 is the number of nets γ such that v1v2 and v2v3 are not chords in γ . So N3 =
ud−2 − 2ud−3.
N4 is the number of nets γ such that vi vi+1 is not a chord in γ for i = 1, 2, 3, 4:
4
i=1
N4 = S −
{γ ∈ S : vi vi+1 ∈ Cγ }
+
γ ∈ S : {v1v2, v3v4} or {v1v2, v4v5} or {v2v3, v4v5} ⊂ Cγ
= ud−2 − 4ud−3 + 3ud−4.
Similar computations yield N5 = ud−2 − 6ud−3 + 10ud−4 − 4ud−5, N6 = ud−2 −
8ud−3 + 21ud−4 − 20ud−5 + 5ud−6, etc. Using these values, we can compute N2d−2.
The values of N2d−2 for some small values of d are
N6 = N2 = u2 = 1,
N8 = N2 + N3 = 2u3 − 2u2 = 2,
N10 = N2 + N3 + N4 = 3u4 − 6u3 + 3u2 = 6,
N12 = N2 + N3 + N4 + N5 = 4u5 − 12u4 + 13u3 − 4u2 = 18,
N14 = N2 + N3 + N4 + N5 + N6 = 5u6 − 20u5 + 34u4 − 24u3 + 5u2 = 57.
1. Browder , F . (ed.): Mathematical developments arising from Hilbert problems . In: Proceedings of Symposia in Pure Mathematics , vol. XXVIII. Am. Math. Soc. , Providence ( 1976 )
2. Degtyarev , A.I. , Kharlamov , V.M.: Topological properties of real algebraic varieties: Rokhlin's way . Uspekhi Mat. Nauk 55 ( 4 ( 334 )), 129  212 ( 2000 )
3. Eremenko , A. , Gabrielov , A. : Degrees of real Wronski maps . Discrete Comput. Geom . 28 , 331  347 ( 2002 )
4. Eremenko , A. , Gabrielov , A. : Rational functions with real critical points and the B . and M. Shapiro conjecture in real enumerative geometry . Ann. of Math. (2) 155 ( 1 ), 105  129 ( 2002 )
5. Eremenko , A. , Gabrielov , A. : Elementary proof of the B . and M. Shapiro conjecture for rational functions ( 2005 )
6. Eremenko , A. , Gabrielov , A. , Shapiro , M. , Vainshtein , A. : Rational functions and real Schubert calculus . Proc. Amer. Math. Soc . 134 ( 4 ), 949  957 ( 2006 )
7. Fulton , W.: Introduction to Intersection Theory in Algebraic Geometry . CBMS Regional Conference Series in Mathematics, vol. 54 . Published for the Conference Board of the Mathematical Sciences , Washington ( 1984 )
8. Goldberg , L. : Catalan numbers and branched coverings by the Riemann sphere . Adv. Math . 85 ( 2 ), 129  144 ( 1991 )
9. Hodge , W. , Pedoe , D. : Methods of Algebraic Geometry, vol . II. Cambridge University Press, Cambridge ( 1952 )
10. Itenberg , I. , Kharlamov , V. , Shustin , E.: Welschinger invariant and enumeration of real rational curves . Int. Math. Res. Not . 49 , 2639  2653 ( 2003 )
11. Mikhalkin , G.: Counting curves via lattice paths in polygons . C. R. Math. Acad. Sci. Paris 336 ( 8 ), 629  634 ( 2003 )
12. Mukhin , E. , Tarasov , V. , Varchenko , A. : The B . and M. Shapiro conjecture in real algebraic geometry and the Bethe ansatz . Ann. Math. 170 ( 2 ), 863  881 ( 2009 )
13. Pandharipande , R. , Solomon , J. , Walcher , J.: Disk enumeration on the quintic 3fold . J. Amer. Math. Soc . 21 ( 4 ), 1169  1209 ( 2008 )
14. Petersen , T.K. , Pylyavskyy , P. , Rhoades , B. : Promotion and cyclic sieving via webs . J. Algebraic Combin . 30 ( 1 ), 19  41 ( 2009 )
15. Purbhoo , K. : Jeu de taquin and a monodromy problem for Wronskians of polynomials . Adv. Math. 224 ( 3 ), 827  862 ( 2010 )
16. Ruffo , J. , Sivan , Y. , Soprunova , E. , Sottile , F. : Experimentation and conjectures in the real Schubert calculus for flag manifolds . Experiment . Math. 15 ( 2 ), 199  221 ( 2006 )
17. Schubert , H.: Kalkül der abzählenden Geometrie . Springer, Berlin ( 1979 ). Reprint of the 1879 original
18. Solomon , J.: Intersection theory on the moduli space of holomorphic curves with lagrangian boundary conditions ( 2006 ). arXiv:math/0606429v1
19. Soprunova , E. , Sottile , F. : Lower bounds for real solutions to sparse polynomial systems . Adv. Math. 204 ( 1 ), 116  151 ( 2006 )
20. Sottile , F. : Real Schubert calculus: polynomial systems and a conjecture of Shapiro and Shapiro . Experiment. Math. 9 ( 2 ), 161  182 ( 2000 )
21. Sottile , F. : Frontiers of reality in Schubert calculus . Bull. Am. Math. Soc . 47 ( 1 ), 31  71 ( 2010 )
22. Sottile , F. : Real solutions to equations from geometry ( 2010 ). arXiv:math/0609829v2
23. Sottile , F. : Enumerative geometry for the real Grassmannian of lines in projective space . Duke Math. J . 87 ( 1 ), 59  85 ( 1997 )
24. Stanley , R.: Enumerative Combinatorics , vol. 2 . Cambridge University Press, Cambridge ( 1999 )
25. Welschinger , J.Y.: Invariants entiers en géométrie énumérative réelle ( 2010 ). arXiv: 1003 .2707v1
26. Welschinger , J.Y.: Invariants of real rational symplectic 4manifolds and lower bounds in real enumerative geometry . C. R. Math. Acad. Sci. Paris 336 ( 4 ), 341  344 ( 2003 )