06.06.09 · riemann-surfaces / jacobians

Weierstrass points and gap sequences

shipped3 tiersLean: none

Anchor (Master): Griffiths-Harris 1978 *Principles of Algebraic Geometry* (Wiley) §2.2; Arbarello-Cornalba-Griffiths-Harris 1985 *Geometry of Algebraic Curves I* (Springer GMW 267) Ch. I; Hensel-Landsberg 1902 *Theorie der algebraischen Funktionen einer Variabeln* (Teubner) §§ on the Lückensatz

Intuition Beginner

Fix a compact Riemann surface — a closed surface dressed with enough data to ask whether a function on it counts as holomorphic — of genus $g \geq 2$ , and pin down one point $p$ on it. Ask a simple question: what are the allowed "strengths" of a meromorphic function that blows up only at $p$ ? The strength is the order of the pole. On a surface with handles, not every pole order is achievable, and the pattern of which ones are forbidden is a fingerprint of the point.

The forbidden orders are the gaps. For a generic point there are exactly $g$ of them, and they are the smallest possible: $1, 2, \dots, g$ . So at a typical point you can never make a function with a single simple pole, nor a double pole, and so on up to order $g$ , but every larger order is reachable.

A Weierstrass point is a special point where this pattern shifts: some gap is pushed higher than it would generically be, opening room for a function of unexpectedly low pole order. These special points are rare and intrinsic to the surface — you cannot move them by changing coordinates. They are the surface's distinguished marked points, and they can be located all at once by a single determinant called the Wronskian.

The count is rigid. Weighting each special point by how far its gap pattern departs from generic, the weights always sum to the same number, $(g - 1) g (g + 1)$ , no matter which genus- $g$ surface you started with. The geometry of a curve forces a fixed budget of special behaviour, distributed among finitely many points.

Visual Beginner

Picture a vertical number line of allowed pole orders $0, 1, 2, 3, \dots$ drawn beside the point $p$ . Mark each order in one of two colours: black for a gap (no function with exactly that pole order at $p$ ) and green for a non-gap (some function achieves it). At a generic point of a genus- $3$ surface the black marks sit at $1, 2, 3$ and everything from $4$ upward is green — a short black band at the bottom, then green forever. A second diagram, for a hyperelliptic point of the same surface, shows black at $1, 3, 5$ and green at $0, 2, 4, 6, \dots$ — the gaps have spread out, skipping into higher orders.

Beside these, a sketch of the whole surface shows a handful of dots, the Weierstrass points, scattered across it, each tagged with a small number, its weight; a caption notes that the tags always add up to the same total for a fixed genus.

Worked example Beginner

Take a hyperelliptic surface of genus $2$ : it is a double cover of the sphere, branched over $2 g + 2 = 6$ points. Pick one branch point $p$ , a place where the two sheets of the cover pinch together. Functions pulled back from the sphere see $p$ with even order only, because going once around $p$ swaps the two sheets. The lowest-order function with a pole concentrated near $p$ is the coordinate that has a double pole there, giving a non-gap at order $2$ .

So the achievable pole orders at $p$ are $0, 2, 4, 6, \dots$ , all the even numbers. The gaps — the forbidden orders — are $1, 3$ , the odd numbers below $2 g = 4$ . There are exactly $g = 2$ of them, as the gap theorem promises, but they are $1, 3$ rather than the generic $1, 2$ . Order $3$ has been pushed up: this is a Weierstrass point.

How special is it? Compare the gap list $1, 3$ against the generic list $1, 2$ position by position. The first gaps agree ( $1$ versus $1$ , no contribution); the second gap is $3$ versus $2$ , a surplus of $1$ . The total surplus, the weight, is $1$ . With $6$ branch points each of weight $1$ , the weights sum to $6$ .

Check the budget. The total-weight formula says the weights across all Weierstrass points add to $(g - 1) g (g + 1)$ . For $g = 2$ this is $1 \cdot 2 \cdot 3 = 6$ . The six branch points, weight $1$ apiece, account for the whole budget — and there are no other Weierstrass points on a genus- $2$ surface.

Check your understanding Beginner

Formal definition Intermediate+

Let $C$ be a compact Riemann surface (equivalently, a smooth projective complex curve) of genus $g \geq 1$ , with canonical bundle $K_{C}$ , so $de g K_{C} = 2 g - 2$ and $h^{0} (C, K_{C}) = g$ . Fix a point $p \in C$ and write $n p$ for the divisor with multiplicity $n$ at $p$ .

Gaps and non-gaps. An integer $n \geq 1$ is a gap at $p$ if there is no meromorphic function on $C$ with a pole of order exactly $n$ at $p$ and no other poles; equivalently, by Riemann-Roch (cf. 06.04.01), $$ h^0(C, \mathcal{O}(np)) = h^0(C, \mathcal{O}((n-1)p)). $$ Otherwise $n$ is a non-gap, and $h^{0} (C, O (n p)) = h^{0} (C, O ((n - 1) p)) + 1$ . Since $h^{0} (O (n p))$ increases by at most $1$ at each step (a pole of order $n$ is detected by a single Laurent coefficient), each $n \geq 1$ is exactly one of the two.

Weierstrass gap theorem (Lückensatz). For $g \geq 1$ there are exactly $g$ gaps, $$ 1 = n_1 < n_2 < \cdots < n_g \leq 2g - 1, $$ the first always $1$ and the last at most $2 g - 1$ . The non-negative integers that are not gaps form a sub-monoid $H (p) \subseteq Z_{\geq 0}$ under addition — the Weierstrass semigroup — containing $0$ and closed because the product of two functions adds pole orders. A numerical semigroup with $g$ gaps is said to have genus $g$ .

Weierstrass point and weight. The weight of $p$ is $$ w(p) := \sum_{i=1}^g (n_i - i) \geq 0, $$ the total amount by which the gap sequence exceeds the generic sequence $1, 2, \dots, g$ term by term. A point $p$ is a Weierstrass point when its gap sequence is non-generic, equivalently $w (p) > 0$ ; an ordinary Weierstrass point has weight exactly $1$ (gap sequence $1, 2, \dots, g - 1, g + 1$ ).

Wronskian. Choose a basis $ω_{1}, \dots, ω_{g}$ of $H^{0} (C, K_{C})$ and a local coordinate $z$ centred at $p$ , writing $ω_{j} = f_{j} (z) d z$ . The Wronskian is the local function $$ W(z) := \det\bigl( f_j^{(k)}(z) \bigr)_{0 \leq k \leq g-1,; 1 \leq j \leq g}. $$ Under a coordinate change $W$ transforms as a section of $K_{C}^{\otimes g (g + 1) /2}$ , so its zero divisor is well defined and global; $ord_{p} W = w (p)$ , identifying the Weierstrass points with the support of $div W$ .

Counterexamples to common slips.

A gap is not the same as a pole order some function avoids; it is an order no function attains while remaining holomorphic away from $p$ . The dichotomy is forced by Riemann-Roch, not by a particular function.
The bound $n_{g} \leq 2 g - 1$ is sharp only at special points: the generic $n_{g}$ is $g$ , far below $2 g - 1$ . The maximal $n_{g} = 2 g - 1$ occurs at hyperelliptic Weierstrass points.
The Weierstrass semigroup determines the gap sequence and conversely; the two are complementary subsets of $Z_{\geq 0}$ , with the non-gaps a semigroup but the gap set not closed under addition (it is the complement of a semigroup).
Weight $w (p) > 0$ is the definition of a Weierstrass point; the count of Weierstrass points and the sum of their weights are different numbers — a hyperelliptic curve has only $2 g + 2$ Weierstrass points but their weights sum to $g^{3} - g$ .

Key theorem with proof Intermediate+

Theorem (total Weierstrass weight). Let $C$ be a compact Riemann surface of genus $g \geq 1$ . Then the Weierstrass points are the zeros of the Wronskian section $W \in H^{0} (C, K_{C}^{\otimes g (g + 1) /2})$ , each $p$ occurring with multiplicity $w (p)$ , and $$ \sum_{p \in C} w(p) = (g-1),g,(g+1) = g^3 - g. $$ In particular for $g \geq 2$ there are at least $2 g + 2$ and at most $g^{3} - g$ Weierstrass points.

Proof. The argument has three steps: the Wronskian is a global section of the stated bundle; its order of vanishing at $p$ equals $w (p)$ ; the degree of that bundle gives the sum.

Step 1 — globality of the Wronskian. Let $ω_{1}, \dots, ω_{g}$ be a basis of $H^{0} (C, K_{C})$ and, in a coordinate chart with coordinate $z$ , write $ω_{j} = f_{j} d z$ . Under a change of coordinate $w = w (z)$ the column vector $(f_{j})$ scales by $(d z / d w)$ and the rows of derivatives mix by the chain rule; the determinant of derivatives up to order $g - 1$ acquires the factor $(d z / d w)^{1 + 2 + \dots + g}$ together with the differential factor, so $W (d z)^{\otimes g (g + 1) /2}$ is coordinate-free. Changing the basis ${ω_{j}}$ multiplies $W$ by the determinant of the change-of-basis matrix, a non-zero constant, which moves no zeros. Hence $W$ defines a global holomorphic section of $K_{C}^{\otimes g (g + 1) /2}$ , independent of choices up to scale.

Step 2 — vanishing order equals the weight. At $p$ , the vanishing sequence of the canonical series $∣ K_{C} ∣$ is the increasing list $0 \leq a_{1} < a_{2} < \dots < a_{g}$ of orders to which sections of $K_{C}$ vanish at $p$ ; a Riemann-Roch computation identifies these with the gaps shifted by one, $a_{i} = n_{i} - 1$ . Choosing the basis adapted to the flag of vanishing orders, the Wronskian's leading term at $p$ has order $\sum_{i = 1}^{g} (a_{i} - (i - 1)) = \sum_{i = 1}^{g} (n_{i} - i) = w (p)$ . So $ord_{p} W = w (p)$ , positive exactly at Weierstrass points.

Step 3 — degree count. The total number of zeros of a section equals the degree of its bundle: $$ \sum_{p} w(p) = \deg!\bigl(K_C^{\otimes g(g+1)/2}\bigr) = \frac{g(g+1)}{2},(2g - 2) = (g-1)g(g+1). $$ For the count: each Weierstrass point has weight at least $1$ , giving the upper bound $g^{3} - g$ ; each has weight at most $(2 g)$ (the maximum permitted by $n_{g} \leq 2 g - 1$ ), and the extremal case $w (p) = (2 g)$ at every Weierstrass point forces exactly $(g^{3} - g) / (2 g) = 2 g + 2$ of them, the hyperelliptic configuration, which is therefore the minimum. $□$

Bridge. The total-weight formula builds toward the entire theory of special divisors, and the gap/non-gap dichotomy that drives it appears again in the Brill-Noether stratification of 06.06.06, where the jump $h^{0} (n p) - h^{0} ((n - 1) p)$ is the local shadow of the global loci $W_{d}^{r}$ . The foundational reason a finite, rigid budget of special points exists is that the Wronskian lives in a fixed line bundle $K_{C}^{\otimes g (g + 1) /2}$ whose degree is a topological invariant; the bridge is that this is exactly the ramification divisor of the canonical map. This generalises: replacing $K_{C}$ by the bundle of any linear system $g_{d}^{r}$ gives a Wronskian whose zeros are the ramification points of that series, and putting these together with the Plücker formula recovers the total-ramification count for arbitrary linear systems. The central insight is that "special point" is not an extrinsic accident but the vanishing locus of a canonically defined section, which is why automorphisms must permute Weierstrass points and why the count is genus-determined rather than curve-dependent.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Show that the gap sequence at a hyperelliptic Weierstrass point (a branch point of the $g_{2}^{1}$ ) is $1, 3, 5, \dots, 2 g - 1$ , and compute its weight.

Hint

At a branch point $p$ the hyperelliptic involution fixes $p$ ; the non-gaps are generated by the pole order $2$ of the degree- $2$ function, so the non-gaps are the even numbers.

Answer

Let $x : C \to P^{1}$ be the degree- $2$ map and $p$ a branch point. The function $x$ (after translating so $x (p) = \infty$ ) has a double pole at $p$ and is regular near the other branch points; its powers give functions with pole orders $2, 4, 6, \dots$ at $p$ . These are all the non-gaps up to $2 g$ : by Riemann-Roch, $h^{0} (2 k p) = k + 1$ for $0 \leq k \leq g$ , so each even order $\leq 2 g$ is a non-gap and contributes the jumps, while $h^{0}$ does not jump at odd orders $\leq 2 g - 1$ . Hence the gaps are the odd numbers $1, 3, 5, \dots, 2 g - 1$ , exactly $g$ of them. The weight is $$ w(p) = \sum_{i=1}^g \bigl((2i - 1) - i\bigr) = \sum_{i=1}^g (i - 1) = \frac{g(g-1)}{2} = \binom{g}{2}. $$ With $2 g + 2$ branch points the weights sum to $(2 g + 2) (2 g) = (g + 1) g (g - 1) = g^{3} - g$ , saturating the total. Rubric: full credit for the even-semigroup argument, the odd gap sequence, and $w (p) = (2 g)$ .

Exercise 4 (medium, short-answer).

Explain why every automorphism of $C$ permutes the set of Weierstrass points, and deduce a finiteness constraint on $Aut (C)$ for $g \geq 2$ .

Hint

The gap sequence at $p$ is defined purely from $h^{0} (O (n p))$ , which an automorphism preserves; the Weierstrass set is finite and non-empty for $g \geq 2$ .

Answer

An automorphism $φ : C \to C$ pulls back $O (n p)$ to $O (n φ^{- 1} (p))$ and induces an isomorphism on global sections, so $h^{0} (O (n p)) = h^{0} (O (n φ (p)))$ for all $n$ ; hence $p$ and $φ (p)$ have the same gap sequence and the same weight. Therefore $φ$ maps Weierstrass points to Weierstrass points, permuting the finite set $W = {p : w (p) > 0}$ (non-empty and finite for $g \geq 2$ , with $2 g + 2 \leq ∣ W ∣ \leq g^{3} - g$ ). This makes $Aut (C)$ act faithfully on a finite set once one knows the action on $W$ together with local data is faithful, which is the structural input to Hurwitz's theorem that $∣ Aut (C) ∣ \leq 84 (g - 1)$ for $g \geq 2$ (Hurwitz 1893). The Weierstrass set is thus a canonical finite rigidifier of the curve. Rubric: full credit for the $h^{0}$ -invariance, the permutation conclusion, and the connection to the automorphism finiteness/Hurwitz bound.

Exercise 5 (hard, short-answer).

Define higher Weierstrass points (the $m$ -Weierstrass points associated to $K_{C}^{\otimes m}$ for $m \geq 1$ ) and compute their total weight for $m \geq 2$ .

Hint

Replace $H^{0} (K_{C})$ by $H^{0} (K_{C}^{\otimes m})$ , of dimension $r_{m} = (2 m - 1) (g - 1)$ for $m \geq 2$ by Riemann-Roch, and form the Wronskian of a basis; track which bundle the Wronskian lives in.

Answer

For $m \geq 2$ , $dim H^{0} (K_{C}^{\otimes m}) = (2 m - 1) (g - 1) =: r_{m}$ by Riemann-Roch and Serre duality ( $de g K_{C}^{\otimes m} = m (2 g - 2) > 2 g - 2$ , so the dual term vanishes). The Wronskian of a basis $s_{1}, \dots, s_{r_{m}}$ of $H^{0} (K_{C}^{\otimes m})$ , formed with derivatives up to order $r_{m} - 1$ , is a section of $$ \bigl(K_C^{\otimes m}\bigr)^{\otimes r_m} \otimes K_C^{\otimes \binom{r_m}{2}} = K_C^{\otimes\left(m r_m + \binom{r_m}{2}\right)}. $$ Its zeros are the $m$ -Weierstrass points, with weight $w^{(m)} (p)$ the order of vanishing. The total weight is the degree of this bundle: $$ \sum_p w^{(m)}(p) = \Bigl(m r_m + \tfrac{r_m(r_m - 1)}{2}\Bigr)(2g - 2) = r_m\Bigl(m + \tfrac{r_m - 1}{2}\Bigr)(2g - 2). $$ The ordinary case $m = 1$ recovers $r_{1} = g$ and $\sum w = g^{3} - g$ . These higher loci refine the canonical Weierstrass points and feature in the study of the pluricanonical maps and the geometry of $\overline{M}_{g}$ . Rubric: full credit for $r_{m} = (2 m - 1) (g - 1)$ , the twisted bundle, and the degree-based total-weight formula.

Exercise 6 (hard, short-answer).

Not every numerical semigroup of genus $g$ occurs as a Weierstrass semigroup of an actual curve. State what the gap theorem does and does not guarantee, and name the obstruction.

Hint

The Lückensatz says a Weierstrass semigroup has genus $g$ and lies in ${0} \cup [g + 1, \infty) \cup \dots$ with $g$ gaps; the converse — every genus- $g$ numerical semigroup is realised — is false (Buchweitz's examples).

Answer

The Weierstrass gap theorem guarantees that the semigroup $H (p)$ of non-gaps at any point of a genus- $g$ curve is a numerical semigroup of genus $g$ (exactly $g$ gaps, the largest at most $2 g - 1$ , $1$ always a gap). It does not guarantee the converse: there exist numerical semigroups of genus $g$ that are not the Weierstrass semigroup of any point on any smooth curve. Buchweitz 1980 gave the first non-realisable examples, using the obstruction that the dimension of a stratum of curves with prescribed semigroup, computed by a deformation-theoretic count on $M_{g}$ , can exceed the actual dimension $3 g - 3$ — a numerical inequality on the gaps (a sum over pairwise sums of gaps) is violated. So "Weierstrass semigroup" is a strictly smaller class than "numerical semigroup of genus $g$ "; the realisation problem is governed by deformation theory of the pointed curve, not by the combinatorics alone. Rubric: full credit for the genus- $g$ guarantee, the failure of the converse, and naming the deformation-theoretic/Buchweitz obstruction.

Advanced results Master

The Weierstrass theory sits at the junction of Riemann-Roch, the ramification of linear systems, and the combinatorics of numerical semigroups. Several strands organise the modern picture.

Ramification of an arbitrary linear series. The Wronskian construction is not special to $K_{C}$ . Given a linear series $g_{d}^{r} = (L, V)$ with $L$ a line bundle of degree $d$ and $V \subseteq H^{0} (C, L)$ of dimension $r + 1$ , the Wronskian of a basis of $V$ is a section of $L^{\otimes (r + 1)} \otimes K_{C}^{\otimes (2 r + 1)}$ , whose zeros are the ramification points of the series, weighted by their ramification order. The total ramification is $de g = (r + 1) d + (r + 1) r (g - 1)$ , the Plücker formula for linear series (Griffiths-Harris §2.2). The canonical series $∣ K_{C} ∣$ , with $r = g - 1$ and $d = 2 g - 2$ , recovers $\sum_{p} w (p) = g^{3} - g$ . Weierstrass points are thus the ramification of the canonical map; for a non-hyperelliptic curve the canonical map is an embedding $C ↪ P^{g - 1}$ and the Weierstrass points are the flexes (hyperosculating points) of the canonical curve, where the osculating hyperplane meets $C$ with multiplicity $> g - 1$ .

The hyperelliptic dichotomy. A curve of genus $g \geq 2$ is hyperelliptic exactly when it carries a $g_{2}^{1}$ , equivalently when some point has gap sequence $1, 3, \dots, 2 g - 1$ . The $2 g + 2$ branch points of the double cover are then the entire Weierstrass set, each of maximal weight $(2 g)$ , saturating the lower bound $2 g + 2$ on the count. Non-hyperelliptic curves have at least $2 g + 3$ Weierstrass points (in fact a generic curve has the maximal $g^{3} - g$ ordinary ones, all of weight $1$ ). The hyperelliptic Weierstrass points are also precisely the $2$ -torsion-related theta-characteristics underlying the genus- $g$ branch data, tying the local gap structure to the global theta geometry of 06.06.06.

Semigroups, the Frobenius number, and symmetry. The non-gaps $H (p)$ form a numerical semigroup of genus $g$ ; its largest gap, the Frobenius number, is at most $2 g - 1$ . The semigroup is symmetric (the gap set is symmetric about $(2 g - 1) /2$ : $n$ is a gap iff $2 g - 1 - n$ is a non-gap) exactly when $p$ is a Weierstrass point of a Gorenstein nature — and every Weierstrass semigroup arising from a smooth point is symmetric, a consequence of Serre duality applied to the multiplication $h^{0} (n p) \cdot h^{0} ((2 g - 2 - n) p)$ . The realisation problem — which genus- $g$ numerical semigroups occur — is open in general; Buchweitz 1980 produced non-Weierstrass semigroups via a dimension obstruction, and Eisenbud-Harris 1987 showed every semigroup with small enough weight is realised, using limit linear series.

Weierstrass points and moduli. The locus in $M_{g}$ of curves with an exceptional Weierstrass point (weight strictly larger than generic, or a special semigroup) stratifies moduli; the stratification by Weierstrass gap sequence refines the gonality stratification. Pinkham, Eisenbud-Harris, and others use degenerations to nodal/cuspidal curves, where the gap sequence becomes the semigroup of a singularity, to control these strata. The total-weight identity $g^{3} - g$ is the genus- $g$ instance of the general principle that ramification of a canonically defined linear system is a topological invariant.

Synthesis. Weierstrass points are the central insight that "specialness" on a curve is the vanishing of a canonical section, and this is exactly the unifying thread: the gap sequence is the local jump pattern of $h^{0} (n p)$ , the Wronskian is its global avatar in $K_{C}^{\otimes g (g + 1) /2}$ , and the foundational reason the total weight is the fixed number $g^{3} - g$ is that this bundle's degree is topological. Putting these together, the theory generalises in two directions at once — laterally to the ramification of arbitrary linear series via the Plücker formula, which is dual to the Brill-Noether existence statements, and vertically to the higher Weierstrass loci of pluricanonical systems that probe the geometry of $\overline{M}_{g}$ . The hyperelliptic case is exactly the extremal configuration of the count, where $2 g + 2$ points of maximal weight exhaust the budget, and this bridge between the local semigroup combinatorics and the global theta geometry of 06.06.06 is what makes Weierstrass points simultaneously a tool for bounding automorphisms (Hurwitz), a rigidifier of the curve, and a window onto the stratification of moduli. The same pattern recurs whenever a numerical invariant of sections is pinned by a degree computation, and the realisation problem for semigroups shows where the combinatorics outruns the geometry.

Full proof set Master

Proposition 1 (gap-jump dichotomy). For $p \in C$ and $n \geq 1$ , $h^{0} (O (n p)) - h^{0} (O ((n - 1) p)) \in {0, 1}$ , and there are exactly $g$ values of $n$ for which the difference is $0$ (the gaps), all lying in $[1, 2 g - 1]$ .

Proof. The inclusion $H^{0} (O ((n - 1) p)) \subseteq H^{0} (O (n p))$ has cokernel measured by the residue/leading-Laurent-coefficient map at $p$ , which lands in a one-dimensional space (the coefficient of $z^{- n}$ ), so the jump is $0$ or $1$ . For $n = 0$ , $h^{0} (O) = 1$ ; for $n \geq 2 g - 1$ , $de g (n p) = n \geq 2 g - 1 > 2 g - 2$ gives $h^{0} (O (n p)) = n + 1 - g$ by Riemann-Roch with vanishing $h^{1}$ , so the jump is $1$ for all $n \geq 2 g - 1$ . Summing the jumps from $n = 1$ to $\infty$ , the non-gaps below any large $N$ number $N - g$ (since $h^{0} (O (N p)) = N + 1 - g$ and the count starts at $h^{0} (O) = 1$ ), so the gaps number exactly $g$ , and all lie in $[1, 2 g - 1]$ because every $n \geq 2 g - 1$ is a non-gap. The smallest gap is $1$ : a function with a single simple pole would give an isomorphism $C ≅ P^{1}$ , impossible for $g \geq 1$ . $□$

Proposition 2 (weight equals Wronskian vanishing order). With notation as in the Formal definition, $ord_{p} W = w (p) = \sum_{i = 1}^{g} (n_{i} - i)$ .

Proof. The vanishing sequence of $∣ K_{C} ∣$ at $p$ is $a_{1} < \dots < a_{g}$ , where $a_{i}$ is the $i$ -th order to which some canonical differential vanishes at $p$ ; Riemann-Roch identifies $a_{i} = n_{i} - 1$ because $h^{0} (K_{C} - j p)$ drops precisely when $j - 1$ is a gap (Serre duality: $h^{0} (K_{C} - j p) = h^{1} (O (j p)) = h^{0} (O (j p)) - (j + 1 - g)$ ). Pick a basis $ω_{1}, \dots, ω_{g}$ adapted to the flag, with $ω_{i}$ vanishing to order exactly $a_{i}$ at $p$ ; in the local coordinate, $ω_{i} = (c_{i} z^{a_{i}} + \dots) d z$ with $c_{i} \neq = 0$ . The Wronskian determinant of $f_{i} = c_{i} z^{a_{i}} + \dots$ with derivatives up to order $g - 1$ has leading term of order $\sum_{i} a_{i} - \sum_{k = 0}^{g - 1} k = \sum_{i} a_{i} - (2 g)$ . Substituting $a_{i} = n_{i} - 1$ gives $\sum_{i} (n_{i} - 1) - (2 g) = \sum_{i} n_{i} - g - (2 g) = \sum_{i} (n_{i} - i)$ , since $\sum_{i} i = (2 g + 1) = (2 g) + g$ . Hence $ord_{p} W = w (p)$ . $□$

Proposition 3 (total weight). $\sum_{p \in C} w (p) = (g - 1) g (g + 1)$ .

Proof. By Proposition 2 the divisor of the global section $W$ of $K_{C}^{\otimes g (g + 1) /2}$ is $\sum_{p} w (p) p$ . The number of zeros of any non-zero holomorphic section of a line bundle equals the degree of that bundle, so $$ \sum_p w(p) = \deg K_C^{\otimes g(g+1)/2} = \frac{g(g+1)}{2},\deg K_C = \frac{g(g+1)}{2},(2g - 2) = (g-1)g(g+1). $$ The Wronskian is not identically zero: for $g \geq 1$ a basis of $H^{0} (K_{C})$ is linearly independent, so its Wronskian is generically non-zero, and a global section with finitely many zeros exists. $□$

Proposition 4 (extremal counts). For $g \geq 2$ the number $N$ of Weierstrass points satisfies $2 g + 2 \leq N \leq g^{3} - g$ , with the lower bound attained exactly by hyperelliptic curves.

Proof. Each Weierstrass point has $w (p) \geq 1$ , so $N \leq \sum_{p} w (p) = g^{3} - g$ , equality iff every Weierstrass point is ordinary (weight $1$ ). For the lower bound: $w (p) \leq (2 g)$ , since the gap sequence is term-by-term bounded by the hyperelliptic sequence $1, 3, \dots, 2 g - 1$ (the maximal allowable gaps given $n_{i} \leq 2 i - 1$ , which follows from $n_{g} \leq 2 g - 1$ and the strict increase). Thus $N \geq (g^{3} - g) / (2 g) = (g + 1) g (g - 1) / (g (g - 1) /2) = 2 (g + 1) = 2 g + 2$ . Equality forces $w (p) = (2 g)$ at every Weierstrass point, i.e. gap sequence $1, 3, \dots, 2 g - 1$ everywhere it is special, which is the hyperelliptic configuration (existence of a $g_{2}^{1}$ ). $□$

Connections Master

Riemann-Roch on compact Riemann surfaces 06.04.01. The entire gap/non-gap dichotomy is a corollary of Riemann-Roch: $n$ is a gap iff $h^{0} (O (n p)) = h^{0} (O ((n - 1) p))$ , and the count of exactly $g$ gaps in $[1, 2 g - 1]$ is the bookkeeping of Riemann-Roch as $n$ runs past $2 g - 2$ . The Wronskian section lives in $K_{C}^{\otimes g (g + 1) /2}$ , and its degree $-$ hence the total weight $-$ is computed from $de g K_{C} = 2 g - 2$ , the Riemann-Roch input. Without 06.04.01 there is no gap theorem and no $g^{3} - g$ .
Jacobi inversion and special divisors 06.06.06. The local jumps $h^{0} (n p) - h^{0} ((n - 1) p)$ are the pointwise shadow of the global Brill-Noether loci $W_{d}^{r} \subset Jac (C)$ studied through Jacobi inversion; a Weierstrass point of high weight signals a special line bundle $O (n p)$ with $h^{0} \geq 2$ , placing $n p$ in a non-generic stratum. The hyperelliptic Weierstrass points are exactly the points whose theta-geometry on the Jacobian is most degenerate, linking the gap combinatorics here to the theta-divisor singularities there.
Hyperelliptic curves and the $g_{2}^{1}$ . A curve is hyperelliptic iff some point has the maximal gap sequence $1, 3, \dots, 2 g - 1$ , iff it carries a degree- $2$ map to $P^{1}$ ; the Weierstrass points are then the $2 g + 2$ branch points, the ramification of that $g_{2}^{1}$ . This is the extremal case of the count in 06.06.06's Brill-Noether picture and the prototype for $k$ -gonal stratifications of the moduli space; the branch-point configuration also fixes the curve up to the choice of $2 g + 2$ points on $P^{1}$ , a rigidity used in reconstruction arguments.

Historical & philosophical context Master

The gap theorem originates with Karl Weierstrass, who in his Berlin lectures of the 1870s on Abelian functions isolated the Lückensatz — the assertion that at each point of an algebraic curve of genus $g$ there are exactly $g$ "gaps" in the sequence of attainable pole orders ^{[Weierstrass 1870s]}. Weierstrass did not publish the result himself in this form; it entered the literature through expositions of his lectures, notably Hensel and Landsberg's 1902 Theorie der algebraischen Funktionen einer Variabeln, which systematised the gap theorem and the points now bearing his name. The conceptual move was characteristically Weierstrassian: to extract an intrinsic, arithmetic invariant (a set of integers, a semigroup) from the transcendental theory of Abelian integrals, replacing analytic accidents with combinatorial structure.

The later reframing through the Wronskian and the ramification of the canonical map — the form in which the total weight $g^{3} - g$ becomes a single degree computation — belongs to the projective-geometric tradition consolidated by Griffiths and Harris, who present Weierstrass points as the hyperosculation loci of the canonical curve, and by Hurwitz, whose 1893 study of curve automorphisms used the permutation action on Weierstrass points to bound $Aut (C)$ by $84 (g - 1)$ . The philosophical lesson is the recurring one in the theory of curves: a finite, rigid, coordinate-free budget of "special behaviour" is forced by a topological invariant, and the combinatorics of the gap semigroup outruns the geometry — not every numerical semigroup of genus $g$ is realised, a gap between the possible and the actual that Buchweitz first exhibited and that remains only partially understood.

Bibliography Master

@book{GriffithsHarris1978,
  author    = {Griffiths, Phillip and Harris, Joseph},
  title     = {Principles of Algebraic Geometry},
  publisher = {Wiley-Interscience},
  year      = {1978},
  note      = {\S2.2: Wronskian, ramification of linear systems, Weierstrass points, total weight $g^3-g$}
}

@book{ACGH1985,
  author    = {Arbarello, Enrico and Cornalba, Maurizio and Griffiths, Phillip A. and Harris, Joseph},
  title     = {Geometry of Algebraic Curves, Volume I},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {267},
  publisher = {Springer},
  year      = {1985},
  note      = {Ch. I and exercise batches: gap sequences, Weierstrass semigroups, ramification of linear series}
}

@book{HenselLandsberg1902,
  author    = {Hensel, Kurt and Landsberg, Georg},
  title     = {Theorie der algebraischen Funktionen einer Variabeln},
  publisher = {Teubner, Leipzig},
  year      = {1902},
  note      = {First systematic published exposition of the Weierstrass Lückensatz and Weierstrass points}
}

@book{Miranda1995,
  author    = {Miranda, Rick},
  title     = {Algebraic Curves and Riemann Surfaces},
  series    = {Graduate Studies in Mathematics},
  volume    = {5},
  publisher = {American Mathematical Society},
  year      = {1995},
  note      = {\S VII.4: gap sequences and Weierstrass points via Riemann-Roch}
}

@book{FarkasKra1992,
  author    = {Farkas, Hershel M. and Kra, Irwin},
  title     = {Riemann Surfaces},
  edition   = {2nd},
  series    = {Graduate Texts in Mathematics},
  volume    = {71},
  publisher = {Springer},
  year      = {1992},
  note      = {\S III.5: Weierstrass points, weights, Wronskian, hyperelliptic branch points}
}

@article{Hurwitz1893,
  author    = {Hurwitz, Adolf},
  title     = {{\"U}ber algebraische Gebilde mit eindeutigen Transformationen in sich},
  journal   = {Mathematische Annalen},
  volume    = {41},
  pages     = {403--442},
  year      = {1893},
  note      = {Automorphisms permute Weierstrass points; the $84(g-1)$ bound}
}

@article{Buchweitz1980,
  author    = {Buchweitz, Ragnar-Olaf},
  title     = {On Zariski's criterion for equisingularity and non-smoothable monomial curves},
  journal   = {Preprint / thesis work},
  year      = {1980},
  note      = {First examples of numerical semigroups of genus $g$ not realised as Weierstrass semigroups}
}

@article{EisenbudHarris1987,
  author    = {Eisenbud, David and Harris, Joe},
  title     = {Existence, decomposition, and limits of certain Weierstrass points},
  journal   = {Inventiones Mathematicae},
  volume    = {87},
  pages     = {495--515},
  year      = {1987},
  note      = {Realisation of Weierstrass semigroups of small weight via limit linear series}
}

Prerequisites

06.04.01
06.06.06

Tier anchors

beginner: Miranda 1995 *Algebraic Curves and Riemann Surfaces* (AMS GSM 5) §VII.4 informal; Kirwan 1992 *Complex Algebraic Curves* (Cambridge LMS 23) informal narrative
intermediate: Griffiths-Harris 1978 *Principles of Algebraic Geometry* (Wiley) §2.2; Miranda 1995 *Algebraic Curves and Riemann Surfaces* (AMS GSM 5) §VII.4; Farkas-Kra 1992 *Riemann Surfaces* (Springer GTM 71) §III.5
master: Griffiths-Harris 1978 *Principles of Algebraic Geometry* (Wiley) §2.2; Arbarello-Cornalba-Griffiths-Harris 1985 *Geometry of Algebraic Curves I* (Springer GMW 267) Ch. I; Hensel-Landsberg 1902 *Theorie der algebraischen Funktionen einer Variabeln* (Teubner) §§ on the Lückensatz

References

Weierstrass — Lückensatz (gap theorem) · Weierstrass lectures of the 1870s on Abelian functions; first published exposition in Hensel-Landsberg 1902 Theorie der algebraischen Funktionen einer Variabeln, Teubner (gap theorem and Weierstrass points)
Griffiths-Harris 1978 — Principles of Algebraic Geometry · Wiley, §2.2 (Wronskian, ramification of linear systems, Weierstrass points, total weight g^3-g)
Arbarello-Cornalba-Griffiths-Harris 1985 — Geometry of Algebraic Curves I · Springer GMW 267, Ch. I and Exercise batch I.E (Weierstrass points, gap sequences, semigroups, ramification of linear series)
Miranda 1995 — Algebraic Curves and Riemann Surfaces · AMS GSM 5, §VII.4 (gap sequences and Weierstrass points via Riemann-Roch)
Farkas-Kra 1992 — Riemann Surfaces · Springer GTM 71, §III.5 (Weierstrass points, weights, Wronskian; hyperelliptic case)
Hurwitz 1893 — Über algebraische Gebilde mit eindeutigen Transformationen in sich · Math. Ann. 41, 403-442 (automorphisms permute Weierstrass points; the 84(g-1) bound)

Estimated time

beginner: 16m
intermediate: 42m
master: 78m