Computational algebraic geometry
Learning coefficients via symbolic and numerical methods
Anton Leykin
Georgia Tech
Ideals
↔
varieties
• Letkbe a field (RorC). • IdealinR=k[x1, . . . , xn]generated byf1, . . . , fris I=hf1, . . . , fri= r X i=1 gifi⊂R • Varietydefined byIis V =V(I) ={p∈kn:∀f ∈I, f(p) = 0}• Thevanishing idealof a varietyV is
I=I(V) ={f ∈R:f(V) = 0} • TheradicalofIis √ I={g:∃m, gm∈I} ⊂R, V(I) =V( √ I) Iis a radical ideal ifI=√I
• Hilbert Nullstellensatz: ifk=CthenI(V(J)) =
√
Gröbner bases, invariants
• Amonomial order>is given by weightsw1, w2, . . .∈Rn≥0:
xα> xβ ⇔ w1·α > w1·β or
w2·α > w2·β or
. . . such that>is total.
Example: >lex is given byw1=e1, w2=e2, . . ..
• The initial term off ∈Ris the term (coefficient·monomial) with the largest monomial occurring with a nonzero coefficient:
e.g. forf = 3x2y+ 4xy7+ 2it isin
lex(f) = 3x2y. • AGröbner basisofIis a set of generatorsGofIsuch that
hin(G)i=hin(I)i.
• Invariantssuch as dimension, degree, Hilbert polynomial can be computed using Gröbner bases.
Computational (symbolic) algebraic geometry
• Cox, Little, O’Shea,Ideals, varieties, and algorithms
• Cox, Little, O’Shea,Using algebraic geometry
• Greuel, Pfister,A Singular Introduction to Commutative Algebra
Software:
• Singular(can resolve singularities) • Macaulay2(can do everything else)
• Magma(Australia),CoCoA(Italy),risa/asir(Japan), ... (other specialized software)
Polynomial homotopy continuation
• Targetsystem: nequations innvariables, F(x) = (f1(x), . . . , fn(x)) =0, wherefi∈R=C[x] =C[x1, ..., xn]fori= 1, ..., n. • Startsystem:nequations innvariables:
G(x) = (g1(x), . . . , gn(x)) =0,
such that it is easy to solve.
• Homotopy: forγ∈C\ {0}consider
Example
target start
f1 = x41x2+ 5x21x23+x31−4 g1 = x51−1
f2 = x21−x1x2+x2−8 g2 = x22−1
Start solutions→target solutions: H(x, t) = 0implies dx dt =− ∂H ∂x −1 ∂H ∂t .
Global picture
Optimal homotopy:
• the continuation paths areregular; • the homotopy establishes a bijection
between the start and target solutions. Possiblesingularscenarios:
Higher-dimensional solution sets
• LetI= (f1, . . . , fN)be an ideal ofC[x1, . . . , xn]. • Goal: Understand the variety
X=V(I) ={x∈Cn| ∀f ∈I, f(x) = 0}. • Awitness setfor an equidimensional componentY ofX:
• a generic “slicing” planeLwithdimL= codimY
• witness pointswY,L=Y ∩L • (generators ofI)
Numerical algebraic geometry
• Sommese, Verschelde, and Wampler,Introduction to Numerical AG(2005)
• Sommese and Wampler,The numerical solution of systems of polynomials(2005)
Software:
• PHCpack(Verschelde); • HOM4PS(group of T.Y.Li); • Bertini(group of Sommese);
• NAG4M2: Numerical Algebraic Geometry for Macaulay2 (L.). and more, e.g.:Maple’s ROOTFINDING[HOMOTOPY].
Analysis
• For a polynomialf, the function|f|sis locally integrableRes >0. • Hence,|f|sis a generalized function defined on
{s: Res >0} ⊂C: for a distributionϕ∈Cc∞,
|f|s(ϕ) = Z
|f(x)|sϕ(x)dx
• Gelfand [1957]: Does it extend to a meromorphic function onC? • I. N. Bernstein [1968]: Yes. The poles are contained in a finite
number of arithmetic progressions.
• Key ingredients:resolution of singularitiesand being able to write afunctional equation
b(s)fs=P·fs+1
whenf is a monomial. (Here:b(s)is a univariate polynomial and P is a linear differential operator with coefficients inC[x, s].)
Invariants in singularity theory
Definition (Multiplier ideal for
f
= (f1
, . . . , f
r)
)
J(fc) = h∈C[x] : |h|2 (P |fi|2)c is locally integrable . Forr= 1, it is the ideal ofh, that make |f|h|
1|c locally integrable.
• Algebrao-geometric definition: via log-canonical resolutions. • Jumping coefficients off: rational numbers
0 =ξ0< ξ1< ξ2<· · ·
such thatJ(fc)is constant exactly forc∈[ξi, ξi+1).
• ξ1is called thelog-canonical threshold.
• These invariants measure singularities of the corresponding variety; in particular, they depend only on the idealhfi.
Weyl algebra
• LetKbe a field of characteristic zero. (Think:K=C) • Affine space: X =Kn.
• Weyl algebra: an associative algebra
DX =Khx,∂i=Khx1, . . . , xn, ∂1, . . . , ∂ni
where[∂i, xi] =∂ixi−xi∂i= 1and all other pairs of generators commute.
• DXis isomorphic to the algebra oflinear differential operators
with polynomial coefficients. • Every element has thenormal form
Q= X
α,β∈Zn
cαβxα∂β,
D
-modules
• DXis simple: only trivial two-sided ideals.
• We consider onlyleftideals andleftDX-modules. • Examples ofD-modules:K[x],K[[x]],C∞(X). • Software:
• kan/sm1(Takayama)
• risa/asir(Noro)
• dmod.lib, Singular(Levandovsky et al.)
Gröbner bases
• DXis Gröbner-friendly: DX is an algebra ofsolvable type. • Gröbner bases can be computed with respect to any
w-compatible monomial order, wherew= (wx, w∂)∈R2n satisfieswx+w∂ ≥0componentwise.
• TheBernstein-Satopolynomialbf(s)6= 0is the monic polynomial b(s)of the minimal degree satisfying
b(s)fs=P·fs+1=P f·fs∈DX[s]fs, whereP ∈DX[s]andDX[s]fsis a cyclicDX[s]-module generated byfs.
Connection between
lct(
f
)
and
b
f(
s
)
• Assumek=Cthen
• The log-canonical thresholdc0is the lowest root ofbf(−s). • Every jumping coefficientc∈[c0, c0+ 1)is a root ofbf(−s). • Fork=Rthere are examples whererlct(f)is not a jumping
coefficient off.
However,rlct(f) mod 1equals some root ofbf(−s).
• M2 packages wherelctis computed: • Dmodules(general case)
• MonomialMultiplierIdeals(monomial ideals)
Improper integrals
To findc= rlct(f)look for the poles=−cclosest to 0 of
|f|s(ϕ) = Z
x∈Rn
|f(x)|sϕ(x)dx. Idea:
• Pickϕsupported on a neighborhood of a singularity of the variety{f = 0}and evaluate|f|s(ϕ)numerically.
• If|f|−c1(ϕ)<∞and|f|−c2(ϕ) =∞thenc
1<rlct(f)≤c2.
Questions
• Can we say something about the denominator ofrlct(f)? • How to compute an improper integral numerically?
How to determine it equals∞? • Can we use statistics to estimaterlct?