Weierstrass preparation and division

Written by Claude Opus 5.5 (Anthropic), October 2026; Exercise 1 and its solution by GPT-6.1 Sol (OpenAI). Original text: CC0.

Near the origin, a holomorphic function that does not vanish identically on the last coordinate axis behaves like a polynomial in the last coordinate. It is a nonvanishing factor times a monic polynomial whose roots tend to zero, and every other holomorphic function can be divided by it with a polynomial remainder. This reading proves both statements, gives division constants that do not depend on the dividend, and deduces that the ring of convergent power series is Noetherian. In this course it supplies the analytic input for Classical finite-order preparation and division.

Conventions. We write π’ͺn=β„‚{z1,…,zn}\mathcal O_n=\mathbb C\{z_1,\ldots,z_n\} for the ring of germs at zero of holomorphic functions, z=(zβ€²,w)z=(z',w) with w=znw=z_n, Δρ={wβˆˆβ„‚:|w|<ρ}\Delta_\rho=\{w\in\mathbb C:|w|<\rho\} and Δδ′={zβ€²βˆˆβ„‚nβˆ’1:|zk|<Ξ΄ for all k}\Delta'_\delta=\{z'\in\mathbb C^{n-1}:|z_k|<\delta\text{ for all }k\}. For n=1n=1 the factor β„‚nβˆ’1\mathbb C^{n-1} is a point. A function of several variables is holomorphic when it is continuous and holomorphic in each variable separately, as in Holomorphic functions of several variables, Definition 1.1; Theorem 2.1 there gives Taylor expansions and holomorphic partial derivatives. The one-variable results used below are proved in the contour foundation: contour integrals and the triangle theorem (Β§16.1), Cauchy’s formula on discs, Taylor coefficients and the identity theorem (Β§16.2), holomorphic integrals depending on a parameter (Β§16.3), Cauchy’s theorem (Β§16.5) and the maximum principle (Β§16.6). Polynomial roots, Β§9, gives the factorization of a complex polynomial into linear factors with multiplicities.

A germ g∈π’ͺng\in\mathcal O_n is regular of order ss in ww if g(0,w)=wsv(w)g(0,w)=w^sv(w) with v(0)β‰ 0v(0)\neq0; for s=0s=0 this means g(0)β‰ 0g(0)\neq0. A Weierstrass polynomial of degree ss is a polynomial P=ws+a1(zβ€²)wsβˆ’1+β‹―+as(zβ€²)P=w^s+a_1(z')w^{s-1}+\cdots+a_s(z') with ak∈π’ͺnβˆ’1a_k\in\mathcal O_{n-1} and ak(0)=0a_k(0)=0.

The statements

Theorem 1 (preparation). If gg is regular of order ss in ww, then g=uPg=uP for a unit u∈π’ͺnu\in\mathcal O_n and a Weierstrass polynomial PP of degree ss. Both factors are unique.

Theorem 2 (division with constants independent of the dividend). Let Dβ€²βŠ‚β„‚nβˆ’1D'\subset\mathbb C^{n-1} be open, sβ‰₯1s\geq1, and P(zβ€²,w)=ws+βˆ‘k=1sak(zβ€²)wsβˆ’kP(z',w)=w^s+\sum_{k=1}^sa_k(z')w^{s-k} with aka_k holomorphic on Dβ€²D'. Let 0≀α<ρ0\leq\alpha<\rho, and suppose that for every zβ€²βˆˆDβ€²z'\in D' all roots of P(zβ€²,β‹…)P(z',\cdot) have modulus at most Ξ±\alpha. Then every holomorphic ff on D′×ΔρD'\times\Delta_\rho has exactly one expression

f=qP+R,q∈π’ͺ(D′×Δρ),R=βˆ‘j=0sβˆ’1cj(zβ€²)wj,cj∈π’ͺ(Dβ€²). f=qP+R,\qquad q\in\mathcal O(D'\times\Delta_\rho),\qquad R=\sum_{j=0}^{s-1}c_j(z')w^j,\quad c_j\in\mathcal O(D').

If |f|≀M|f|\leq M on D′×ΔρD'\times\Delta_\rho, then |R|≀C1M|R|\leq C_1M and |q|≀C2M|q|\leq C_2M there, where

C1=sρ(ρ+Ξ±)sβˆ’1(Οβˆ’Ξ±)s,C2=1+C1(Οβˆ’Ξ±)s.(6) C_1=\frac{s\rho(\rho+\alpha)^{s-1}}{(\rho-\alpha)^s},\qquad C_2=\frac{1+C_1}{(\rho-\alpha)^s}. \qquad\text{(6)}

Corollary 3 (division by a regular germ). If gg is regular of order ss in ww, every f∈π’ͺnf\in\mathcal O_n has exactly one expression f=qg+Rf=qg+R with q∈π’ͺnq\in\mathcal O_n and R∈π’ͺnβˆ’1[w]R\in\mathcal O_{n-1}[w] of degree less than ss. Hence π’ͺn/(g)\mathcal O_n/(g) is a free π’ͺnβˆ’1\mathcal O_{n-1}-module with basis the classes of 1,w,…,wsβˆ’11,w,\ldots,w^{s-1}.

Proposition 4 (adapted coordinates). For finitely many nonzero germs there is a linear change of coordinates after which each of them is regular in ww, of order equal to its order at zero.

Proposition 5 (size of the coefficients). If gg is regular in ww of order ss and ss is also the order of gg at zero, the coefficients of its Weierstrass polynomial satisfy |ak(zβ€²)|≀(sk)Ck|zβ€²|k|a_k(z')|\leq\binom skC^k|z'|^k near zero, for some constant CC. Here |zβ€²|=max⁡k|zk||z'|=\max_k|z_k|.

Theorem 6. The ring π’ͺn\mathcal O_n is Noetherian.

Three tools

Lemma A (integrals over a fixed circle). Let UβŠ‚β„‚mU\subset\mathbb C^m be open and Οƒ>0\sigma>0. Let Ο†\varphi be continuous on UΓ—{|ΞΆ|=Οƒ}U\times\{|\zeta|=\sigma\} and holomorphic in x∈Ux\in U for each fixed ΞΆ\zeta. Then Ξ¦(x)=∫|ΞΆ|=σφ(x,ΞΆ)dΞΆ\Phi(x)=\int_{|\zeta|=\sigma}\varphi(x,\zeta)\,d\zeta is holomorphic on UU.

Proof. For a compact neighbourhood KK of a point of UU, the integrand is uniformly continuous and bounded on KΓ—{|ΞΆ|=Οƒ}K\times\{|\zeta|=\sigma\}; hence Ξ¦\Phi is continuous. Fix all variables but one. The integrand is holomorphic in the remaining variable and bounded on compact sets, so Β§16.3 of the contour foundation, with the circle and its arc-length measure, shows that Ξ¦\Phi is holomorphic in that variable. Thus Ξ¦\Phi is continuous and holomorphic in each variable separately. β–«\square

Lemma B (zeros inside a circle). Let hh be holomorphic on Δτ\Delta_\tau, 0<Οƒ<Ο„0<\sigma<\tau, and h(ΞΆ)β‰ 0h(\zeta)\neq0 for |ΞΆ|=Οƒ|\zeta|=\sigma. Then hh has finitely many zeros in Δσ\Delta_\sigma. List them as b1,…,bNb_1,\ldots,b_N, each repeated according to its multiplicity. For every integer jβ‰₯0j\geq0,

12Ο€i∫|ΞΆ|=σ΢jhβ€²(ΞΆ)h(ΞΆ)dΞΆ=b1j+β‹―+bNj,(1) \frac1{2\pi i}\int_{|\zeta|=\sigma}\zeta^j\,\frac{h'(\zeta)}{h(\zeta)}\,d\zeta=b_1^j+\cdots+b_N^j, \qquad\text{(1)}

with b0=1b^0=1, so that the case j=0j=0 gives NN.

Proof. The function hh is not identically zero on the connected disc Δτ\Delta_\tau, so its zeros are isolated by the identity theorem. The compact disc {|ΞΆ|≀σ}\{|\zeta|\leq\sigma\} therefore contains only finitely many of them, all in the open disc. If cc is a zero of multiplicity mm, Taylor expansion at cc gives h(ΞΆ)=(ΞΆβˆ’c)mhc(ΞΆ)h(\zeta)=(\zeta-c)^mh_c(\zeta), with hch_c holomorphic on Δτ\Delta_\tau and hc(c)β‰ 0h_c(c)\neq0. Removing the distinct zeros c1,…,cpc_1,\ldots,c_p in Δσ\Delta_\sigma one after another gives

h(ΞΆ)=∏i=1p(ΞΆβˆ’ci)mik(ΞΆ), h(\zeta)=\prod_{i=1}^p(\zeta-c_i)^{m_i}\,k(\zeta),

where kk is holomorphic on Δτ\Delta_\tau and has no zero with |ΞΆ|≀σ|\zeta|\leq\sigma. By compactness and continuity, kk has no zero on some larger disc Δσ′\Delta_{\sigma'}, Οƒ<Οƒβ€²<Ο„\sigma<\sigma'<\tau. There, away from the cic_i,

ΞΆjhβ€²(ΞΆ)h(ΞΆ)=βˆ‘i=1pmiΞΆjΞΆβˆ’ci+ΞΆjkβ€²(ΞΆ)k(ΞΆ). \zeta^j\frac{h'(\zeta)}{h(\zeta)}=\sum_{i=1}^pm_i\frac{\zeta^j}{\zeta-c_i}+\zeta^j\frac{k'(\zeta)}{k(\zeta)} .

The last term is holomorphic on the convex set Δσ′\Delta_{\sigma'}, so its integral over the circle is zero by Cauchy’s theorem. Cauchy’s formula for the polynomial ΞΆj\zeta^j gives 12Ο€i∫|ΞΆ|=σ΢j(ΞΆβˆ’ci)βˆ’1dΞΆ=cij\frac1{2\pi i}\int_{|\zeta|=\sigma}\zeta^j(\zeta-c_i)^{-1}\,d\zeta=c_i^j. Summing gives βˆ‘imicij\sum_im_ic_i^j, which is the right side of (1). β–«\square

Lemma C (coefficients from power sums). Let b1,…,bNβˆˆβ„‚b_1,\ldots,b_N\in\mathbb C, repetitions allowed, Sj=βˆ‘lbljS_j=\sum_lb_l^j for jβ‰₯1j\geq1, and ∏l=1N(wβˆ’bl)=wN+A1wNβˆ’1+β‹―+AN\prod_{l=1}^N(w-b_l)=w^N+A_1w^{N-1}+\cdots+A_N. With A0=1A_0=1,

kAk=βˆ’(S1Akβˆ’1+S2Akβˆ’2+β‹―+SkA0),1≀k≀N.(C) kA_k=-\bigl(S_1A_{k-1}+S_2A_{k-2}+\cdots+S_kA_0\bigr),\qquad1\leq k\leq N. \qquad\text{(C)}

Thus every AkA_k is a polynomial with rational coefficients in S1,…,SkS_1,\ldots,S_k; no ordering or labelling of the blb_l enters.

Proof. Put Q(w)=∏l(wβˆ’bl)Q(w)=\prod_l(w-b_l) and Ξ²=max⁡l|bl|\beta=\max_l|b_l|. For |w|>Ξ²|w|>\beta (any wβ‰ 0w\neq0 if Ξ²=0\beta=0), expanding each 1/(wβˆ’bl)1/(w-b_l) as a geometric series gives the absolutely convergent expansion

Qβ€²(w)Q(w)=βˆ‘l=1N1wβˆ’bl=βˆ‘jβ‰₯0Sjwβˆ’jβˆ’1,S0=N. \frac{Q'(w)}{Q(w)}=\sum_{l=1}^N\frac1{w-b_l}=\sum_{j\geq0}S_jw^{-j-1},\qquad S_0=N .

Multiply by Q(w)Q(w), substitute w=1/tw=1/t and multiply by tNβˆ’1t^{N-1}. For 0<|t|<1/Ξ²0<|t|<1/\beta this gives

βˆ‘k=0Nβˆ’1(Nβˆ’k)Aktk=βˆ‘i=0Nβˆ‘jβ‰₯0AiSjti+j. \sum_{k=0}^{N-1}(N-k)A_kt^k=\sum_{i=0}^N\sum_{j\geq0}A_iS_j\,t^{i+j}.

Both sides are power series in tt converging for |t|<1/Ξ²|t|<1/\beta (for all tt if Ξ²=0\beta=0), and they agree off t=0t=0, hence also at t=0t=0 by continuity. Their coefficients therefore agree by the uniqueness of Taylor coefficients (Β§16.2 of the contour foundation). The coefficient of tkt^k on the right is NAk+βˆ‘j=1kSjAkβˆ’jNA_k+\sum_{j=1}^kS_jA_{k-j}. On the left it is (Nβˆ’k)Ak(N-k)A_k for k<Nk<N and 00 for k=Nk=N. Comparing the two gives (C). β–«\square

Preparation by the zeros in one fibre

Proof of Theorem 1. If s=0s=0, then g(0)β‰ 0g(0)\neq0. The reciprocal 1/g1/g is holomorphic near zero, so gg is a unit and g=gβ‹…1g=g\cdot1. Assume sβ‰₯1s\geq1.

Choice of a circle. Let gg be holomorphic on Δδ0′×Δρ0\Delta'_{\delta_0}\times\Delta_{\rho_0}. Since g(0,w)=wsv(w)g(0,w)=w^sv(w) with v(0)β‰ 0v(0)\neq0, there is Οƒβˆˆ(0,ρ0)\sigma\in(0,\rho_0) with g(0,w)β‰ 0g(0,w)\neq0 for 0<|w|≀σ0<|w|\leq\sigma. Put ΞΌ=min⁡|ΞΆ|=Οƒ|g(0,ΞΆ)|>0\mu=\min_{|\zeta|=\sigma}|g(0,\zeta)|>0. The function gg is uniformly continuous on the compact set {|zk|≀δ0/2}Γ—{|ΞΆ|=Οƒ}\{|z_k|\leq\delta_0/2\}\times\{|\zeta|=\sigma\}. Hence some δ∈(0,Ξ΄0/2]\delta\in(0,\delta_0/2] gives |g(zβ€²,ΞΆ)βˆ’g(0,ΞΆ)|<ΞΌ|g(z',\zeta)-g(0,\zeta)|<\mu, and so g(zβ€²,ΞΆ)β‰ 0g(z',\zeta)\neq0, whenever zβ€²βˆˆΞ”Ξ΄β€²z'\in\Delta'_\delta and |ΞΆ|=Οƒ|\zeta|=\sigma.

Power sums of the zeros. For zβ€²βˆˆΞ”Ξ΄β€²z'\in\Delta'_\delta and jβ‰₯0j\geq0 define

Sj(zβ€²)=12Ο€i∫|ΞΆ|=σ΢jβˆ‚wg(zβ€²,ΞΆ)g(zβ€²,ΞΆ)dΞΆ.(2) S_j(z')=\frac1{2\pi i}\int_{|\zeta|=\sigma}\zeta^j\,\frac{\partial_wg(z',\zeta)}{g(z',\zeta)}\,d\zeta . \qquad\text{(2)}

The derivative βˆ‚wg\partial_wg is holomorphic, so the integrand is continuous on Δδ′×{|ΞΆ|=Οƒ}\Delta'_\delta\times\{|\zeta|=\sigma\} and holomorphic in zβ€²z'. By Lemma A, each SjS_j is holomorphic on Δδ′\Delta'_\delta. By Lemma B applied to g(zβ€²,β‹…)g(z',\cdot) on Δρ0\Delta_{\rho_0}, S0(zβ€²)S_0(z') is the number of zeros of g(zβ€²,β‹…)g(z',\cdot) in Δσ\Delta_\sigma, counted with multiplicity, and Sj(zβ€²)S_j(z') is the sum of their jj-th powers. A continuous integer-valued function on the connected polydisc Δδ′\Delta'_\delta is constant. At zβ€²=0z'=0 the only zero in Δσ\Delta_\sigma is w=0w=0, of multiplicity ss. Hence every fibre g(zβ€²,β‹…)g(z',\cdot) has exactly ss zeros in Δσ\Delta_\sigma, counted with multiplicity; list them as b1(zβ€²),…,bs(zβ€²)b_1(z'),\ldots,b_s(z'). We do not claim, and do not need, that individual zeros depend holomorphically on zβ€²z'. For w2βˆ’z1w^2-z_1, no holomorphic choice of a square root of z1z_1 exists near z1=0z_1=0.

The polynomial. Put a0=1a_0=1 and ak=βˆ’(S1akβˆ’1+β‹―+Ska0)/ka_k=-(S_1a_{k-1}+\cdots+S_ka_0)/k for 1≀k≀s1\leq k\leq s. These functions are holomorphic on Δδ′\Delta'_\delta. By Lemma C, for every zβ€²βˆˆΞ”Ξ΄β€²z'\in\Delta'_\delta,

P(zβ€²,w)=ws+a1(zβ€²)wsβˆ’1+β‹―+as(zβ€²)=∏l=1s(wβˆ’bl(zβ€²)).(3) P(z',w)=w^s+a_1(z')w^{s-1}+\cdots+a_s(z')=\prod_{l=1}^s\bigl(w-b_l(z')\bigr). \qquad\text{(3)}

At zβ€²=0z'=0 all blb_l vanish, so P(0,w)=wsP(0,w)=w^s and ak(0)=0a_k(0)=0. Thus PP is a Weierstrass polynomial of degree ss.

The unit. Fix zβ€²βˆˆΞ”Ξ΄β€²z'\in\Delta'_\delta. Off the points bl(zβ€²)b_l(z'), the quotient g(zβ€²,w)/P(zβ€²,w)g(z',w)/P(z',w) is holomorphic on Δρ0\Delta_{\rho_0}. If a value bb occurs exactly mm times among the bl(zβ€²)b_l(z'), then both g(zβ€²,β‹…)g(z',\cdot) and P(zβ€²,β‹…)P(z',\cdot) vanish at bb to order exactly mm. Write g(zβ€²,w)=(wβˆ’b)mgb(w)g(z',w)=(w-b)^mg_b(w) and P(zβ€²,w)=(wβˆ’b)mPb(w)P(z',w)=(w-b)^mP_b(w) with gb(b)β‰ 0β‰ Pb(b)g_b(b)\neq0\neq P_b(b). The quotient then extends across bb with the nonzero value gb(b)/Pb(b)g_b(b)/P_b(b). Call the extended function Ezβ€²E_{z'}. It is holomorphic on Δρ0\Delta_{\rho_0} and has no zero in Δσ\Delta_\sigma, because every zero of g(zβ€²,β‹…)g(z',\cdot) there is one of the bl(zβ€²)b_l(z'). All roots of P(zβ€²,β‹…)P(z',\cdot) lie in Δσ\Delta_\sigma, so P(zβ€²,ΞΆ)β‰ 0P(z',\zeta)\neq0 for |ΞΆ|=Οƒ|\zeta|=\sigma. Cauchy’s formula gives, for |w|<Οƒ|w|<\sigma,

Ezβ€²(w)=12Ο€i∫|ΞΆ|=Οƒg(zβ€²,ΞΆ)P(zβ€²,ΞΆ)(ΞΆβˆ’w)dΞΆ. E_{z'}(w)=\frac1{2\pi i}\int_{|\zeta|=\sigma}\frac{g(z',\zeta)}{P(z',\zeta)(\zeta-w)}\,d\zeta .

The integrand is continuous on Δδ′×Δσ×{|ΞΆ|=Οƒ}\Delta'_\delta\times\Delta_\sigma\times\{|\zeta|=\sigma\} and holomorphic in (zβ€²,w)(z',w). By Lemma A, u(zβ€²,w)=Ezβ€²(w)u(z',w)=E_{z'}(w) is holomorphic on V=Δδ′×ΔσV=\Delta'_\delta\times\Delta_\sigma. It has no zero there, so 1/u1/u is holomorphic as well. The functions gg and uPuP are continuous on VV and agree off the finitely many points bl(zβ€²)b_l(z') of each fibre, so they agree on VV. Hence g=uPg=uP with u(0)β‰ 0u(0)\neq0.

Uniqueness. First a bound for roots: if Ο„>0\tau>0 and βˆ‘k=1s|ck|Ο„βˆ’k<1\sum_{k=1}^s|c_k|\tau^{-k}<1, then every root of ws+c1wsβˆ’1+β‹―+csw^s+c_1w^{s-1}+\cdots+c_s lies in Δτ\Delta_\tau. Indeed, for |w|β‰₯Ο„|w|\geq\tau the lower terms have modulus at most |w|sβˆ‘k|ck||w|βˆ’k≀|w|sβˆ‘k|ck|Ο„βˆ’k<|w|s|w|^s\sum_k|c_k||w|^{-k}\leq|w|^s\sum_k|c_k|\tau^{-k}<|w|^s.

Now suppose g=u1P1g=u_1P_1 as germs, with u1u_1 a unit and P1=ws1+c1(zβ€²)ws1βˆ’1+β‹―+cs1(zβ€²)P_1=w^{s_1}+c_1(z')w^{s_1-1}+\cdots+c_{s_1}(z') a Weierstrass polynomial. At zβ€²=0z'=0 this reads g(0,w)=u1(0,w)ws1g(0,w)=u_1(0,w)w^{s_1}, so s1=ss_1=s. Choose Ο„1∈(0,Οƒ]\tau_1\in(0,\sigma] and Ξ΄1∈(0,Ξ΄]\delta_1\in(0,\delta] such that on Δδ1′×Δτ1\Delta'_{\delta_1}\times\Delta_{\tau_1} the identity g=u1P1g=u_1P_1 holds with u1u_1 holomorphic and without zeros. Then choose Ο„βˆˆ(0,Ο„1)\tau\in(0,\tau_1) and shrink Ξ΄1\delta_1 so that βˆ‘k|ak(zβ€²)|Ο„βˆ’k<1\sum_k|a_k(z')|\tau^{-k}<1 and βˆ‘k|ck(zβ€²)|Ο„βˆ’k<1\sum_k|c_k(z')|\tau^{-k}<1 for zβ€²βˆˆΞ”Ξ΄1β€²z'\in\Delta'_{\delta_1}. This is possible because ak(0)=ck(0)=0a_k(0)=c_k(0)=0 and the coefficients are continuous. For such zβ€²z', the zeros of g(zβ€²,β‹…)g(z',\cdot) in Δτ1\Delta_{\tau_1}, with multiplicities, are the roots of P1(zβ€²,β‹…)P_1(z',\cdot), and these lie in Δτ\Delta_\tau. The zeros of g(zβ€²,β‹…)g(z',\cdot) in Δσ\Delta_\sigma are the roots of P(zβ€²,β‹…)P(z',\cdot), and they also lie in Δτ\Delta_\tau. Two monic polynomials of degree ss with the same roots and multiplicities are equal, by the factorization of complex polynomials into linear factors. Hence P1=PP_1=P. Then u1=g/P=uu_1=g/P=u wherever Pβ‰ 0P\neq0, and by continuity everywhere on the common domain. β–«\square

Proof of Proposition 4. Let gβ‰ 0g\neq0 have order mm, so that its Taylor series is βˆ‘dβ‰₯mgd\sum_{d\geq m}g_d with homogeneous parts gdg_d of degree dd and gmβ‰ 0g_m\neq0. If vβˆˆβ„‚nv\in\mathbb C^n satisfies gm(v)β‰ 0g_m(v)\neq0, choose an invertible linear map AA with Aen=vAe_n=v and put gΜƒ(z)=g(Az)\tilde g(z)=g(Az). For small ww, gΜƒ(0,w)=g(wv)=βˆ‘dβ‰₯mgd(v)wd=gm(v)wm+O(wm+1)\tilde g(0,w)=g(wv)=\sum_{d\geq m}g_d(v)w^d=g_m(v)w^m+O(w^{m+1}), so gΜƒ\tilde g is regular of order mm in ww. For finitely many nonzero germs g(1),…,g(p)g^{(1)},\ldots,g^{(p)} with lowest homogeneous parts gmi(i)g^{(i)}_{m_i}, the product ∏igmi(i)\prod_ig^{(i)}_{m_i} is a nonzero polynomial, since the polynomial ring is an integral domain. A nonzero polynomial does not vanish on all of β„‚n\mathbb C^n: by induction on nn, fix the first nβˆ’1n-1 variables so that some coefficient of the highest power of the last variable is nonzero, and use that a nonzero polynomial in one variable has finitely many roots. One vector vv with ∏igmi(i)(v)β‰ 0\prod_ig^{(i)}_{m_i}(v)\neq0 serves for all of them. β–«\square

Proof of Proposition 5. Write the lowest homogeneous part as gs(zβ€²,w)=cws+βˆ‘k=0sβˆ’1hk(zβ€²)wkg_s(z',w)=cw^s+\sum_{k=0}^{s-1}h_k(z')w^k, where c=gs(0,1)β‰ 0c=g_s(0,1)\neq0 and hkh_k is homogeneous of degree sβˆ’ks-k. Choose HH with |hk(zβ€²)|≀H|zβ€²|sβˆ’k|h_k(z')|\leq H|z'|^{s-k} for all kk. The terms of order at least s+1s+1 satisfy |g(z)βˆ’gs(z)|≀K|z|s+1|g(z)-g_s(z)|\leq K|z|^{s+1} for |z|=max⁡(|zβ€²|,|w|)≀η|z|=\max(|z'|,|w|)\leq\eta, with suitable K,Ξ·>0K,\eta>0. Indeed, if βˆ‘Ξ²|cΞ²|Ξ·|Ξ²|<∞\sum_\beta|c_\beta|\eta^{|\beta|}<\infty for the Taylor coefficients cΞ²c_\beta of gg, then βˆ‘|Ξ²|β‰₯s+1|cΞ²||zΞ²|≀|z|s+1Ξ·βˆ’sβˆ’1βˆ‘Ξ²|cΞ²|Ξ·|Ξ²|\sum_{|\beta|\geq s+1}|c_\beta||z^\beta|\leq|z|^{s+1}\eta^{-s-1}\sum_\beta|c_\beta|\eta^{|\beta|} for |z|≀η|z|\leq\eta.

Choose Cβ‰₯1C\geq1 with Hβˆ‘k=0sβˆ’1Cβˆ’(sβˆ’k)≀|c|/4H\sum_{k=0}^{s-1}C^{-(s-k)}\leq|c|/4. Let 0<|w|≀min⁡(Ξ·,|c|/(4K))0<|w|\leq\min(\eta,|c|/(4K)) and |w|β‰₯C|zβ€²||w|\geq C|z'|. Then |z|=|w||z|=|w|, and

|g(zβ€²,w)|β‰₯|c||w|sβˆ’βˆ‘kH|zβ€²|sβˆ’k|w|kβˆ’K|w|s+1β‰₯|c||w|sβˆ’|c|4|w|sβˆ’|c|4|w|s>0. |g(z',w)|\geq|c||w|^s-\sum_kH|z'|^{s-k}|w|^k-K|w|^{s+1}\geq|c||w|^s-\tfrac{|c|}4|w|^s-\tfrac{|c|}4|w|^s>0 .

So in that neighbourhood every zero of gg other than the origin satisfies |w|<C|zβ€²||w|<C|z'|, and the origin satisfies |w|≀C|zβ€²||w|\leq C|z'|. Shrink the base polydisc of the proof of Theorem 1, using the bound for roots, so that for every zβ€²z' in it the points (zβ€²,bl(zβ€²))(z',b_l(z')) lie in that neighbourhood. They are zeros of gg, so |bl(zβ€²)|≀C|zβ€²||b_l(z')|\leq C|z'|. Since aka_k is, up to sign, the kk-th elementary symmetric function of the bl(zβ€²)b_l(z'), |ak(zβ€²)|≀(sk)Ck|zβ€²|k|a_k(z')|\leq\binom skC^k|z'|^k. β–«\square

Division, finite generation and elementary algebra

Proof of Theorem 2. The divided difference. For complex numbers b1,…,bsb_1,\ldots,b_s, ΞΆ\zeta and ww, telescoping through the products in which the first β„“\ell factors use ww and the others use ΞΆ\zeta gives

∏i=1s(ΞΆβˆ’bi)βˆ’βˆi=1s(wβˆ’bi)=(ΞΆβˆ’w)βˆ‘β„“=1s∏i<β„“(wβˆ’bi)∏i>β„“(ΞΆβˆ’bi).(5) \prod_{i=1}^s(\zeta-b_i)-\prod_{i=1}^s(w-b_i) =(\zeta-w)\sum_{\ell=1}^s\ \prod_{i<\ell}(w-b_i)\prod_{i>\ell}(\zeta-b_i). \qquad\text{(5)}

Apply this to the roots of P(zβ€²,β‹…)P(z',\cdot). The divided difference K(zβ€²;ΞΆ,w)=(P(zβ€²,ΞΆ)βˆ’P(zβ€²,w))/(ΞΆβˆ’w)K(z';\zeta,w)=\bigl(P(z',\zeta)-P(z',w)\bigr)/(\zeta-w) is the polynomial

K(zβ€²;ΞΆ,w)=βˆ‘j=0sβˆ’1wjKj(zβ€²,ΞΆ),Kj(zβ€²,ΞΆ)=βˆ‘k=0sβˆ’1βˆ’jak(zβ€²)ΞΆsβˆ’1βˆ’jβˆ’k, K(z';\zeta,w)=\sum_{j=0}^{s-1}w^jK_j(z',\zeta),\qquad K_j(z',\zeta)=\sum_{k=0}^{s-1-j}a_k(z')\zeta^{s-1-j-k},

where a0=1a_0=1. This follows from ΞΆdβˆ’wd=(ΞΆβˆ’w)βˆ‘i+j=dβˆ’1ΞΆiwj\zeta^d-w^d=(\zeta-w)\sum_{i+j=d-1}\zeta^iw^j. By (5), |K(zβ€²;ΞΆ,w)|≀s(Οƒ+Ξ±)sβˆ’1|K(z';\zeta,w)|\leq s(\sigma+\alpha)^{s-1} when |ΞΆ|=Οƒ|\zeta|=\sigma and |w|≀σ|w|\leq\sigma.

Existence on a smaller disc. Fix Οƒ\sigma with Ξ±<Οƒ<ρ\alpha<\sigma<\rho. For |ΞΆ|=Οƒ|\zeta|=\sigma every factor of P(zβ€²,ΞΆ)=∏l(ΞΆβˆ’bl(zβ€²))P(z',\zeta)=\prod_l(\zeta-b_l(z')) has modulus at least Οƒβˆ’Ξ±\sigma-\alpha, so |P(zβ€²,ΞΆ)|β‰₯(Οƒβˆ’Ξ±)s>0|P(z',\zeta)|\geq(\sigma-\alpha)^s>0. For zβ€²βˆˆDβ€²z'\in D' and |w|<Οƒ|w|<\sigma, define

qΟƒ(zβ€²,w)=12Ο€i∫|ΞΆ|=Οƒf(zβ€²,ΞΆ)P(zβ€²,ΞΆ)(ΞΆβˆ’w)dΞΆ,RΟƒ(zβ€²,w)=βˆ‘j=0sβˆ’1wj12Ο€i∫|ΞΆ|=Οƒf(zβ€²,ΞΆ)Kj(zβ€²,ΞΆ)P(zβ€²,ΞΆ)dΞΆ.(4) \begin{aligned} q_\sigma(z',w)&=\frac1{2\pi i}\int_{|\zeta|=\sigma}\frac{f(z',\zeta)}{P(z',\zeta)(\zeta-w)}\,d\zeta,\\ R_\sigma(z',w)&=\sum_{j=0}^{s-1}w^j\,\frac1{2\pi i}\int_{|\zeta|=\sigma}\frac{f(z',\zeta)K_j(z',\zeta)}{P(z',\zeta)}\,d\zeta . \end{aligned} \qquad\text{(4)}

Lemma A shows that qΟƒq_\sigma is holomorphic on D′×ΔσD'\times\Delta_\sigma and that the coefficients of RΟƒR_\sigma are holomorphic on Dβ€²D'. Cauchy’s formula for f(zβ€²,β‹…)f(z',\cdot) on the circle, together with

1ΞΆβˆ’w=P(zβ€²,w)P(zβ€²,ΞΆ)(ΞΆβˆ’w)+K(zβ€²;ΞΆ,w)P(zβ€²,ΞΆ), \frac1{\zeta-w}=\frac{P(z',w)}{P(z',\zeta)(\zeta-w)}+\frac{K(z';\zeta,w)}{P(z',\zeta)},

gives f=qΟƒP+RΟƒf=q_\sigma P+R_\sigma on D′×ΔσD'\times\Delta_\sigma.

Uniqueness. Let Ξ±<τ≀ρ\alpha<\tau\leq\rho, and suppose qP+R=0qP+R=0 on D′×ΔτD'\times\Delta_\tau, with qq holomorphic and RR a polynomial in ww of degree less than ss with holomorphic coefficients. Fix zβ€²z'. A root bb of multiplicity mm of P(zβ€²,β‹…)P(z',\cdot) lies in Δτ\Delta_\tau, and R(zβ€²,w)=βˆ’q(zβ€²,w)P(zβ€²,w)R(z',w)=-q(z',w)P(z',w) vanishes at bb to order at least mm. The polynomial R(zβ€²,β‹…)R(z',\cdot) thus has at least ss zeros counted with multiplicity. A nonzero polynomial of degree less than ss has at most sβˆ’1s-1, so R(zβ€²,β‹…)=0R(z',\cdot)=0. Then q(zβ€²,β‹…)P(zβ€²,β‹…)=0q(z',\cdot)P(z',\cdot)=0. Since P(zβ€²,β‹…)P(z',\cdot) has only finitely many zeros, continuity gives q(zβ€²,β‹…)=0q(z',\cdot)=0.

The whole domain. For Ξ±<Οƒ1<Οƒ2<ρ\alpha<\sigma_1<\sigma_2<\rho, uniqueness on D′×Δσ1D'\times\Delta_{\sigma_1} shows that qΟƒ2q_{\sigma_2} and RΟƒ2R_{\sigma_2} restrict to qΟƒ1q_{\sigma_1} and RΟƒ1R_{\sigma_1}. The functions therefore define qq and RR on all of D′×ΔρD'\times\Delta_\rho with f=qP+Rf=qP+R. This representation is unique by the uniqueness step with Ο„=ρ\tau=\rho.

Estimates. Let |f|≀M|f|\leq M. For |w|<Οƒ|w|<\sigma, the bounds on KK and 1/P1/P and the length 2πσ2\pi\sigma of the circle give

|R(zβ€²,w)|≀sΟƒ(Οƒ+Ξ±)sβˆ’1(Οƒβˆ’Ξ±)sM. |R(z',w)|\leq\frac{s\sigma(\sigma+\alpha)^{s-1}}{(\sigma-\alpha)^s}\,M .

The function RR does not depend on Οƒ\sigma; letting σ→ρ\sigma\to\rho gives |R|≀C1M|R|\leq C_1M on D′×ΔρD'\times\Delta_\rho. On the circle |w|=Οƒ|w|=\sigma we have |q|=|fβˆ’R|/|P|≀(1+C1)M/(Οƒβˆ’Ξ±)s|q|=|f-R|/|P|\leq(1+C_1)M/(\sigma-\alpha)^s. The maximum modulus principle for q(zβ€²,β‹…)q(z',\cdot) on the closed disc |w|≀σ|w|\leq\sigma extends this bound to the disc. Letting σ→ρ\sigma\to\rho gives |q|≀C2M|q|\leq C_2M. β–«\square

The constants (6) depend only on the degree ss and the radii α\alpha and ρ\rho. In particular, one fixed polydisc serves for the division of every bounded holomorphic function.

Proof of Corollary 3. If s=0s=0, then gg is a unit and f=(f/g)gf=(f/g)g; a remainder of negative degree is zero. Let sβ‰₯1s\geq1. By the proof of Theorem 1, g=uPg=uP on a product Δδ′×Δσ\Delta'_\delta\times\Delta_\sigma, with uu zero-free. Given ff, choose Ο„βˆˆ(0,Οƒ]\tau\in(0,\sigma] and Ξ΄β€²βˆˆ(0,Ξ΄]\delta'\in(0,\delta] such that ff is holomorphic on Δδ′′×Δτ\Delta'_{\delta'}\times\Delta_\tau. Then choose α∈(0,Ο„)\alpha\in(0,\tau) and Ξ΄β€³βˆˆ(0,Ξ΄β€²]\delta''\in(0,\delta'] with βˆ‘k|ak(zβ€²)|Ξ±βˆ’k<1\sum_k|a_k(z')|\alpha^{-k}<1 on Δδ″′\Delta'_{\delta''}. By the bound for roots, all roots of P(zβ€²,β‹…)P(z',\cdot) then lie in Δα\Delta_\alpha. Theorem 2 on Δδ″′×Δτ\Delta'_{\delta''}\times\Delta_\tau gives f=qP+R=(q/u)g+Rf=qP+R=(q/u)g+R.

If q1g+R1=q2g+R2q_1g+R_1=q_2g+R_2 as germs, then (q1βˆ’q2)uP+(R1βˆ’R2)=0(q_1-q_2)uP+(R_1-R_2)=0 on a product neighbourhood. By the bound for roots, that neighbourhood contains a product Δδ2′×Δτ2\Delta'_{\delta_2}\times\Delta_{\tau_2} on which all roots of P(zβ€²,β‹…)P(z',\cdot) have modulus at most some Ξ±2<Ο„2\alpha_2<\tau_2. The uniqueness step of Theorem 2 gives R1=R2R_1=R_2 and (q1βˆ’q2)u=0(q_1-q_2)u=0, so q1=q2q_1=q_2.

By uniqueness, the remainder map f↦Rf\mapsto R is π’ͺnβˆ’1\mathcal O_{n-1}-linear. Its kernel is (g)(g). It restricts to the identity on polynomials of degree less than ss. Hence the classes of 1,w,…,wsβˆ’11,w,\ldots,w^{s-1} generate π’ͺn/(g)\mathcal O_n/(g). They are independent: if βˆ‘jcjwj=qg\sum_jc_jw^j=qg, uniqueness of the remainder of this polynomial forces every cj=0c_j=0. β–«\square

Proof of Theorem 6. We use the following module fact. If a commutative ring AA is Noetherian, every submodule NβŠ‚AsN\subset A^s is finitely generated. For s=1s=1 this is the definition. For s>1s>1, the last coordinates of the elements of NN form an ideal of AA. Choose finitely many elements of NN whose last coordinates generate it. Every element of NN differs from an AA-combination of these by an element of N∩(Asβˆ’1Γ—0)N\cap(A^{s-1}\times0), which is finitely generated by induction.

Now induct on nn. The ring π’ͺ0=β„‚\mathcal O_0=\mathbb C is a field. Let nβ‰₯1n\geq1 and JβŠ‚π’ͺnJ\subset\mathcal O_n a nonzero ideal. A linear change of coordinates is a ring automorphism of π’ͺn\mathcal O_n. By Proposition 4 we may therefore assume that some f∈Jf\in J is regular of order ss in ww. If s=0s=0, then ff is a unit and J=π’ͺnJ=\mathcal O_n. Otherwise Corollary 3 writes each h∈Jh\in J as h=qf+Rhh=qf+R_h, where Rh=hβˆ’qf∈JR_h=h-qf\in J lies in the free module ⨁j<sπ’ͺnβˆ’1wj\bigoplus_{j<s}\mathcal O_{n-1}w^j. The remainders RhR_h, h∈Jh\in J, form an π’ͺnβˆ’1\mathcal O_{n-1}-submodule of this free module. By the induction hypothesis and the module fact, they are generated by finitely many R1,…,Rp∈JR_1,\ldots,R_p\in J. Then JJ is generated by f,R1,…,Rpf,R_1,\ldots,R_p. β–«\square

Exercises and complete solutions

1. Multiplicity in division

Difficulty: Introductory.

For P(w)=w3P(w)=w^3, divide f(w)=ewf(w)=e^w. Explain why checking distinct zeros alone does not prove uniqueness.

Solution. The remainder is R=1+w+w2/2R=1+w+w^2/2 and q=(ewβˆ’1βˆ’wβˆ’w2/2)/w3q=(e^w-1-w-w^2/2)/w^3, extended holomorphically at zero with value 1/61/6. A polynomial such as ww vanishes at the only distinct zero, but is not divisible holomorphically by w3w^3. Uniqueness requires the three zeros counted with multiplicity, or equality of the first three Taylor coefficients.

2. A prepared exponential

Difficulty: Introductory.

On β„‚2\mathbb C^2 with coordinates (z,w)(z,w), let g=ewβˆ’1βˆ’zg=e^w-1-z. Show that gg is regular of order one in ww, and find its Weierstrass polynomial and unit.

Solution. g(0,w)=ewβˆ’1=w(1+w/2+β‹―)g(0,w)=e^w-1=w(1+w/2+\cdots), so the order is one. For |z|<1|z|<1 let L(z)=log⁡(1+z)L(z)=\log(1+z) be the principal logarithm, so L(0)=0L(0)=0 and eL(z)=1+ze^{L(z)}=1+z. Let E(t)=(etβˆ’1)/tE(t)=(e^t-1)/t for tβ‰ 0t\neq0 and E(0)=1E(0)=1; this is an entire function. Then

ewβˆ’1βˆ’z=eL(ewβˆ’Lβˆ’1)=(1+z)E(wβˆ’L(z))(wβˆ’L(z)). e^w-1-z=e^{L}\bigl(e^{w-L}-1\bigr)=(1+z)\,E\bigl(w-L(z)\bigr)\,\bigl(w-L(z)\bigr).

Thus P=wβˆ’L(z)P=w-L(z), with coefficient βˆ’L(z)-L(z) vanishing at z=0z=0, and u=(1+z)E(wβˆ’L(z))u=(1+z)E(w-L(z)), with u(0,0)=1u(0,0)=1. Uniqueness in Theorem 1 shows these are the factors that its proof constructs.

3. Coefficients without holomorphic roots

Difficulty: Intermediate.

  1. Let g=w2βˆ’z3g=w^2-z^3. Find its Weierstrass polynomial, show that no holomorphic function b(z)b(z) near 00 satisfies b(z)2=z3b(z)^2=z^3, and check Proposition 5. (b) Let g=w2βˆ’zg=w^2-z. Show that gg is regular of order two in ww, but that the coefficient estimate of Proposition 5 fails. Which hypothesis is missing?

Solution. (a) gg is already a Weierstrass polynomial of degree two, so P=gP=g and u=1u=1. If bb were holomorphic with b2=z3b^2=z^3, the order of vanishing of b2b^2 at 00 would be even, while z3z^3 vanishes to order three. The coefficients are a1=0a_1=0 and a2=βˆ’z3a_2=-z^3, so |a2|≀|z|2|a_2|\leq|z|^2 for |z|≀1|z|\leq1, as Proposition 5 predicts: the order of gg at zero is 22, equal to its order in ww. Power sums give the same coefficients: S1=b1+b2=0S_1=b_1+b_2=0 and S2=b12+b22=2z3S_2=b_1^2+b_2^2=2z^3, so (C) gives a1=βˆ’S1=0a_1=-S_1=0 and a2=βˆ’(S1a1+S2)/2=βˆ’z3a_2=-(S_1a_1+S_2)/2=-z^3.

  1. g(0,w)=w2g(0,w)=w^2, so gg is regular of order two in ww, and P=gP=g. Here a2=βˆ’za_2=-z, which is not bounded by a constant times |z|2|z|^2 near 00. The order of gg at zero is one, because of the linear term βˆ’z-z, so the hypothesis that the order in ww equals the order at zero fails. The roots Β±z1/2\pm z^{1/2} satisfy |b|=|z|1/2|b|=|z|^{1/2}, which is not bounded by a constant times |z||z|.

4. Explicit division constants

Difficulty: Introductory.

Let P=w2βˆ’zP=w^2-z on Dβ€²={|z|<1/4}D'=\{|z|<1/4\}. Take Ξ±=1/2\alpha=1/2 and ρ=1\rho=1. Verify the hypothesis of Theorem 2, divide f=w3f=w^3, and compare with (6).

Solution. The roots Β±z1/2\pm z^{1/2} have modulus |z|1/2<1/2|z|^{1/2}<1/2. Since w3=w(w2βˆ’z)+zww^3=w(w^2-z)+zw, we get q=wq=w and R=zwR=zw; by uniqueness these are the quotient and remainder of Theorem 2. On Dβ€²Γ—Ξ”1D'\times\Delta_1 we have |f|≀M=1|f|\leq M=1, |R|≀1/4|R|\leq1/4 and |q|≀1|q|\leq1. Formula (6) gives C1=2β‹…1β‹…(3/2)/(1/2)2=12C_1=2\cdot1\cdot(3/2)/(1/2)^2=12 and C2=13/(1/2)2=52C_2=13/(1/2)^2=52. The bounds hold with room to spare: the constants are uniform in ff, not sharp.

An algebraic proof of the coefficient recursion

This proof was written by GPT-6 Astra (OpenAI), October 2026, for this course. It gives (C) again, in the notation Aj=(βˆ’1)jejA_j=(-1)^je_j of (N6), without power series in 1/w1/w.

A finite identity that does not distinguish repeated roots

For any list Ξ»1,…,Ξ»kβˆˆβ„‚\lambda_1,\ldots,\lambda_k\in\mathbb C, repetitions allowed, define

Sβ„“=βˆ‘i=1kΞ»iβ„“(β„“β‰₯1),e0=1,ej=βˆ‘1≀i1<β‹―<ij≀kΞ»i1β‹―Ξ»ij(1≀j≀k).(N1) S_\ell=\sum_{i=1}^k\lambda_i^\ell\quad(\ell\ge1),\qquad e_0=1,\qquad e_j=\sum_{1\le i_1<\cdots<i_j\le k} \lambda_{i_1}\cdots\lambda_{i_j}\quad(1\le j\le k). \qquad\text{(N1)}

The indices in the definition of eje_j select occurrences in the list, even when their values coincide. Set ej=0e_j=0 for j>kj>k. The finite Newton identities are

jej=βˆ‘β„“=1j(βˆ’1)β„“βˆ’1ejβˆ’β„“Sβ„“,1≀j≀k.(N2) j e_j=\sum_{\ell=1}^j(-1)^{\ell-1}e_{j-\ell}S_\ell, \qquad 1\le j\le k. \qquad\text{(N2)}

Proof. Work first in the polynomial ring R=β„‚[Ξ»1,…,Ξ»k]R=\mathbb C[\lambda_1,\ldots,\lambda_k], treating the Ξ»i\lambda_i as indeterminates. Introduce a formal variable XX and the finite polynomial

E(X)=∏i=1k(1+Ξ»iX)=βˆ‘j=0kejXj.(N3) E(X)=\prod_{i=1}^k(1+\lambda_iX) =\sum_{j=0}^k e_jX^j. \qquad\text{(N3)}

Each factor has constant coefficient one, so it is invertible in R[[X]]R[[X]], with formal inverse βˆ‘mβ‰₯0(βˆ’Ξ»iX)m\sum_{m\ge0}(-\lambda_iX)^m. Formal differentiation of the finite product gives

Eβ€²(X)=βˆ‘i=1kΞ»i∏hβ‰ i(1+Ξ»hX)=E(X)βˆ‘i=1kΞ»i1+Ξ»iX=E(X)βˆ‘β„“β‰₯1(βˆ’1)β„“βˆ’1Sβ„“Xβ„“βˆ’1.(N4) \begin{aligned} E'(X) &=\sum_{i=1}^k\lambda_i\prod_{h\ne i}(1+\lambda_hX)\\ &=E(X)\sum_{i=1}^k\frac{\lambda_i}{1+\lambda_iX}\\ &=E(X)\sum_{\ell\ge1}(-1)^{\ell-1}S_\ell X^{\ell-1}. \end{aligned} \qquad\text{(N4)}

Equality of the coefficients of Xjβˆ’1X^{j-1} is exactly (N2). There is no analytic convergence or division by a root. To make the argument entirely finite, truncate each geometric inverse after degree kβˆ’1k-1 and work modulo XkX^k; its product with 1+Ξ»iX1+\lambda_iX is one modulo XkX^k. This determines all the coefficients used in (N2). Finally specialize the indeterminates to the given complex numbers. Polynomial identities remain valid when values coincide or vanish. β–«\square

Since division by the positive integer jj is allowed in β„‚\mathbb C, (N2) recursively expresses every eje_j, 1≀j≀k1\le j\le k, as a polynomial with rational coefficients in S1,…,SjS_1,\ldots,S_j. For indices not exceeding kk,

e1=S1,e2=S12βˆ’S22,e3=S13βˆ’3S1S2+2S36.(N5) e_1=S_1,\qquad e_2=\frac{S_1^2-S_2}{2},\qquad e_3=\frac{S_1^3-3S_1S_2+2S_3}{6}. \qquad\text{(N5)}

For the descending coefficients of

P(w)=∏i=1k(wβˆ’Ξ»i)=wk+A1wkβˆ’1+β‹―+Ak,(N6) P(w)=\prod_{i=1}^k(w-\lambda_i) =w^k+A_1w^{k-1}+\cdots+A_k, \qquad\text{(N6)}

we have A0=1A_0=1 and Aj=(βˆ’1)jejA_j=(-1)^je_j. Thus the equivalent coefficient recursion is

jAj+βˆ‘β„“=1jAjβˆ’β„“Sβ„“=0,1≀j≀k.(N7) jA_j+\sum_{\ell=1}^j A_{j-\ell}S_\ell=0, \qquad 1\le j\le k. \qquad\text{(N7)}

For example, where the indices do not exceed kk, A1=βˆ’S1A_1=-S_1, A2=(S12βˆ’S2)/2A_2=(S_1^2-S_2)/2, and A3=(βˆ’S13+3S1S2βˆ’2S3)/6A_3=(-S_1^3+3S_1S_2-2S_3)/6. These signs correspond to the factors wβˆ’Ξ»iw-\lambda_i.

Further reading

Jean-Pierre Demailly, Complex Analytic and Differential Geometry, version of 21 June 2012, Chapter II, Β§2.A, treats preparation and division by C. L. Siegel’s contour method.