diff a1e72ff..29bb8d1 ambient_homeo_mult_2_3_v2.txt
diff --git a/ambient_homeo_mult_2_3_v2.txt b/ambient_homeo_mult_2_3_v2.txt
index e69de29..51feb75 100644
--- a/ambient_homeo_mult_2_3_v2.txt
+++ b/ambient_homeo_mult_2_3_v2.txt
@@ -0,0 +1,2741 @@
+AMBIENTLY HOMEOMORPHIC ISOLATED HYPERSURFACES OF MULTIPLICITIES TWO AND THREE
+=============================================================================
+
+A standalone ASCII explainer (for 16-year-olds) of the OpenAI paper
+of the same title, dated 24 September 2026.
+
+One-sentence version: in a high enough complex dimension there are two
+hypersurface germs with an isolated singularity each, one visibly
+starting with a degree-2 term and the other visibly starting with a
+degree-3 term, whose zero sets are indistinguishable by any continuous
+change of coordinates near the origin. This disproves the embedded
+Zariski multiplicity conjecture and three other open problems.
+
+The source is the LaTeX article in this directory, concatenated into
+ambient_homeo_mult_2_3_original.txt (sections: introduction, background,
+seeds, local, global, conclusion, appendix/certificate). All theorems,
+lemmas, propositions, corollaries, and the finite integer certificate
+from the appendix are transcribed. Each exercise ends with a line
+"Original: ..." identifying the theorem, lemma, or section it covers,
+for cross-checking against the source.
+
+Conventions used throughout this explainer:
+  C, R, Q, Z denote the complex numbers, real numbers, rationals,
+  integers.
+  C^N means N-dimensional complex space with coordinates z_1,...,z_N.
+  (C^N, 0) means a neighbourhood of the origin in C^N, and a "germ" is
+  an object defined only on an arbitrarily small such neighbourhood.
+  ord_0(f) = "multiplicity" of a holomorphic f at the origin
+  = the smallest total degree of a monomial actually occurring in the
+  power-series expansion of f with nonzero coefficient.
+  V(f) = f^(-1)(0) is the zero set.
+  Hess_0(f) is the complex Hessian matrix (d^2 f / dz_i dz_j)(0).
+  f = g o phi means the composition (g o phi)(z) = g(phi(z)).
+  For a finite-rank free abelian group L with a bilinear form S and an
+  operator M, "(L, M, S)" bundles the three pieces of data. An
+  "equivariant isometry" between two such triples is an isomorphism
+  intertwining M and pulling S back.
+
+==============================================================
+SECTION A: WHAT THE PAPER PROVES
+==============================================================
+
+Exercise: Multiplicity of a hypersurface germ. What is the
+"multiplicity" of a hypersurface singularity at the origin, and
+why does it not change under a holomorphic change of coordinates?
+
+Let f : (C^N, 0) -> (C, 0) be a holomorphic germ. Its Taylor expansion
+at the origin is a power series in z_1,...,z_N with no constant term.
+The order ord_0(f) is the smallest total degree d for which some mono-
+mial z_1^(a_1) ... z_N^(a_N) with a_1+...+a_N = d occurs with nonzero
+coefficient. When additionally f is reduced (is not the square, cube,
+or higher power of another holomorphic function) and has an isolated
+critical point (grad f vanishes only at 0), the integer ord_0(f) is
+exactly the algebraic multiplicity of V(f) at 0.
+
+If phi : (C^N, 0) -> (C^N, 0) is a holomorphic (or real-analytic)
+diffeomorphism with phi(0) = 0, and g = f o phi, then ord_0(g) =
+ord_0(f). Reason: phi's Jacobian at 0 is invertible, so phi maps the
+ideal of functions vanishing to order d at 0 exactly onto itself. Con-
+cretely, z_1^2 + ... + z_N^2 has order 2 in every smooth coordinate
+system, and z_1^3 + ... + z_N^3 has order 3 in every smooth coordinate
+system.
+
+Original: Section 1, definition of ord_0 and multiplicity.
+
+Exercise: Ambient homeomorphism of pairs. What is the precise
+equivalence relation weakening smooth coordinate changes that Zariski
+asked about?
+
+An ambient homeomorphism of hypersurface germs V_1 = f_1^(-1)(0) and
+V_2 = f_2^(-1)(0) is a homeomorphism
+
+    H : (U_1, 0) -> (U_2, 0)
+
+between small open neighbourhoods U_j of the origin in C^N such that
+H(0) = 0 and H(V_1 intersect U_1) = V_2 intersect U_2.
+
+Crucial weakening vs. a coordinate change: H is only required to be a
+homeomorphism (continuous, bijective, continuous inverse). It is NOT
+required to be differentiable, holomorphic, real-analytic, bi-
+Lipschitz, or anything else regular. Because of this the Jacobian
+argument above fails: H need not preserve vanishing orders of
+functions, and may in principle "straighten out" a quadratic cone
+into a cubic one.
+
+Original: Section 1, paragraph defining "ambiently homeomorphic".
+
+Exercise: Zariski's multiplicity question. What did Zariski ask, what
+was known, and what does this paper prove?
+
+Zariski (1971) asked: if two reduced hypersurface germs V_1, V_2 in
+C^N are ambiently homeomorphic, must ord_0(f_1) = ord_0(f_2)?
+
+Positive results before this paper:
+  - Zariski (1932): for irreducible plane curves (N = 2), the
+    embedded topology of the branch at 0 determines its
+    characteristic data, in particular its multiplicity.
+    Summing over branches settles every reduced plane curve, so
+    the answer is YES for N = 2.
+  - A'Campo (1973) and Le (1973): if a reduced hypersurface germ is
+    ambiently homeomorphic to a smooth one, it is actually smooth.
+    So multiplicity one cannot appear in a bad pair, in any ambient
+    dimension.
+  - Yau (1988): isolated quasihomogeneous surface singularities in
+    C^3 have multiplicity determined by embedded topology (with Xu,
+    added in proof).
+  - Fernandes-Jelonek-Sampaio (2025): multiplicity two is preserved
+    by ambient bi-Lipschitz equivalence (a much stronger notion
+    than ambient homeomorphism), and Hessian corank is preserved by
+    ambient homeomorphism in three complex variables.
+
+Theorem 1.1 of the paper (main theorem). There is a finite integer
+N > 3, divisible by eight, and there are reduced polynomials
+f_1, f_2 in R[z_1,...,z_N], weighted homogeneous for positive rational
+weights, such that:
+  (i)   each f_j has an isolated critical point at the origin;
+  (ii)  ord_0(f_1) = 2 and ord_0(f_2) = 3;
+  (iii) there is an ambient homeomorphism of pairs
+            (C^N, V(f_1), 0) is isomorphic to (C^N, V(f_2), 0).
+
+So the answer to Zariski's question is NO in high enough dimension.
+The construction determines one such N by exact integer arithmetic; it
+does not try to minimise N.
+
+Original: Section 1, Zariski 1971/1932, A'Campo 1973, Le 1973,
+Yau 1988, Fernandes-Jelonek-Sampaio 2025; Theorem 1.1.
+
+Exercise: Right-topological equivalence. What is it, and what two
+conjectures does the main theorem also refute?
+
+Let f, g : (C^N, 0) -> (C, 0) be holomorphic germs. They are
+topologically right equivalent if there is a homeomorphism germ
+phi : (C^N, 0) -> (C^N, 0) with
+
+    f = g o phi.
+
+This is strictly stronger than ambient homeomorphism of pairs: if
+f = g o phi, then phi carries V(f) to V(g). But the equation f =
+g o phi also says phi preserves the value of the holomorphic function
+at every point; the function is not just its zero set.
+
+The right-topological multiplicity conjecture (Sampaio 2025) asks
+whether reduced topologically right equivalent germs must have equal
+multiplicity. Its mod-2 version asks only whether equality holds mod 2.
+
+Corollary of the main theorem. For the SAME f_1, f_2 and SAME N,
+there is a homeomorphism germ phi : (C^N, 0) -> (C^N, 0) with
+f_1 = f_2 o phi. Nevertheless ord_0(f_1) = 2 and ord_0(f_2) = 3 are
+unequal in Z and unequal mod 2. So neither the multiplicity conjecture
+nor its mod-2 version survives.
+
+Note. The homeomorphism phi furnished by the corollary need NOT
+coincide with the ambient homeomorphism H from Theorem 1.1. The
+statement is only f_1 = f_2 o phi.
+
+Original: Section 1, Sampaio 2025 Milnor Conjectures 2 and "Z mod 2";
+Section 6, Corollary 6.4.
+
+Exercise: Arnold's corank problem. What is Hessian corank, and what
+did Arnold ask, and what does the main theorem answer?
+
+Let f : (C^N, 0) -> (C, 0) be holomorphic with a critical point at the
+origin. Its Hessian at 0 is the N x N complex matrix Hess_0(f) =
+(d^2 f / dz_i dz_j)(0). Its rank is the number of nonzero eigenvalues
+over C. The corank is
+
+    corank_0(f) = N - rank_C Hess_0(f).
+
+A Morse nondegenerate critical point has corank 0; a cone z_1^2 + ...
++ z_k^2 (with N > k) has corank N - k.
+
+Arnold's corank problem (Sampaio 2026, Problem 1). Suppose phi : (C^N,
+0) -> (C^N, 0) is a homeomorphism germ carrying V(f) to V(g), for
+reduced holomorphic f, g with critical points at 0. Must
+corank_0(f) = corank_0(g)?
+
+Corollary of the main theorem. For the SAME f_1, f_2 in Theorem 1.1,
+the two germs have DIFFERENT Hessian coranks, even though their zero
+sets are ambiently homeomorphic. So the answer to Arnold's question is
+also NO.
+
+Concretely: the quadratic part of f_1 turns out to be a sum of 16E
+independent squares Q = sum_{k=1}^{16E} u_k^2 (with E > 0 from the
+integer certificate), so rank_C Hess_0(f_1) = 16E and corank_0(f_1)
+= N - 16E < N. The polynomial f_2 has order 3, so its Hessian
+vanishes entirely: corank_0(f_2) = N. Both coranks are unequal but
+both Hessian ranks 16E and 0 are EVEN, so a weaker "corank-parity"
+version of Arnold's problem is not contradicted (and indeed a corank-
+parity theorem of Sampaio 2026 states it holds in all dimensions).
+
+Original: Section 1, Arnold (Sampaio 2026 Problem 1); Section 6,
+Corollary 6.3.
+
+Exercise: Initial-form Milnor fiber (Budur-Fernandez de Bobadilla-
+Le-Nguyen). What is it, and what is refuted?
+
+For a nonzero holomorphic germ g : (C^n, 0) -> (C, 0), let g_* be its
+lowest-degree nonzero homogeneous part (ordinary total degree, NOT a
+weighted grading). So g_* is a homogeneous polynomial of degree
+ord_0(g). Its unit fiber is the complex manifold
+
+    g_*^(-1)(1).
+
+Budur, Fernandez de Bobadilla, Le, Nguyen (2022) call the homotopy
+type of g_*^(-1)(1) the initial-form Milnor fiber of g. Motivated by
+contact-locus / monodromy theory, they conjectured (Conjecture 1.7 of
+arXiv v3):
+
+    If two reduced holomorphic germs f, g : (C^n, 0) -> (C, 0) are
+    topologically right equivalent (as in the third exercise above),
+    then their initial-form Milnor fibers f_*^(-1)(1) and
+    g_*^(-1)(1) are homotopy equivalent.
+
+The full nonreduced lowest part g_* is retained: e.g. the germ z_2^2
+has lowest part z_2^2 (not the reduced z_2), even though z_2^2 and z_2
+have the same zero set.
+
+Corollary of the main theorem. Take f_1, f_2 from Theorem 1.1 and add
+the SAME square in one new variable t:
+
+    f~_j(z, t) = f_j(z) + t^2,    j = 1, 2,    in C^(N+1).
+
+The two germs f~_1, f~_2 are topologically right equivalent (multiply
+the homeomorphism phi by identity in t). Each has an isolated critical
+point at the origin. But
+
+    H_0( (f~_1)_*^(-1)(1); Z ) is isomorphic to Z     (one component),
+    H_0( (f~_2)_*^(-1)(1); Z ) is isomorphic to Z^2   (two components).
+
+So the initial-form Milnor fibers are not even homotopy equivalent, and
+the Budur et al. homotopy conjecture fails in degree zero.
+
+Why the counts. The lowest parts are
+    (f~_1)_* = Q + t^2        with Q = sum_{k=1}^{16E} u_k^2
+    (f~_2)_* = t^2.
+Their unit fibers:
+    (f~_1)_*^(-1)(1)  =  {v in C^(16E+1) : sum v_k^2 = 1} x C^(N-16E).
+        Writing v = x + i y with x, y in R^(16E+1), the quadric
+        equation becomes |x|^2 - |y|^2 = 1 with x . y = 0. The map
+        (u, y) -> sqrt(1+|y|^2) u + i y over u in S^(16E) and y
+        perpendicular u identifies it with the tangent bundle of the
+        connected sphere S^(16E); hence it is connected.
+    (f~_2)_*^(-1)(1)  =  C^N x {t in C : t^2 = 1} = C^N x {+1, -1}.
+        Two components.
+
+Sampaio 2025 (Conjecture 4) and Budur et al. 2022 (Conjecture 1.7
+arXiv v3) are the two formulations refuted.
+
+Original: Section 1 paragraphs on Budur et al.; Section 6, Corollary
+6.5 (initial-form Milnor-fibre counterexamples).
+
+Exercise: Family constructions vs. arbitrary equivalent pairs. Why is
+the main theorem not in contradiction with Le-Ramanujam (1976) and
+equimultiplicity theorems?
+
+Le-Ramanujam (1976): if a one-parameter holomorphic deformation f_t of
+isolated hypersurface singularities has constant Milnor number mu(f_t)
+and ambient complex dimension != 3, its topological type is constant.
+Greuel (1986) and O'Shea (1987): constant Milnor number also implies
+constant multiplicity for quasihomogeneous deformations. Fernandez de
+Bobadilla and Pelka (2022) established the general equimultiplicity
+theorem for continuous mu-constant families.
+
+These results all sit inside a DEFORMATION or FAMILY: they require a
+continuous family f_t joining the two singularities, plus a strong
+hypothesis (constant Milnor number).
+
+The pair f_1, f_2 in Theorem 1.1 is NOT connected by a mu-constant
+family, because
+    - their Hessian ranks differ (16E vs. 0, but in fact as coranks
+      they differ), and
+    - Milnor number changes under even the mildest "quadratic to
+      cubic" change.
+So the family theorems never apply, and there is no contradiction: the
+main theorem is specifically about ARBITRARY ambiently-equivalent
+pairs, not about pairs seen along a deformation.
+
+Fernandez-de Bobadilla and Pelka's Example 8.6 already exhibits
+topologically equivalent germs in C^4 (both multiplicity two) that
+cannot be joined by a mu-constant family, and these are a precursor of
+the construction here (adding two independent squares to plane-curve
+germs with the same integral Seifert form).
+
+Metric hypotheses also fall outside the theorem: the main ambient
+homeomorphism is topological, not bi-Lipschitz, and the ambient dimen-
+sion is larger than three, so the Fernandes-Jelonek-Sampaio low-
+dimensional theorems do not apply.
+
+Original: Section 1, "Families, metric equivalence" subsection;
+Le-Ramanujam 1976, Greuel 1986, O'Shea 1987, Bobadilla-Pelka 2022,
+Fernandes-Jelonek-Sampaio 2025.
+
+==============================================================
+SECTION B: THE ALGEBRAIC MACHINERY
+==============================================================
+
+Exercise: Milnor fibration. What does Milnor's theorem associate to
+an isolated singularity, and what is the Milnor fiber?
+
+(Milnor 1968). Let s >= 2 and let f : (C^s, 0) -> (C, 0) be a
+holomorphic germ with an isolated critical point at 0. Choose a small
+sphere S^(2s-1)_eps around the origin whose interior contains no
+other critical point. Then the link
+
+    K_f = f^(-1)(0) intersect S^(2s-1)_eps
+
+is a smooth closed manifold of complex dimension s-2 (real dimension
+2s-3), and the map
+
+    f / |f| : S^(2s-1)_eps \ K_f  ->  S^1
+
+is a locally trivial fibration (the Milnor fibration). Its fibre is
+the interior of a compact manifold F_f with boundary K_f, called the
+Milnor fiber. F_f is homotopy equivalent to a wedge of (s-1)-spheres;
+concretely
+
+    H_j(F_f; Z) = 0 for j != 0, s-1,
+    H_{s-1}(F_f; Z) is a finite-rank free abelian group.
+
+The link K_f is (s-3)-connected. All of this is independent of the
+particular choice of eps small enough.
+
+Original: Section 2, Theorem (Milnor); Milnor 1968 Theorems 4.8, 5.2,
+6.5, Lemma 6.1.
+
+Exercise: Milnor lattice, monodromy operator, and Seifert form. What
+three algebraic objects does the paper attach to an isolated
+singularity?
+
+From the Milnor fibration of the previous exercise (with s >= 2):
+
+(1) Milnor lattice L_f:
+    L_f = reduced H_{s-1}(F_f; Z),
+    a finite-rank free abelian group. Its rank is the (middle) Milnor
+    number mu(f).
+
+(2) Monodromy operator M_f: the geometric monodromy h : F_f -> F_f is
+    obtained by traversing once around the base S^1. Its induced map
+    on middle homology is a linear automorphism
+        M_f : L_f -> L_f.
+    The paper uses a specific convention (see (2.1) below). For
+    isolated weighted homogeneous singularities (the ones in the main
+    theorem), M_f has FINITE ORDER.
+
+(3) Seifert form S_f:
+        S_f(x, y) = lk(x, y^+),
+    where lk is the linking number in S^(2s-1)_eps, and y^+ is the
+    push-off of y in the positive page direction (a neighboring
+    Milnor fiber). S_f is a unimodular, integral, bilinear form.
+
+Consistency / convention. The paper uses COLUMN vectors and fixes the
+monodromy via
+
+    M_f = (-1)^s S_f^(-1) S_f^t,     (2.1)
+
+where (-1)^t on a matrix means transpose. This inverts the more
+standard convention S_f(M x, y) = (-1)^s S_f(y, x); inverting
+monodromy throughout preserves every comparison made in the paper
+(inverting an operator preserves cyclicity and its characteristic
+polynomial's roots).
+
+Isomorphisms of Seifert forms mean integral congruence (equivalence
+over Z) unless a different coefficient field or ring is specified.
+
+Original: Section 2, paragraphs after Theorem (Milnor);
+Balnojan-Hertling 2019, Section 6, Equations 6.1-6.2.
+
+Exercise: Thom-Sebastiani-Sakamoto. What does adding two isolated
+singularities in disjoint variables do to their Milnor data?
+
+Theorem (Thom-Sebastiani-Sakamoto). Let f and g be isolated
+singularity germs in a and b variables respectively, and let f + g be
+their sum in a + b variables (disjoint variable sets). Then there is
+an isomorphism of integral lattices
+
+    L_(f+g)  is isomorphic to  L_f (tensor over Z) L_g,
+
+under which
+
+    M_(f+g)  =  M_f (tensor) M_g,
+    S_(f+g)  =  (-1)^(a b) S_f (tensor) S_g.
+
+The statement includes the case f = x_1^p (a one-variable power);
+its Milnor lattice is H_0 of the fiber, appropriately reduced.
+
+References. Sebastiani-Thom 1971 (monodromy); Sakamoto 1974 Theorem 2,
+p. 715 (integral Seifert form); Balnojan-Hertling 2019, Section 6,
+Equations 6.3-6.5 (lattice and monodromy). The tensor sign (-1)^(ab)
+depends only on the ambient dimensions a and b. So when comparing sums
+with the SAME factor dimensions on both sides, the tensor signs are
+the SAME and therefore cancel.
+
+Original: Section 2, Theorem (Thom-Sebastiani-Sakamoto);
+Sebastiani-Thom 1971; Sakamoto 1974; Balnojan-Hertling 2019.
+
+Exercise: Weighted homogeneous singularity. What is it, and what is
+its spectral polynomial?
+
+A polynomial f in s variables is weighted homogeneous with positive
+rational weights w_1,...,w_s if
+
+    f(lambda^(w_1) z_1, ..., lambda^(w_s) z_s) = lambda^d f(z)
+
+for some positive degree d, understood after clearing denominators (so
+that w_j and d are integers when necessary). It is convenient to
+rescale so that d = 1; then the weights w_1,...,w_s are the
+"normalised" weights. Clearing denominators gives a weighted quasi-
+homogeneous polynomial in the classical sense.
+
+Its spectrum (Steenbrink) is a finite multiset of rational "spectral
+exponents" beta, each in the open interval (0, s), counted with
+multiplicity. The paper uses the convention: positive spectral
+exponents, shifted by 1 from the classical convention Sp in
+(-1, s-1). The shift by 1 does not change any congruence.
+
+Spectral polynomial:
+
+    P_f(t)  =  sum over beta of  t^beta
+
+            =  prod_{j=1}^{s}  ( t^(w_j) - t ) / ( 1 - t^(w_j) ).   (2.2)
+
+Why (2.2)? The quotient ring R = C[z_1,...,z_s] / (partial_1 f, ...,
+partial_s f) is finite-dimensional. The derivatives partial_j f are
+weighted homogeneous of degree 1 - w_j (since partial_j of a monomial
+z^a of degree 1 has degree 1 - w_j). They form a REGULAR SEQUENCE
+(their common zero set is the origin only, so the ideal has height s
+in the Cohen-Macaulay polynomial ring). Successive multiplication
+exact sequences therefore give the Hilbert series
+
+    Hilb(R, t)  =  prod_j ( 1 - t^(1 - w_j) ) / ( 1 - t^(w_j) ).   (2.3)
+
+Multiplying (2.3) by t^(sum_j w_j) yields (2.2). Steenbrink's monomial
+formula assigns the exponent (k_1+1) w_1 + ... + (k_s+1) w_s to a
+monomial z^k in a homogeneous basis of R; the weighted Hilbert series
+converts the multiset of these exponents directly to (2.2).
+
+Semisimplicity. The weighted coordinate rotation z_j -> exp(2 pi i
+w_j) z_j defines geometric monodromy. A common multiple of the weight
+denominators (i.e. of the denominators of the w_j) kills this rotation
+(its D-th power is identity). So M_f has finite order, hence is
+semisimple over C (finite-order operators in characteristic zero are
+diagonalisable). The classes of beta mod Z are exactly the eigenvalues
+of M_f up to the monodromy convention, and equal to exp(2 pi i beta)
+with a fixed sign.
+
+Original: Section 2, Theorem (weighted homogeneous spectrum) and its
+proof; Steenbrink 1989, Section 1, p. 165; Stacks Project Tags 00ND
+and 02JN.
+
+Exercise: Nemethi / Balnojan-Hertling. Under what congruence of
+spectra are the real Seifert forms of two weighted homogeneous
+singularities isomorphic?
+
+Theorem (real Seifert forms from the spectrum). Let f, g be isolated
+weighted homogeneous singularities in the SAME number s of variables.
+If their multisets of positive spectral exponents agree modulo 2 Z
+(i.e. for every alpha in R / 2 Z the number of spectral exponents
+congruent to alpha mod 2 is the same for f and g), then their REAL
+Seifert forms (S_f tensor R, S_g tensor R) are isomorphic over R.
+
+This is a consequence of the spectral-pair classification; it is NOT
+an assertion about integral forms. The relevant theorems are Nemethi
+1995 (Theorem 6.5) and Balnojan-Hertling 2019 (Theorem 4.4(b)). For
+weighted homogeneous isolated singularities, the monodromy is
+finite-order (previous exercise), so nilpotent logarithms vanish and
+the spectral pair data picks out the form.
+
+Details of the equivalence. Put m = s - 1. For a Steenbrink spectral
+pair (alpha, w) the first coordinate is a spectral exponent, the
+second is a cohomological weight; for weighted homogeneous isolated
+singularities the second coordinate is m everywhere except at
+eigenvalue 1, where it is m + 1. The definition of spectral pairs
+subtracts the eigenvalue-1 shift, so all second coordinates end up
+equal to m; this coordinate contributes no data. The translation
+beta -> beta - 1 respects congruence mod 2 Z. The cohomology-to-
+homology sign is the fixed number (-1)^((m+1)(m+2)/2), and the duality
+converts the cohomological Seifert form to the homological one
+(Balnojan-Hertling 2019 Eq 6.14). The common sign and duality
+preserve isomorphism. So the conclusion holds cleanly, including the
+eigenvalue-1 part.
+
+Original: Section 2, Theorem (real Seifert forms from the spectrum);
+Nemethi 1995; Balnojan-Hertling 2019.
+
+Exercise: Hertling-Mase / Orlik-Randell. When does the eigenvalue
+multiset determine the integral monodromy module, and for which class
+of singularities?
+
+A chain polynomial is a sum of the form
+
+    x_1^(a_1) x_2 + x_2^(a_2) x_3 + ... + x_(r-1)^(a_(r-1)) x_r
+        + x_r^(a_r),                     a_i >= 2 integers,
+
+where r >= 1 and x_1,...,x_r are independent variables (r = 1 case:
+just x_1^(a_1), a pure power).
+
+Theorem (Orlik-Hertling-Mase). Suppose f is a sum of chain
+polynomials in disjoint variables. Write the characteristic polynomial
+of M_f as
+
+    char(M_f, t)  =  prod_d  Phi_d(t)^(c_d),
+
+where Phi_d is the d-th cyclotomic polynomial and c_d >= 0 are
+integers. Then the pair (L_f, M_f) (as a Z[t]-module with t acting by
+M_f) decomposes as
+
+    (L_f, M_f)  =  direct sum_{j=1}^{max_d c_d}  Z[t] / (p_j(t)),   (2.3)
+
+with
+
+    p_j(t)  =  prod_{d : c_d >= j}  Phi_d(t).
+
+Each summand is a cyclic Z[t]-module with a single cyclic generator,
+whose characteristic polynomial is p_j. Since t maps to M_f, this
+describes (L_f, M_f) completely up to isomorphism as a module with
+operator.
+
+What it does and does not do. It controls the lattice TOGETHER with
+its monodromy operator; it does NOT identify the integral Seifert
+form. The Seifert form requires the additional arithmetic step in
+Sections D-E of this explainer.
+
+Where the theorem comes from. Orlik 1972 formulated the integral
+decomposition; Orlik-Randell 1977 gave the chain calculation;
+Hertling-Mase 2022 Theorem 1.3(a),(c),(d) and Definition 2.6 supply
+the precise statement. Reversing the variable order puts our chains in
+that paper's convention. Each standard block has an integral cyclic
+generator, hence is exactly the companion module in (2.3). Inverting
+the operator preserves cyclicity and characteristic polynomial (its
+roots are symmetric), so the difference between M_f and M_f^(-1)
+in (2.1) causes no change.
+
+Original: Section 2, Theorem (Hertling-Mase); Orlik 1972;
+Orlik-Randell 1977; Hertling-Mase 2022.
+
+Exercise: Hasse-Minkowski for quadratic forms over number fields.
+State it, and explain how the paper uses it.
+
+Theorem (Hasse-Minkowski). Two nondegenerate quadratic forms over a
+number field F are isometric over F if and only if, for every place v
+of F (archimedean or finite), their extensions to the local field
+F_v are isometric over F_v.
+
+Usage. In Section D (local and rational isometries) the paper
+constructs a rational isometry R between two quadratic-form-valued
+operator modules by constructing isometries at every place. The Hasse-
+Minkowski step is the linear-algebra version.
+
+Reference: Milne 2022 Theorem 25.64(c) (with a short discussion);
+O'Meara 1963 Chapter VI.
+
+Original: Section 2, Theorem (Hasse-Minkowski); Milne 2022; O'Meara
+1963.
+
+Exercise: Strong approximation. State it precisely, with the
+connected / simply-connected / almost-simple hypotheses.
+
+Theorem (strong approximation). Let G be a connected, simply
+connected, absolutely almost simple algebraic group over a number
+field F. If the archimedean product
+
+    prod_{v | infinity}  G(F_v)
+
+is NONCOMPACT, then G(F) (the F-points, a discrete subgroup) is DENSE
+in the finite adelic group
+
+    G(A_(F, f))  =  restricted product over finite v of G(F_v),
+
+restricted with respect to integral compact open subgroups.
+
+References: Platonov 1969 Theorem A, with the correction Platonov
+1970; Rapinchuk 2014 Theorem 2.3.
+
+Application in this paper. G is a product of special unitary groups
+SU(h_d) of nondegenerate hermitian forms over totally imaginary
+quadratic extensions E_d of totally real number fields F_d (arising
+from cyclotomic components of the form). In dimension >= 2 such a
+group is connected, simply connected, absolutely almost simple; at
+each real place the form has signature (p,q) with p,q > 0, making the
+archimedean component noncompact. Hypothesis satisfied; strong
+approximation applies.
+
+Original: Section 2, Theorem (strong approximation); Platonov 1969,
+1970; Rapinchuk 2014.
+
+Exercise: Smale (Poincare in high dimensions) and Levine (knot
+classification). State both, and explain the connection.
+
+Statement 1 (Smale 1961). A smooth closed manifold of dimension d >= 5
+that is HOMOTOPY equivalent to the sphere S^d is HOMEOMORPHIC to S^d.
+(No smoothness of the homeomorphism is claimed; the smooth structure
+of the source may be exotic.)
+
+Statement 2 (Levine 1970). Let q >= 3 and let K_1, K_2 be smooth
+oriented submanifolds of S^(2q+1), each homeomorphic to S^(2q-1). Call
+a knot K_j "simple" if
+
+    pi_1( S^(2q+1) \ K_j )  =  Z,
+    pi_i( S^(2q+1) \ K_j )  =  0    for 2 <= i <= q-1.
+
+If both knots are simple and their integral Seifert matrices are
+congruent over Z (correspond to equivalent integral Seifert forms),
+then they are ambiently isotopic in S^(2q+1). In particular their
+oriented sphere pairs are homeomorphic.
+
+Connection with the paper. Put n = N - 1 in the main theorem. In
+Section H (spherical links and simplicity), the two links K_1, K_2 are
+first proved to be homotopy spheres (via det(1 - M_f) = 1 and the Wang
+sequence), then simple knots (via the homotopy exact sequence of the
+Milnor fibration). Levine's theorem then converts the integral Seifert
+congruence into an ambient isotopy; smooth isotopy extension gives an
+oriented diffeomorphism of pairs (S^(2N-1), K_1) -> (S^(2N-1), K_2).
+
+Levine's statement allows smooth homotopy spheres that are NOT the
+standard smooth sphere; no standard-smooth-sphere assertion is needed.
+Related formulations appear in Durfee 1974 and Saeki 1999; see also
+Neumann 2017.
+
+Original: Section 2, Theorem (Smale and Levine); Smale 1961;
+Levine 1970; Durfee 1974; Saeki 1999; Neumann 2017.
+
+==============================================================
+SECTION C: TWO SEEDS WITH DIFFERENT ORDERS
+==============================================================
+
+The whole construction starts with two polynomials p_1, p_2 that
+already have the desired different orders (2 vs. 3) but whose
+topological-type invariants (spectrum, real Seifert form, integral
+monodromy module) agree modulo 2. The "more exotic" work of the later
+sections turns agreement of these invariants into actual integral
+Seifert-form congruence, and then into ambient homeomorphism.
+
+Every construction below uses sums of polynomials in disjoint sets of
+variables, so the exact variable count is a matter of bookkeeping.
+
+Exercise 1: The finite product relation. What two multiplicative
+identities among positive rational numbers underlie the seeds?
+
+Fix
+
+    d = 3^11 = 177147.
+
+Let
+
+    B = { b in Z : 0 < b < d/2,  3 does not divide b }
+
+and define, for each b in B, the positive rationals
+
+    A_b  =  (d - b) / b,
+    R_b  =  ( (2d + b)(2d - b) ) / ( (3d - b)(d + b) ).
+
+Lemma 3.1 of the paper. There are integers e_b, indexed by b in B,
+such that
+
+    E  :=  sum_{b in B}  e_b  >  0,                                (3.1)
+    prod_{b in B} A_b^(e_b)  =  1,
+    prod_{b in B} R_b^(e_b)  =  1.
+
+Proof strategy. Form the integer matrix M with column indexed by b
+whose entries are the prime valuations v_p(A_b) and v_p(R_b), for all
+primes p at which either is non-integer; two separate families of
+rows, one for A and one for R. Finding e_b as in (3.1) is finding a
+single integer kernel vector whose coordinate SUM is positive.
+
+The paper exhibits this by an EXPLICIT FINITE certificate. Retain only
+the columns for which each of
+
+    d-b,  b,  2d+b,  2d-b,  3d-b,  d+b
+
+has no prime factor > 4450. Then repeatedly delete any column that is
+the only column supported on some nonzero row; discard zero rows. The
+resulting matrix M and its augmentation M~ (append a final row of
+ones) have the following parameters, computed exactly:
+
+    After the prime cutoff:      1118 columns, 1100 nonzero rows.
+    After all deletions:          896 columns,  867 nonzero rows.
+    rank over F_1009 of  M~:      868.
+
+This is the sole finite calculation needed for the existence of the
+two product identities. The 896 columns and 867 rows of M are
+integers; M~ is M with a final row of ones appended. Since M~ is
+868 x 896 with rank 868 > 867 (its last column set), the augmented
+matrix is rank 868 while the unaugmented is at most 867.
+
+Consequence: an 868 x 868 square submatrix H of M~ with det H != 0
+exists. Set
+
+    z  =  adj(H) (0, ..., 0, 1)^t,
+
+where adj(H) = adjugate matrix, so H z = det(H) (0, ..., 0, 1)^t.
+Then z is integral, satisfies the valuation equations on those 896
+columns, and has coordinate sum det H (because the last row of M~ is
+all ones). Extend z by zeros to all columns b in B, multiply by -1 if
+necessary so E = |det H| > 0. Every valuation equation holds, so every
+prime valuation of both products in (3.1) vanishes; a positive
+rational with vanishing every prime valuation is 1. This gives e_b.
+
+The lexicographically first nonsingular minor works (with columns
+ordered by b and rows by (family, prime)); no size bound is needed.
+
+Original: Section 3, Lemma 3.1 and its proof.
+
+Exercise 2: Why does the "certificate" translation to code produce
+896 and 868 exactly, and what is the exact integer program?
+
+The paper's Appendix A gives a self-contained Python program that
+uses only integer arithmetic and Python's standard library, no
+floating-point, no third-party packages. It:
+
+  (a) Sieves the integers up to d + d/2 + 2 = 3d/2+2 ~ 265,721 to
+      produce a small prime divisor table for every integer > 1. From
+      that table it reads off, by repeated integral division, exact
+      prime valuations of the six small factors d-b, b, 2d+b, 2d-b,
+      3d-b, d+b.
+  (b) Builds one integer column per b in B (about 59,049 b's, since
+      |B| = phi(3) * d/3 = (2/3) d / 2 = d/3 ... actually |B| ~ d/3
+      = 59049). Rows are indexed by (family, prime) pairs; six factors
+      produce their signed valuation columns.
+  (c) Restricts to b for which each factor has no prime > 4450:
+      reduces to 1118 columns.
+  (d) Repeatedly deletes any column that is the only column
+      supported on some nonzero row. Discards zero rows. Result: 896
+      columns, 867 nonzero rows.
+  (e) Appends a final row of ones, obtaining the augmented matrix M~.
+  (f) Performs Gaussian elimination over the finite field F_1009,
+      storing one normalised pivot per leading row. The number of
+      stored pivots equals the rank; result 868.
+
+Correctness of the rank check. The prime 1009 is prime: none of
+2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31 divides it, and 32^2 > 1009.
+The rank mod a prime equals the rank over Q minus the number of prime
+divisors of the relevant minors; since the mod-1009 rank is at most
+the Q-rank, and since the determinant of H (a minor of M~) is not
+divisible by 1009 (if we assert this), rank over Q is also 868. The
+paper's argument is careful on this point and uses the adjugate
+minor whose determinant survives reduction mod 1009; equivalently,
+the elementary integral computation certifies rank 868 over Q.
+
+Supplementary certificate: explicit-relation.json. The paper's
+Appendix A also mentions a separately supplied file
+"verification/support/explicit-relation.json" containing 849 nonzero
+integer coefficients (each with at most 140 decimal digits) whose
+coordinate sum is positive. A short standard-library program
+"verification/validation/explicit_relation.py" verifies both product
+identities in (3.1) by brute-force trial-division of the six integers
+in every row, without needing the matrix or its rank. This is an
+independent computational witness to Lemma 3.1. The rank certificate
+above and the adjugate argument already prove the lemma; the JSON is
+supplementary.
+
+Original: Section 3, Lemma 3.1; Appendix A (certificate).
+
+Exercise 3: Completing a rational weight to an isolated chain
+singularity. State and prove Lemma 3.2.
+
+Lemma 3.2. Let D be a positive odd integer, and let c be an integer
+with 0 < c < D/2. There is a chain polynomial
+
+    F_c(x_1, ..., x_l)
+        =  x_1^(a_1) x_2 + x_2^(a_2) x_3 + ... + x_(l-1)^(a_(l-1)) x_l
+             + x_l^(a_l),                                    (3.2)
+
+of weighted degree 1, whose first weight is c/D, all of whose weights
+lie strictly in (0, 1/2), and all of whose ordinary monomial degrees
+are >= 3. It has an isolated critical point at the origin. Removing
+its first variable and first monomial gives a chain with exactly the
+remaining weights (when l > 1).
+
+Construction. Set c_1 = c and perform successively
+
+    a_i  =  floor( D / c_i ),        c_(i+1)  =  D - a_i c_i.    (3.3)
+
+Stop at the first zero remainder. Since each remainder 0 <= c_(i+1) <
+c_i, the process terminates in finitely many steps.
+
+Give x_i weight c_i/D. Then a_i c_i + c_(i+1) = D for every
+nonterminal term, and a_l c_l = D for the terminal term. All a_i >= 2
+(because c_i < D/2 < D, so floor(D/c_i) >= 2). And since c_i < D/2,
+weight c_i/D < 1/2.
+
+Each nonterminal monomial x_i^(a_i) x_(i+1) has ordinary degree
+a_i + 1 >= 3. The terminal monomial x_l^(a_l) has ordinary degree
+a_l = D/c_l; since c_l divides the odd integer D and is not 1 (c_l <
+c_(l-1) < ... <= c < D/2) and is not 2 (D is odd), a_l is at least 3
+(dividing odd D with no factor 2 means a_l cannot be 2).
+
+Isolation. Suppose a nonzero point x is critical. Let j be the
+SMALLEST index with x_j != 0. If j = l, the derivative in x_l is
+a_l x_l^(a_l - 1) != 0. Otherwise the derivative in x_j is
+a_j x_j^(a_j - 1) x_(j+1) + (possible tail contributions). Because the
+coefficient x_j^(a_j-1) is nonzero and no other term of that derivative
+contains x_j, the derivative forces x_(j+1) = 0. Then the derivative
+in x_(j+1) is x_j^(a_j) + (tail terms that vanish since a_(j+1) >= 2
+and x_(j+1) = 0 and x_(j+2) = 0, ...). This derivative equals x_j^(a_j)
+!= 0, a contradiction. So there is no nonzero critical point.
+
+Tail structure. Removing x_1 and the first monomial leaves
+x_2^(a_2) x_3 + ... + x_l^(a_l), which is again a chain (in variables
+x_2, ..., x_l), with weights c_2/D, ..., c_l/D. Apply Lemma 3.2 again
+if needed. The isolation proof limits to the smaller set of variables
+the same way.
+
+Original: Section 3, Lemma 3.2 and its proof.
+
+Exercise 4: The weight table. How are the two seed polynomials p_1,
+p_2 assembled, and where does the "4E" balancing come from?
+
+Fix D = 3 d = 3^12 and use the integers e_b of Lemma 3.1. Prepare
+signed counts:
+
+    w = b / d:          count of side (2) minus side (1)  =  + 2 e_b.
+    w = b / (3d):       count of side (2) minus side (1)  =  - 2 e_b.
+    w = (d - b) / (3d): count of side (2) minus side (1)  =  + 2 e_b.
+    w = (d + b) / (3d): count of side (2) minus side (1)  =  + 2 e_b.               (3.4)
+
+All these weights lie in (0, 1/2): the numerators are
+3b, b, d-b, d+b and each is < D/2 = 3d/2. Their common denominator is
+D = 3d.
+
+For each b:
+    side 2 receives 2e_b copies of weight b/d, 2e_b copies of weight
+    (d-b)/(3d), 2e_b copies of weight (d+b)/(3d);
+    side 1 receives 2e_b copies of weight b/(3d).
+
+Then for each starting weight c/D (one of these), realize each of its
+occurrences by a copy of the chain in Lemma 3.2 (for c/D), inserted
+into the side where that occurrence lives. On the OPPOSITE side
+insert just the TAIL of that same chain (the chain with its first
+variable and first monomial removed; empty if the chain was a pure
+power).
+
+Then: tail weights appear EQUALLY on both sides (each chain on side
+2 contributes its tail to side 1, and vice versa), so they cancel in
+a "difference of variable counts" sense. Only the FIRST weights are
+unbalanced.
+
+Difference in variable counts (side 2 minus side 1) after this stage:
+
+    sum over b of  (2e_b + 2e_b + 2e_b) - (2e_b)  =  sum over b of 4 e_b = 4 E.        (3.5)
+
+So side 2 has 4E MORE variables than side 1. Fix by adding 4E
+independent squares (each z^2) to side 1.
+
+Now both sides have equal variable count. Also:
+
+    - Every starting weight < 1/2, every chain weight < 1/2, and 1/2
+      for squares, so all weights are in (0, 1/2].
+    - Every chain monomial has degree >= 3. Every square has degree 2.
+    - 4E is even, and the number of copies of each chain tail that was
+      included on each side is even; both sides have EVEN variable
+      count when finished.
+
+Finally, add TWO INDEPENDENT PURE CUBES (x^3 each) to BOTH sides, to
+eliminate the case 3 | k in the spectrum comparison (below). The total
+variable count on each side is then s_0 (even), and the two seeds
+p_1, p_2 are constructed.
+
+Resulting orders:
+
+    p_1 (side 1) contains at least one square (from 4E >= 4 squares
+    since E >= 1), and all other monomials have ordinary degree >= 3.
+    So ord_0(p_1) = 2.
+
+    p_2 (side 2) contains no squares, no linear, no constant terms; all
+    monomials have ordinary degree >= 3. So ord_0(p_2) = 3.
+
+Coefficients can all be taken to be +1.
+
+Original: Section 3, Proposition 3.3, proof (construction); weight
+table (3.4), balance (3.5).
+
+Exercise 5: Proposition 3.3, stated in full.
+
+Proposition 3.3. There are real-coefficient weighted homogeneous
+polynomials p_1, p_2, each in the SAME even number s_0 of variables,
+such that:
+
+  (ii-a)  both are sums of chains (in disjoint variables) with
+          exponents >= 2, both have isolated critical points;
+  (ii-b)  ord_0(p_1) = 2 and ord_0(p_2) = 3;
+  (ii-c)  their spectral multisets are EQUAL modulo 2 (in the sense of
+          the Nemethi / Balnojan-Hertling theorem from Section B);
+  (ii-d)  their real Seifert forms are isomorphic; SEPARATELY, their
+          integral monodromy modules are isomorphic;
+  (ii-e)  both monodromy operators have order dividing D = 3^12.
+
+At the top we already know (ii-a), (ii-b), (ii-e). The hard part is
+(ii-c), from which (ii-d) follows via the two structure theorems.
+
+Original: Section 3, Proposition 3.3.
+
+Exercise 6: The spectrum comparison. Why do p_1 and p_2 have equal
+spectra modulo 2? Explain all three cases.
+
+Use the weighted spectrum formula (2.2) from Section B:
+
+    P_p(t)  =  sum_{alpha} t^alpha  =  prod_w  ( t^w - t ) / ( 1 - t^w ).
+
+All exponents of either p_j lie in (1/(2D)) Z (D = 3^12 = 3d, squares
+have weight 1/2, chains have denominator dividing D). Reduction mod 2
+puts multiplicities in the group algebra of the finite cyclic group
+(1/(2D)) Z / 2 Z. Its characters are the maps
+alpha  ->  exp(pi i k alpha) for k = 0, 1, ..., 4D-1 (integers).
+
+By Fourier inversion on the finite group, it suffices to prove
+
+    P_{p_1}  =  P_{p_2}  evaluated at  t^u = exp( (pi i k + eps) u ),
+    let eps -> 0+,
+
+for each character k. When an individual quotient ( t^w - t ) /
+( 1 - t^w ) is 0/0 at eps = 0, evaluate its limit by l'Hopital.
+
+CASE 1: k is even. Put k = 2m with m an integer >= 0.
+
+For a single weight w, the factor at character exp(pi i k w) has value
+
+    (t^w - t)/(1 - t^w)|_{k=2m}     =
+        -1                    if m w is not an integer,
+        (1 - w)/w             if m w is an integer.                  (3.6)
+
+(The second line comes from differentiating numerator and denominator
+and taking the ratio; both diverge to 0 as eps -> 0+, and the ratio
+has the finite limit (1 - w)/w.)
+
+None of these values is zero, so common weights can be cancelled.
+Which weights give EXCEPTIONAL value (1-w)/w vs. NONEXCEPTIONAL value
+-1? The weight w has DENOMINATOR d or 3d in lowest terms:
+
+    - Starting weight b/d (denominator d): m w is an integer iff
+      m b is divisible by d, i.e. iff m is divisible by d (since
+      3 does not divide b, i.e. b and d coprime? actually just: b is
+      coprime with d = 3^11 by definition of B, so b/d is in lowest
+      terms, and m b / d is an integer iff d | m).
+      So either ALL b/d-type weights are exceptional or NONE are.
+    - Starting weights b/(3d), (d-b)/(3d), (d+b)/(3d) (denominator 3d):
+      m w is an integer iff 3d | m numerator, i.e. iff 3d | m (since
+      all numerators b, d-b, d+b are coprime with 3d - this is the
+      reason 3 is excluded from B and why the b/(3d) etc. are in
+      lowest terms). So: all exceptional, or none.
+    - Chain tail weights: their denominators divide D = 3d; same
+      grouping.
+    - Weight 1/2 for squares: m w is an integer iff m is even iff
+      2 | m.
+
+So the exceptional / nonexceptional condition is uniform within each
+denominator class {d, 3d, 2}. So we can pair up factor values between
+p_1 and p_2 in two blocks:
+
+Block d (denominator d): signed count 2e_b. Exceptional value:
+    prod_b ((1 - b/d)/(b/d))^(2e_b) = prod_b A_b^(2e_b) = 1.
+
+Block 3d (denominator 3d): signed counts -2e_b, 2e_b, 2e_b for the
+three weights. Exceptional value:
+
+    prod_b  [ ( b/(3d) ) / ( 1 - b/(3d) ) ]^(2e_b)  -- for the b/(3d) weights
+        x [ ( 1 - (d-b)/(3d) ) / ( (d-b)/(3d) ) ]^(2e_b)
+        x [ ( 1 - (d+b)/(3d) ) / ( (d+b)/(3d) ) ]^(2e_b)
+    = prod_b ( R_b / A_b )^(2e_b)
+    = (prod_b R_b^(e_b))^2 / (prod_b A_b^(e_b))^2
+    = 1.
+
+Nonexceptional signs are (-1)^(signed count). Since every signed count
+is even, these contribute 1. So contributes nothing.
+
+Squares: exceptional value is (1 - 1/2)/(1/2) = 1, nonexceptional
+value is -1. Side 1 has 4E squares, side 2 has none, so ILL-DEFINED
+unless 4E is even; then (-1)^(4E) = 1.
+
+Combining: P_{p_1}(k) = P_{p_2}(k) for k even. This includes k = 0
+because the shift t^(2E) from squares (see Case 3 below) has 2E
+divisible by 2 (mod 2).
+
+CASE 2: k is odd and 3 does not divide k.
+
+For k odd, the factor eval is
+
+    (t^w - t)/(1 - t^w)|_{k odd}  =  i cot( pi k w / 2 ).          (3.7)
+
+No factor is zero or pole: a zero of cot would require k w / 2 to
+vanish mod 1, so 2 | k (but k odd); a pole would require pi w k / 2
+= pi/2 mod pi, i.e. wk = 1 mod 2, i.e. k w is an odd integer, so
+denominator of w in lowest terms divides k; but starting weights and
+chain weights have denominators nontrivial powers of 3, and 3 does not
+divide k by hypothesis of Case 2. Squares have denominator 2, and
+their factor is exp(pi i k / 2)^{1/2}-ish combination, finite and
+nonzero and can be handled separately: 4E of them, total contribution
+1 (as in Case 1 with even power).
+
+Common weights can be cancelled. What remains is the exact signed
+cotangent identity for each b:
+
+    C(b/d)  =  C(b/(3d)) / ( C((d-b)/(3d)) * C((d+b)/(3d)) ),       (3.8)
+
+where C(u) = cot(pi k u / 2).
+
+Proof of (3.8): Put x = pi k b / (6d). Then use the standard triple-
+angle product
+
+    cot(x) * cot(x + k pi/3) * cot(x + 2k pi/3)  =  - cot(3 x).     (3.9)
+
+This is a standard trigonometric identity for any x not on the pole
+lattice. The shifts by k pi/3 give the same three factors modulo pi as
+the shifts by pi/3, because 3 does NOT divide k (by Case 2 hypothesis),
+so k = 1 or 2 mod 3 and shifts by k pi/3 permute the shifts by pi/3.
+
+Then with k odd:
+
+    cot(x + k pi/3)     =  1 / C((d-b)/(3d)),
+    cot(x + 2 k pi/3)   =  -1 / C((d+b)/(3d)),
+
+and cot(3 x) = cot(pi k b / (2d)) = C(b/d). Substituting into (3.9)
+gives exactly (3.8). Raising (3.8) to the 2e_b-th power and multiplying
+over b gives exactly the asserted signed cotangent product = 1.
+
+Squares contribute C(1/2) = cot(pi k / 4), which is (+1) or (-1) for
+k odd, and 4E of them (even) gives 1.
+
+Finally, the powers of i (not the cotangents) are equal because the
+two complete polynomials have the same total number s_0 of variables
+(the exponent of i in the product is (s_0) / 2, which is a half
+integer; the ratio is i^0 = 1 because s_0 is the same).
+
+CASE 3: k is odd and 3 divides k.
+
+A pure cube has spectrum t^(1/3) + t^(2/3). At the character k ->
+exp(pi i k u):
+
+    t = exp(pi i k): t^(1/3) evaluates to exp(pi i k / 3),
+                     t^(2/3) evaluates to exp(2 pi i k / 3).
+
+    If k is an odd multiple of 3, k = 3 m with m odd:
+
+    exp(pi i k / 3) = exp(pi i m) = -1,
+    exp(2 pi i k / 3) = exp(2 pi i m) = +1.
+    Sum: -1 + 1 = 0.
+
+Crucially, BEFORE the two common cubes were added, both sides were
+isolated sums of chains (roots of unity of order a power of 3) and
+squares (weight 1/2). Their spectra were finite polynomials (as
+opposed to analytic functions), so multiplying by the cube spectrum
+makes both character evaluations equal to 0. No accidental
+cancellation of a zero against a pole of an individual weight
+quotient is required; the argument uses those finite polynomials
+prior to the common cube factor.
+
+This completes the Fourier comparison of Case 3.
+
+COMBINING all three cases: P_{p_1}(k) = P_{p_2}(k) for every character
+k of the finite group (1/(2D)) Z / 2 Z. By Fourier inversion, the
+spectra agree modulo 2.
+
+Original: Section 3, Proposition 3.3, proof of spectra; Fourier
+inversion argument; trigonometric identities (3.6)-(3.9).
+
+Exercise 7: Deduce real Seifert form and integral monodromy isomorphy
+from equal spectra mod 2.
+
+Real Seifert form. Below we have weighted homogeneous monodromy (finite
+order hence semisimple). Theorem 2.1 (Nemethi/Balnojan-Hertling, in
+Section B) converts equality of spectra modulo 2 into real Seifert form
+isomorphism. The common number of variables s_0 is even; that is
+consistent with the symmetrization convention (2.1) used in the paper.
+
+Integral monodromy module. All summands of p_1 and p_2 are chains
+(including the pure powers as degenerate r=1 chains). Theorem (Hertling-
+Mase) applied to sums of chains gives that the integral monodromy
+module is determined by the eigenvalue MULTIPLICITIES; equal spectra
+give equal eigenvalue multiplicities; hence (via Hertling-Mase) the
+integral monodromy modules are isomorphic.
+
+These are two SEPARATE conclusions. There is NO claim of a single
+integral map compatible with both the Seifert form and the monodromy;
+constructing such a map is the goal of later sections.
+
+Monodromy orders. Every nonsquare weight has denominator dividing
+D = 3d = 3^12. Each square (weight 1/2) contributes t^(1/2) spectrum;
+4E of them contribute t^(2E) (with integer exponent 2E = 2*sum_b e_b;
+2E is up to us, but in any case integer). All spectral exponents of
+either seed are therefore in (1/D) Z. Weighted homogeneous monodromy
+eigenvalues are D-th roots of unity, so M_{p_j}^D = 1 (semisimplicity
+gives equality of the operator to identity, not just of eigenvalues).
+So both monodromies have order dividing D. Claim (ii-e) proven.
+
+Original: Section 3, proof of Proposition 3.3 (closing paragraphs).
+
+Exercise 8: Corollary 3.4: preparing the seeds by removing eigenvalue
+1.
+
+Corollary 3.4. Add the two same new variables x, y with pure powers
+x^5 and y^7 to each seed:
+
+    q_1  =  p_1 + x^5 + y^7,
+    q_2  =  p_2 + x^5 + y^7.                                (3.10)
+
+Then q_1, q_2 are real-coefficient isolated sums of chains in the same
+even number s = s_0 + 2 of variables, with orders 2 and 3. Their real
+Seifert forms are isomorphic; their integral monodromy modules are
+isomorphic. Both monodromies have odd order dividing
+
+    m  =  3^12 * 5 * 7,                                     (3.11)
+
+and neither has eigenvalue 1.
+
+Proof. Isolation and ordinary orders are preserved because the added
+terms x^5, y^7 have degrees 5 and 7 >= 3 (they do not touch the lowest
+order term). The spectrum formula (2.2) multiplies each seed spectrum
+by the same two pure-power spectra
+
+    S_5(t)  =  (t^(1/5) - t) / (1 - t^(1/5))  =  ...  (finite power),
+    S_7(t)  =  (t^(1/7) - t) / (1 - t^(1/7)),
+
+so spectral equality mod 2 is preserved. Nemethi/Balnojan-Hertling and
+Hertling-Mase apply again (chains plus pure powers).
+
+Order of the new monodromy. Each eigenvalue has the form
+
+    alpha * beta * gamma,
+    alpha^D = 1,  beta^5 = 1,  gamma^7 = 1,
+    beta  != 1,  gamma != 1,
+
+where beta, gamma are primitive 5th and 7th roots of unity. Since
+3^12, 5, 7 are pairwise coprime, this common eigenvalue has order
+dividing 5 * 7 * D = 35 D, actually multiplying 3^12 * 5 * 7 = m
+(because D = 3^12 already). The order is divisible by 5 because beta
+!= 1 forces a nontrivial 5-primary component (5 is coprime to D and
+7); similarly for 7. In particular a common eigenvalue cannot be 1.
+Semisimplicity gives M_{q_j} = identity iff every eigenvalue is 1, so
+M_{q_j} has order dividing m and no eigenvalue 1; same holds for the
+operator itself.
+
+Original: Section 3, Corollary 3.4 and its proof.
+
+==============================================================
+SECTION D: FROM TWO SEEDS TO THE SAME LATTICE, LOCALLY AND
+RATIONALLY
+==============================================================
+
+We now have two seed polynomials q_1, q_2 in the same number s of
+variables, with real Seifert forms isomorphic and integral monodromy
+modules isomorphic. The two identifications are SEPARATE. The goal of
+this section is to make them compatible, but only first rationally and
+locally (one prime at a time). Section E will combine the local pieces
+into one integral statement.
+
+The standing set-up: L_j = Milnor lattice of q_j; S_j = integral
+Seifert form; M_j = monodromy operator, of odd order dividing
+m = 3^12 * 5 * 7. Neither M_j has eigenvalue 1.
+
+Exercise 1 (Lemma 4.1, symmetrization). How does an integral Seifert
+form become an integral symmetric form?
+
+Fix an odd integer m divisible by the orders of both M_1 and M_2 (so
+m works; e.g. m = 3^12 * 5 * 7 = 3^{12} * 35). Choose r in Z with
+
+    2 r  congruent to  1  modulo m.                          (4.1)
+
+(Any such r exists since m is odd: 2 is invertible mod m, and we take
+r = (m+1)/2 mod m.) Put
+
+    T_j  =  S_j^{-1} S_j^t,        (same as M_j by (2.1))
+    B_j  =  S_j T_j^r.                                       (4.2)
+
+Lemma 4.1. Each B_j is an INTEGRAL, UNIMODULAR, SYMMETRIC form,
+preserved by T_j. The pair (B_j, T_j) recovers S_j via S_j = B_j
+T_j^{-r}. An isometry of S_j intertwining T_j preserves B_j.
+
+Proof. Suppress j. Transposing S^t = S T gives S = T^t S^t = T^t S T,
+so T^t S T = S. Raising to the r-th power and multiplying on the right
+by S,
+
+    (T^t)^r S T^r  =  S,   hence   (T^t)^r S  =  S T^{-r}.
+
+Then
+
+    B^t  =  (T^r)^t S^t  =  (T^t)^r (S T)  =  (S T^{-r}) T  =  S T^{1-r}.
+
+Since 2r = 1 mod m, we have 1 - r = r mod m, so T^{1-r} = T^r and
+B^t = S T^r = B. Symmetry holds.
+
+Integrality and unimodularity. B = S T^r is integral (product of two
+integral matrices); det B = det S * (det T)^r = (+/-1) * (unit)^r =
++/-1; so B is unimodular.
+
+T-invariance. T^t B T = T^t S T^r T = (T^t S T) T^r = S T^r = B.
+
+Isometry. An isometry phi of (S, T) satisfies phi^t S phi = S and
+phi T = T phi. Using phi^t S = S phi^{-1} and T^r phi = phi T^r:
+
+    phi^t B phi  =  phi^t S T^r phi  =  (S phi^{-1}) (phi T^r)  =  S T^r = B.
+So it preserves B. QED.
+
+Tensor products. Combining (4.2) with the tensor-product conventions:
+if the two factors both have EVEN ambient dimension, then by Thom-
+Sebastiani-Sakamoto the tensor product of the symmetric forms is
+again of the same form; the (-1)^{ab} tensor sign is same on both
+sides when the factor dimensions agree, so it cancels. Concretely:
+    B_{x (x) y}  =  (S_x (x) S_y) (T_x (x) T_y)^r
+              =  (S_x T_x^r) (x) (S_y T_y^r)  =  B_x (x) B_y.
+So symmetrization commutes with tensor products of even-variable
+factors up to the SAME sign on both sides.
+
+Original: Section 4, Lemma 4.1 and its proof.
+
+Exercise 2: Why does the paper want an ODD r in (4.1), and where does
+the prime 2 difficulty come from?
+
+A tensor square can have eigenvalue 1 even if both factors do not:
+    M_f has no eigenvalue 1, but M_f (x) M_f can have eigenvalue 1
+    (for instance if the eigenvalues of M_f on the two factors are
+    lambda and lambda^{-1} with lambda != 1).
+This makes quadratic form theory over Q and Z_2 harder (a hermitian
+form vs. an ordinary orthogonal form appears at eigenvalue 1).
+
+The choice m = 3^12 * 5 * 7 makes m ODD (all prime factors are odd).
+So the monodromy operator is odd-order. Over the rationals, this
+implies no eigenvalue -1 (since -1 has order 2 and 2 does not divide
+m). Over Z_2, it implies the reduction of x^m - 1 mod 2 is separable,
+which allows the Z_2-algebra Z_2[X]/(X^m - 1) to decompose as a
+product of integer rings of UNRAMIFIED extensions (see Prop 4.4).
+
+The oddness of m and absence of eigenvalue 1 in the ORIGINAL M_j (not
+the tensor) is what lets the prime-2 argument be a clean "all forms
+of the same rank are isometric" statement, meaning that the prime-2
+step is decided by RANK ALONE, with no further invariants.
+
+Original: Section 4, beginning and Prop 4.4 statement.
+
+Exercise 3 (Lemma 4.2, lifting in a finite operator algebra).
+
+Lemma 4.2. Let l be an odd prime and let A be a commutative Z_l-
+algebra that is finite as a Z_l-module. Write J for its Jacobson
+radical. Then A/J is a product of finite fields, and A is complete for
+the J-adic topology. For a unit u in A:
+
+    (1) if its image in every residue field is a square, then u is a
+        square in A;
+    (2) u = a^2 + b^2 for some a, b in A.
+
+Proof. Since A is a finite Z_l-algebra, A/ l A is finite and l is in
+its Jacobson radical; the same for A itself. So l A subset of J (the
+Jacobson radical contains the nilpotent-after-quotient l A; since
+A/lA is finite commutative, its nilradical is its Jacobson radical).
+So J contains l A, and A is complete for the l-adic (hence J-adic)
+topology. Quotient A/J = A/lA / (J/lA) is a product of finite fields
+by structure theory of finite commutative algebras.
+
+(1) Newton iteration. Suppose u is a unit whose image in each residue
+field is a square. Choose a lift v of a chosen square root in A/J
+(A/J is a product of finite fields; pick any square root of the image
+of u in each field, assemble by CRT; v may be chosen to be a unit
+because u is a unit). Let e = v^2 - u in J^k for some k >= 1. Define
+
+    v'  =  v  -  e (2 v)^{-1}.
+
+The new error is
+    e'  =  (v')^2 - u
+        =  v^2 - 2 v e/(2v) + (e/(2v))^2 - u
+        =  e - e + (e/(2v))^2
+        =  e^2 / (2v)^2.
+So e' is in J^{2k}. Iteration converges to a square root of u.
+
+(2) In a finite field k of odd cardinality, the set of squares has
+(|k|+1)/2 elements, and for any element v in k, the set v - (squares)
+also has (|k|+1)/2 elements. These sum to |k|+1 > |k|, so they must
+intersect; there exist a, b with a^2 + b^2 = v. (Two subsets of k with
+sum of cardinalities > |k| intersect.)
+
+Choose, in each residue field of A, a, b with a^2 + b^2 = v, and lift
+to a_0, b_0 in A. Then c = a_0^2 + b_0^2 is a unit of A: in every
+residue field, c maps to v, a unit. Now u/c is a unit congruent to 1
+mod J, so by (1) it is a square: u/c = w^2 for some unit w in A.
+Take a = w a_0, b = w b_0. Then a^2 + b^2 = w^2 (a_0^2 + b_0^2) = w^2 c
+= u. QED.
+
+Original: Section 4, Lemma 4.2 and its proof.
+
+Exercise 4 (Proposition 4.3, odd-prime tensor squares).
+
+Proposition 4.3. Let l be an ODD prime, L a finite free Z_l-module,
+T an automorphism of L. If B_1, B_2 are two unimodular symmetric
+forms on L, both T-invariant, then
+
+    (L tensor L, T tensor T, B_1 tensor B_1)
+    ~=  (L tensor L, T tensor T, B_2 tensor B_2)                  (4.3)
+
+via an equivariant integral isometry.
+
+Proof. Put P = B_1^{-1} B_2. Unimodularity makes P an integral auto-
+morphism of L. Symmetry of B_2 makes it SELF-ADJOINT for B_1, and
+invariance of both forms for T makes P commute with T.
+
+Decompose by the characteristic polynomial of P mod l: factor char(P,
+X) as prod_i f_i(X)^{e_i} over F_l, with distinct irreducible f_i.
+Lift this coprime factorisation to char(P, X) = prod_i F_i(X) over
+Z_l, with F_i mod l = f_i^{e_i}. (Illustration: the harder direction
+is lifting a coprime factorisation mod l to a factorisation over Z_l;
+the paper includes the Hensel-lifting argument - see next para.)
+The Chinese-remainder idempotents for the F_i, evaluated at P, give
+
+    L  =  direct sum_i  L_i,     F_i(P_i) = 0 on L_i,  P_i = P|L_i.
+
+These idempotents are polynomials in P; they commute with T and are
+self-adjoint for B_1, so the sum is orthogonal for both B_1 and B_2
+and each L_i is T-stable. Each B_{1,i} = B_1|L_i is unimodular.
+
+Hensel lifting of coprime factors. Suppose we need to lift a coprime
+factorisation g * h mod l, with deg g = a, deg h = b, to
+Z_l factorisation G * H of a given monic polynomial F. Iteratively
+mod l^{n+1} from mod l^n: seek DeltaG, DeltaH (deg < a, < b) to fix the
+product. The correction map
+
+    (u, v)  ->  u g + v h,    deg u < a, deg v < b,
+
+is a bijection onto polynomials of degree < a + b: its kernel is zero
+by coprimality, and dimensions match (both sides have dimension a+b).
+So we can correct the product by any prescribed error at the next
+power of l without changing monicity. Completeness of Z_l gives the
+lift.
+
+Case i = j (diagonal block): consider a residue field k of A_{ii},
+the Z_l-algebra freely generated by P_i tensor 1 and 1 tensor P_i on
+L_i tensor L_i. Let x, y in k be the images of these two generators.
+The relation F_i(P_i) = 0 implies both x, y are roots of the same
+irreducible polynomial f_i over F_l. In a finite field, roots of the
+same irreducible polynomial are FROBENIUS CONJUGATES: y = x^(l^h) for
+some h >= 0. Then
+
+    x y  =  x^{1 + l^h}
+         =  ( x^{(1 + l^h)/2} )^2,
+
+a square (exponent (1 + l^h)/2 is an integer because l^h is odd).
+Since this holds in every residue field of A_{ii}, Lemma 4.2(1) gives
+a in A_{ii} with a^2 = U_{ii} := P_i tensor P_i.
+
+Because a is a product of polynomials in P_i (self-adjoint for B_1),
+a is self-adjoint for B_1 tensor B_1. Thus for all z, w in
+L_i tensor L_i:
+
+    (B_1 tensor B_1)(a z, a w)  =  (B_1 tensor B_1)(z, a^t a w)
+                                =  (B_1 tensor B_1)(z, U_{ii} w)
+                                =  (B_2 tensor B_2)(z, w),
+
+since U_{ii} = P_i tensor P_i relates the two tensor forms on the
+diagonal block by definition. So a is the required isometry on L_i (x)
+L_i.
+
+Case i < j (cross block): pair L_i tensor L_j with L_j tensor L_i.
+Factor swap identifies their operators (T tensor T) and both forms
+(B_1 tensor B_1 and its version for i, j swapped). On the resulting
+two equal blocks, use Lemma 4.2(2) to find a, b in A_{ij} (the
+algebra generated by P_i tensor 1 and 1 tensor P_j) with
+
+    a^2 + b^2  =  U_{ij}  :=  P_i tensor P_j.
+
+Then the block matrix
+
+    M  =  ( a    b )
+          ( -b   a )
+
+satisfies M^* M = U_{ij} * identity, is invertible, and commutes with
+T tensor T. (Sesquilinear adjoint * with respect to the first tensor
+form.)
+
+Combining the diagonal-block maps and the paired-block maps gives one
+integral equivariant isometry from B_2 tensor B_2 to B_1 tensor B_1
+(up to transposition), and therefore (4.3). QED.
+
+Original: Section 4, Proposition 4.3 and its proof.
+
+Exercise 5 (Proposition 4.4, dyadic uniqueness). What is special
+about the prime 2 when the order is odd and no eigenvalue is 1?
+
+Proposition 4.4. Let L be a finite free Z_2-module, and T an
+automorphism of L satisfying T^m = 1 for some ODD m. Suppose that 1
+is NOT an eigenvalue of T on L tensor Q_2. Then ANY two T-invariant
+unimodular symmetric forms on L are equivariantly isometric over Z_2.
+
+Proof. The reduction of X^m - 1 mod 2 is separable (its derivative is
+m X^{m-1} = X^{m-1} mod 2, no common roots with X^m - 1 since X = 0
+is not a root of X^m - 1 mod 2). So X^m - 1 factors into distinct
+irreducible polynomials mod 2, which by Hensel lift to a factorisation
+
+    O  :=  Z_2[X] / (X^m - 1)  =  prod_i  O_i,                (4.4)
+
+each O_i the integer ring of an UNRAMIFIED extension of Q_2 (since
+the residue extension is separable of finite degree and Z_2 integral).
+The action X -> T decomposes L = sum_i M_i, each M_i finite free over
+O_i (module over a DVR, torsion-free, finite -> free).
+
+Invariance of a form B gives: the adjoint of X (i.e. of T) is X^{-1}.
+So the involution sigma of O = prod O_i given by X -> X^{-1} is the one
+respecting forms.
+
+Case A: sigma EXCHANGES factors i and i'. Then M_i and M_{i'} are
+isotropic over each other: B(M_i, M_i) = 0 (skew-symmetric under a
+swap), and B(M_i, M_{i'}) identifies M_{i'} with the Z_2-dual of M_i
+with sigma-transported action. Such a pairing has a UNIQUE equivariant
+isometry class for prescribed modules: an isomorphism on M_i extends
+by its inverse dual to an isometry of the pair (M_i, M_{i'}).
+
+Case B: sigma PRESERVES the factor i. Write D = O_i and D_0 = D^sigma. The
+involution sigma is nontrivial on D: if sigma acted trivially on D, the image
+zeta = image of X in D would satisfy zeta = zeta^{-1}, so zeta^2 = 1, so zeta = +/-1.
+zeta^2 = 1 with zeta in D of odd order (divides m) forces zeta = 1 (since 1
+is the only odd-order root of unity in Q_2, and D is unramified). But
+1 is not an eigenvalue of T on L (x) Q_2, so zeta != 1. Contradiction.
+So sigma is nontrivial involution of D.
+
+D / D_0 is then an unramified quadratic extension. Its residue
+extension k / k_0 (k finite of char 2 with |k| = |k_0|^2) is
+quadratic of separable residue fields.
+
+Trace hermitian form. The trace pairing of D over Z_2 is perfect:
+modulo 2, it reduces to the nondegenerate trace pairing of a finite
+separable field extension k / k_0 (separability ensures
+Tr_{k/k_0}: k -> k_0 not identically zero). So there is a unique
+D-valued form h with
+
+    Tr_{D / Z_2}( a  h(x, y) )  =  B( a x, y )   (a in D).    (4.5)
+
+The adjoint relation and symmetry of B give h(ax, by) = a sigma(b) h(x,
+y), h(y, x) = sigma(h(x, y)): h is hermitian, linear in first argument.
+h is unimodular (details: the B-dual of M_i equals the h-dual after
+projecting via the perfect trace pairing).
+
+Classification of unimodular hermitian forms over D/D_0 in residue
+char 2: every such form is determined by its RANK (rank-M_i). Reason:
+there is a vector of nonzero length (if all lengths vanished, then
+Tr(sigma(a) h(x, y)) = h(x+ay, x+ay) - h(x,x) - h(y,y)... by
+polarisation; the trace pairing of a finite separable extension is
+nondegenerate, so all sigma(a) h(x,y) vanish, so h(x,y) = 0, contradicting
+nonsingularity - unless k = k_0; but k != k_0). A lift of a vector of
+nonzero length has unit length u in D_0^x. The line splits off
+integrally by projection onto that vector. Induct, diagonalising h.
+
+Normalise the diagonal to 1. Every u in D_0^x is a norm from D^x
+because the residue norm k^x -> k_0^x is surjective. Do it
+successively: lift a norm preimage mod 2, improve mod 2^n to mod
+2^{n+1} using
+
+    N_{D/D_0}(1 + 2^n z)  =  1 + 2^n Tr_{D/D_0}(z)  (mod 2^{n+1}).
+
+Tr mod 2 is surjective (residue char 2, separable), so the correction
+exists; convergence. Rescale each diagonal vector to make its entry 1.
+The ranks of the M_i depend only on the module, so the form is
+uniquely determined by (rank_i). QED.
+
+Original: Section 4, Proposition 4.4 and its proof.
+
+Exercise 6 (Lemma 4.7, hermitian forms from their lengths). Why do
+we get a rational isometry of hermitian forms from a rational isometry
+of their length forms?
+
+Lemma 4.7. Let E / F be a quadratic field extension with char F
+!= 2, and let h_1, h_2 be nonsingular hermitian forms. If the
+quadratic forms
+
+    Q_{h_j}(x)  =  h_j(x, x)     (a real F-valued quadratic)   (4.6)
+
+on the underlying F-vector spaces are isometric, then h_1, h_2 are
+E-linearly isometric.
+
+Proof. Polar pairing of Q_h is (1/2) Tr_{E/F} h(x, y). If a vector x
+is orthogonal to all y with respect to this polar form, testing
+scalar multiples of y in the trace and using nondegenerate trace
+shows h(x, y) = 0 for every y. So Q_h is nonsingular. In particular,
+a nonzero hermitian space contains a vector of NONZERO LENGTH.
+
+Reflection / Witt cancellation. For a quadratic form Q with polar
+form b and a vector z of NONZERO length Q(z) != 0 in F (avoiding the
+char-2 subtleties by R), the reflection
+
+    rho_z(v)  =  v  -  ( 2 b(v, z) / Q(z) ) z                    (4.7)
+
+is an isometry of Q (direct expansion). Suppose x and e have equal
+nonzero length Q(x) = Q(e) = a != 0. Then Q(x - e) + Q(x + e) = 4 a
+!= 0, so at least one of x - e, x + e has nonzero length. If
+Q(x - e) != 0, reflection in x - e sends x to e; else Q(x + e) != 0,
+reflection in x + e sends x to -e, then reflection in e sends -e to
+e. So any isometry <a> perp V ~= <a> perp W matching displayed vectors
+restricts to an isometry V ~= W. This is Witt cancellation in the
+form we need.
+
+Induction. Write E = F(sqrt d) for d in F^x nonsquare. Take the first
+nonsingular hermitian line Ex inside h_1, choose a nonzero-length
+vector x with a = h_1(x, x) != 0 in F. Under an isometry of length
+forms Q_{h_1} -> Q_{h_2}, the image y has the same length a. The two
+lines Ex and Ey are hermitian-isometric (send x to y and use sigma-
+linearity), and their quadratic length forms are both
+
+    a N_{E/F}  in eigenvalues <a, -a d>.                        (4.8)
+
+Split off the hermitian lines Ex and Ey. Then cancel the two
+displayed quadratic lines <a> and <-ad> in the sense of (4.7).
+The complementary hermitian forms again have isometric length forms,
+and induction on E-dimension of the complementary hermitian space
+finishes. QED.
+
+Original: Section 4, Lemma 4.7 and its proof.
+
+Exercise 7 (Lemma 4.8, equivariant rational local-global principle).
+
+Lemma 4.8. Let (V_j, U_j, beta_j) be rational vector spaces (finite-
+dimensional over Q) with nonsingular symmetric forms beta_j preserved
+by finite-order-Odd operators U_j. If the two triples are
+equivariantly isometric over R and over Q_l for every prime l, then
+they are equivariantly isometric over Q.
+
+Proof. Take a common odd exponent for the operators (their orders
+divide m). The cyclotomic decomposition of each V_j is ORTHOGONAL for
+beta_j: the idempotents are unchanged under U_j -> U_j^{-1} (the
+adjoint involution), so the forms respect the decomposition.
+Corresponding components have equal dimensions invariant under any
+local isometry.
+
+Component where U_j has eigenvalue 1. This component carries an
+ordinary rational SYMMETRIC form. Local isometries give isometry over
+Q_l for each l, and by assumption also over R. Hasse-Minkowski
+(4.7 above) gives a rational isometry.
+
+Component with eigenvalue -1. Odd order excludes -1 (a real root of
+an odd-degree cyclotomic polynomial cannot be -1). So no such
+component.
+
+Component with primitive roots of unity of order d > 1. Put
+
+    E = Q(zeta_d),      F = Q(zeta_d + zeta_d^{-1}).
+
+The component is an E-vector space, with U_j acting as multiplication
+by zeta_d. Complex conjugation sigma is a nontrivial involution of E
+(nontrivial because d > 2; d > 1 and (Z/d)^x has an element of order
+> 1 acting as zeta_d -> zeta_d^{-1}). The trace construction gives a
+nonsingular E / F hermitian form h_j with
+
+    Tr_{E/Q}( a h_j(x, y) )  =  beta_j( a x, y )    (a in E).   (4.9)
+
+Local-to-global for hermitian forms. An equivariant isometry over a
+rational completion is linear over E tensor Q_l (or E tensor R) and
+preserves the trace identity (4.9), so it becomes a hermitian isometry
+for the extensions. Decomposing F tensor Q_l into its completion
+factors supplies an isometry at every place of F above l; decomposing
+F tensor R supplies every archimedean place. This covers finite places
+where E / F SPLITS (the trace pairing of the resulting quadratic etale
+algebra is still nondegenerate).
+
+So the nonsingular quadratic length forms Q_{h_1}, Q_{h_2} are
+isometric over every completion of F. Hasse-Minkowski over F gives a
+global quadratic isometry (Milne 2022 Theorem 25.64(c)). Lemma 4.7
+upgrades it to an E-linear hermitian isometry. Such an isometry
+preserves the rational symmetric form and intertwines U_j, giving
+the required equivariant rational isometry on this component.
+
+Combining cyclotomic components proves the claim. QED.
+
+Original: Section 4, Lemma 4.8 and its proof.
+
+Exercise 8 (Proposition 4.9, local and rational square isometries).
+
+Proposition 4.9. Write
+
+    A_j  =  (L_j, T_j, B_j) tensor (L_j, T_j, B_j)             (4.10)
+
+for the FIRST tensor square of the symmetrised data. Then there are
+
+    - an equivariant RATIONAL isometry
+        R : A_1 tensor Q -> A_2 tensor Q, and
+    - for every prime l, an equivariant isometry
+        D_l : A_1 tensor Z_l -> A_2 tensor Z_l,
+
+chosen so that, after extension to Q_l, D_l = R for ALL but finitely
+many primes l.
+
+Proof. Use the integral operator-module isomorphism of Corollary 3.4
+(Hertling-Mase) to transport the two forms B_1, B_2 to a COMMON module
+with a COMMON operator T. (Compatibility of the monodromy module
+isomorphisms is not required: any integral T-equivariant isomorphism
+works.)
+
+Odd primes. Proposition 4.3 applies to the common module L (x) Z_l with
+operator T (x) Z_l. It supplies an integral T-equivariant isometry
+between the two tensor square forms B_1 (x) B_1 and B_2 (x) B_2 over
+Z_l. This is D_l for odd l.
+
+Prime 2. Proposition 4.4 applies to the common module L (x) Z_2 with
+operator T (x) Z_2, because the prepared operators (the T_j) have odd
+order dividing m and no eigenvalue 1. It gives that B_1 and B_2 are
+T-equivariantly isometric over Z_2. Tensoring the isometry with itself
+gives B_1 (x) B_1 ~= B_2 (x) B_2 over Z_2. This is D_2.
+
+Archimedean. Lemma 4.1 (symmetrization) is compatible with the real
+Seifert form isometry from Corollary 3.4: symmetrizing an isometry
+of S_j through the (common) tensor sign yields an R-valued isometry
+of B_j, which tensors to a real isometry between A_1 (x) R and A_2 (x) R.
+This supplies the R-equivariant isometry.
+
+Rational. Lemma 4.8 (Hasse descent) with l ranging over all primes
+(including 2) and separately over R now gives R rational. Finally,
+the matrices of R and R^{-1} in integral lattice bases have
+denominators supported at finitely many primes (they are rational
+linear maps between Z-lattices). At every other prime l, R itself is
+already integral, and we take D_l = R. QED.
+
+Original: Section 4, Proposition 4.9 (also labelled local:rational,
+local:conclusion in the paper's sources).
+
+Exercise 9: What does Section D accomplish, and what still needs to
+be done?
+
+Accomplished. We have a rational equivariant isometry R of the first
+tensor squares A_1, A_2, and local integral isometries D_l of the
+same tensor squares, agreeing with R for all but finitely many l.
+
+But R and D_l are independent: at the finitely many "bad" primes, R^{-1}
+D_l is an R-rational isometry that need not be integral and may have
+nontrivial determinant on each eigenspace of T (x) T. The second tensor
+square (Section E) will kill these determinants by a symmetry coming
+from complex conjugation (using the real coefficients of q_1, q_2).
+Then strong approximation will choose one rational point that enjoys
+all local integrality at once.
+
+So Section D reduces the problem from "S_j congruence" to "compatible
+local data with free rational choice", and shifts the remaining
+difficulty into an arithmetic question about a special unitary group.
+This is precisely the setup strong approximation solves.
+
+Original: Section 4 (whole); Section 5.1 (opening paragraphs) for the
+statement of what remains.
+
+==============================================================
+SECTION E: ONE INTEGRAL SEIFERT-FORM ISOMETRY
+==============================================================
+
+Now the goal is to pass from the local and rational data of Section
+D to a single integral Seifert-form isometry S_1 ~= S_2 over Z. The
+plan has three parts:
+
+  (1) Use the REAL coefficients of q_1, q_2 (complex conjugation) to
+      build an integral involution C_j that reverses M_j. Take a
+      SECOND tensor square with C_j^(x)2 built in to make the rational
+      vs. local determinant discrepancy disappear.
+  (2) Adjoin two COMMON factors (six cubes and a pair of pure powers
+      z_7^11 + z_8^13) chosen so that (a) all eigenspaces of the
+      resulting form are INDEFINITE (needed for real-place
+      noncompactness in strong approximation), and (b) no eigenvalue
+      of the resulting operator is real (so the special unitary
+      reduction is clean).
+  (3) Apply strong approximation to the unitary isometry group,
+      obtaining one rational point that simultaneously realizes all
+      local lattice isometries.
+
+Throughout, "symmetric data" means (L, T, B) with B integral,
+unimodular, symmetric, T-invariant.
+
+Exercise 1 (Lemma 5.1, integral reversers). Why does complex
+conjugation give an integral reversing involution on the Milnor
+lattice, and what does it do to the symmetric form B?
+
+Lemma 5.1. There is an integral involution C_j of L_j with
+
+    C_j^{-1} T_j C_j  =  T_j^{-1},                             (5.1)
+    (C_j)^t B_j C_j  =  B_j.                                   (5.2)
+
+Proof. Complex conjugation of coordinates (z_1,...,z_s) -> (conj z_1,
+..., conj z_s) preserves the origin, preserves the sphere S^(2s-1)_eps,
+and exchanges the two normal directions of a positive and negatively
+oriented Milnor page. So it induces an involution C_j on middle
+homology L_j = H_{s-1}(F_j), integral, involutive (C_j^2 = 1).
+
+The effect on the Seifert form is the integral congruence
+
+    C_j^t S_j C_j  =  S_j^t.                                    (*)
+
+Milnor 1968 (and Section 5 of the source paper) derives this from the
+linking conventions: complex conjugation is degree 1 on the ambient
+(2s-1)-sphere (s even); the linking pairing changes sign under
+interchange of its two arguments (s-1 odd); and the positive and
+negative page push-offs y^+ = -y^- cancel these signs in the linking
+computation. The full calculation is Milnor 1968 Section 8 and the
+paper's Section 5.
+
+Taking monodromy of (*) and using T_j = S_j^{-1} S_j^t gives
+
+    C_j^{-1} T_j C_j  =  T_j^{-1}.                              (**)
+
+And then
+
+    C_j^t B_j C_j  =  C_j^t S_j T_j^r C_j
+                   =  (C_j^t S_j C_j) (C_j^{-1} T_j C_j)^r
+                   =  S_j^t T_j^{-r}
+                   =  (S_j T_j) T_j^{-r}
+                   =  S_j T_j^{1-r}
+                   =  S_j T_j^r    (since 1 - r = r mod m)
+                   =  B_j.
+So C_j preserves B_j as well as reversing T_j. QED.
+
+Note. The point is that the result is THEOREMATIC: an integral
+involution reversing the operator exists, because the singularity has
+real coefficients.
+
+Original: Section 5, Lemma 5.1 and its proof.
+
+Exercise 2: Build the second tensor square with the reversers in.
+
+Put
+    C_j^{(2)}  =  C_j tensor C_j,       an involution of  L_j (x) L_j  =  A_j.
+Then C_j^{(2)} preserves b_j = B_j (x) B_j (because each factor of C_j
+preserves B_j) and reverses tau_j = T_j (x) T_j (because each factor of
+C_j reverses T_j). So C_j^{(2)} is a "reverser" for the tensor square.
+
+Define, on the FOURTH tensor power (A_j)^{(x)2} = L_j^{(x)4}:
+
+    \widehat R   =  R  (x)  ( C_2^{(2)} R (C_1^{(2)})^{-1} ),              (5.3)
+    \widehat D_l =  D_l  (x)  ( C_2^{(2)} D_l (C_1^{(2)})^{-1} ).          (5.4)
+
+Both are defined on A_1^{(x)2} = L_1^{(x)4} and land in A_2^{(x)2} = L_2^{(x)4}.
+
+Claim: \widehat R is a rational equivariant isometry (commuting with
+\tau_1 (x) \tau_1), and \widehat D_l is an equivariant isometry of the
+FOURTH-power lattices over Z_l.
+
+Reason: each of the two tensor factors of \widehat R (and of
+\widehat D_l) is equivariant. For instance \widehat R's second factor
+C_2^{(2)} R (C_1^{(2)})^{-1} intertwines tau_1 with tau_2:
+    (C_2^{-1} R C_1) tau_1 (C_2^{-1} R C_1)^{-1}
+    = C_2^{-1} R tau_1 R^{-1} C_2
+    = C_2^{-1} tau_2 C_2
+    = tau_2^{-1},
+which is what we want (R intertwines, C reverses). Both factors also
+preserve the symmetric forms. So the products do too.
+
+Original: Section 5, Equations 5.1-5.2 (also labelled (5.3)-(5.4) in
+some drafts).
+
+Exercise 3 (Lemma 5.2, eigenspace determinant cancellation). State it
+and prove it.
+
+Lemma 5.2. For every prime l, the ratio \widehat R^{-1} \widehat D_l
+has DETERMINANT 1 on every eigenspace of tau_1 (x) tau_1 over the
+algebraic closure of Q_l. The same remains true after tensoring both
+maps with identity on any COMMON finite-order operator data.
+
+Proof. Put
+    a  =  R^{-1} D_l,        a rational Q_l-isometry of tau_1-equivariant
+                              (hence also of tau_1 (x) tau_1 = (T_1 (x) T_1)^2)
+    C  =  C_1^{(2)},         integral reverser of tau_1
+                              (hence of tau_1 (x) tau_1).
+Then multiplying the two maps in (5.3)-(5.4) gives the EXACT identity
+
+    \widehat R^{-1} \widehat D_l  =  a  (x)  ( C a C^{-1} ).       (5.5)
+
+In particular NO compatibility of R with C_1, C_2 is required.
+
+Let W_gamma be the gamma-eigenspace of tau_1 over \overline{Q_l}, and
+set
+    m_gamma  =  dim W_gamma,
+    t_gamma  =  det( a | W_gamma ).
+
+Invariance and nondegeneracy of B_1 (the form on tau_1) give a perfect
+pairing between W_gamma and W_{gamma^{-1}}: the form B_1 is nondegenerate
+and invariant, so B_1(W_gamma, W_delta) = 0 unless gamma delta = 1.
+Since a is an isometry, a preserves this pairing, and we get
+
+    m_{gamma^{-1}}  =  m_gamma,
+    t_{gamma^{-1}}  =  t_gamma^{-1}.                             (5.6)
+
+For gamma = 1, this says t_1^2 = 1. No choice of sign is needed: we
+never need to know what the sign is.
+
+The reverser C = C_1^{(2)} carries W_{eta^{-1}} to W_eta. So the
+determinant of C a C^{-1} on W_eta is t_{eta^{-1}}.
+
+For each eigenvalue lambda of tau_1 (x) tau_1, its eigenspace in the
+tensor square is
+
+    direct sum_{gamma * eta = lambda}  W_gamma (x) W_eta,
+
+where (gamma, eta) ranges over ordered pairs of occurring eigenvalues
+whose product is lambda. By the standard determinant formula
+det(A (x) B) = det(A)^{dim B} det(B)^{dim A}, the determinant of
+(5.5) on this eigenspace is
+
+    prod_{gamma eta = lambda}   t_gamma^{m_eta}  t_{eta^{-1}}^{m_gamma}
+    =  prod_{gamma eta = lambda}  t_gamma^{m_eta}
+       /  prod_{gamma eta = lambda}  t_eta^{m_gamma}
+    =  1,
+
+where the last equality exchanges gamma and eta in the denominator. This
+holds INCLUDING diagonal pairs (gamma = eta) and all coincidences among
+product eigenvalues lambda. So determinant 1.
+
+Finally, tensoring with an identity map on any common finite-order
+operator data makes the determinant on each NEW eigenspace a product of
+POWERS of the already-1 determinants. So determinant 1 is preserved. QED.
+
+Original: Section 5, Lemma 5.2 and its proof.
+
+Exercise 4: Adjoining the two common summands (the "six cubes" and the
+"11 and 13 powers"). What are they, and why?
+
+Use new disjoint variables to form
+
+    u  =  z_1^3 + z_2^3 + z_3^3 + z_4^3 + z_5^3 + z_6^3,     (5.7)
+    v  =  z_7^11 + z_8^13.                                     (5.8)
+
+Both u and v have an EVEN number of variables (6 and 2 respectively),
+and both have odd-order monodromy (3 for u, 11*13 = 143 for v).
+
+Let (U, T_U, B_U) and (V, T_V, B_V) be their symmetric-form data, and
+define the FINAL data
+
+    H_j  =  (Lambda_j, Theta_j, beta_j)
+         =  A_j^{(x)2}  (x)  U  (x)  V.                              (5.9)
+
+So a "new" tensor product over the second tensor square A_j^{(x)2} with
+the common factors U and V. Extend the maps \widehat R and \widehat D_l
+by identity on U (x) V, keeping the same letters.
+
+Two roles of the common factors:
+  (a) SIX CUBES make every eigenspace form INDEFINITE (signature has
+      both positive and negative parts). Need for noncompactness at
+      real places in the SU(p,q) strong approximation setup.
+  (b) DEGREES 11 and 13 exclude real (in particular +1 and -1) from
+      the final operator's eigenvalues. EVEN THOUGH the second tensor
+      square can acquire eigenvalue 1, the primes 11 and 13 cannot
+      cancel against an earlier eigenvalue. This forces every final
+      eigenvalue's order to be divisible by 11 * 13, hence in
+      particular not a real number +/-1. Also relevant: the topo-
+      logical Section 6 uses these two primes to prove det(1 - T) = 1,
+      which makes the LINKS homotopy spheres.
+
+Original: Section 5, Equations 5.6 (or 5.7) and surrounding text.
+
+Exercise 5 (Lemma 5.3, spectral properties of the added factors). State
+it and give the signatures.
+
+Lemma 5.3. Every eigenvalue of Theta_j has odd order divisible by both
+11 and 13. On every complex eigenspace of Theta_j, the hermitian form
+
+    (x, y)  ->  beta_j( x, conj y )                              (5.10)
+
+is NONdegenerate and INDEFINITE. The complex dimension of every
+eigenspace is at least 21.
+
+Proof. Order. Before adjoining V, the operator order divides
+3^a * 5 * 7 for some a (by the preparation in Section C: m =
+3^12 * 5 * 7). The eigenvalues of V are
+
+    zeta_11^b  zeta_13^c,   1 <= b <= 10,  1 <= c <= 12,
+
+where zeta_11, zeta_13 are primitive 11th and 13th roots of unity. The
+nontrivial 11- and 13-primary components cannot cancel against an eigen-
+value of the earlier tensor factors (which live in a group of order
+3^a * 5 * 7, coprime to 11 and 13). So every final order is divisible
+by 11 and by 13, and remains odd.
+
+Indefiniteness. We compute the signatures of the six-cube symmetric
+form. Let
+
+    A  =  ( 1  -1 )
+          ( 0   1 )
+
+A one-cube Seifert matrix is S_c = -A (per Sakamoto 1974, Corollary 3,
+in the paper's convention). Its chosen monodromy is
+
+    t  =  - A^{-1} A^t  =  ( 0  -1 )
+                            ( 1  -1 )
+
+(the minus sign is the one-variable convention). The alternating form
+
+    J  =  A t^2  =  ( 0   1 )
+                     ( -1  0 )
+
+is t-invariant. Iterating Thom-Sebastiani gives S_u = (-1)^{15}
+(-A)^{(x)6} = -A^{(x)6}. The sixfold monodromy is t^{(x)6}, and since
+2 * 2 = 1 mod 3, its symmetrised form is
+
+    B_u  =  S_u (t^{(x)6})^2  =  - J^{(x)6}.
+
+We compute signatures of J^{(x)6} below; B_u has OPPOSITE sign in each.
+
+Diagonalise. Set omega = e^{2 pi i / 3}, e_+ = (1, -omega),
+e_- = (1, -omega^2). These are t-column eigenvectors with eigenvalues
+omega and omega^2 respectively, and
+
+    J(e_+, conj e_+)   =  i sqrt 3,
+    J(e_-, conj e_-)   =  - i sqrt 3,
+    J(e_+, conj e_-)   =  0.
+
+In the resulting orthogonal tensor basis (over C), a vector with k
+copies of e_+ has eigenvalue omega^{-k} (i.e., omega^{6-k}) and length
+27 (-1)^{k+1}; there are C(6, k) such vectors. Summed over k, we get
+
+    Eigenvalue   k             Signature          Dimension
+    ---------------------------------------------------------
+    1            k = 0, 3, 6    (20, 2)            22
+    omega        k = 2, 5       (6, 15)            21
+    omega^2      k = 1, 4       (6, 15)            21
+
+Each eigenspace of this factor alone is INDEFINITE: both signs occur.
+(Passing from J^{(x)6} to B_u = -J^{(x)6} reverses every sign but keeps
+indefiniteness.)
+
+General principle for symmetry: for any real nondegenerate symmetric
+form b invariant under a finite-order operator, its C-bilinear extension
+pairs the gamma-eigenspace perfectly with the gamma^{-1}-eigenspace.
+So b(x, conj y) restricts to a NONdegenerate hermitian form on each
+eigenspace. Distinct eigenspaces are orthogonal.
+
+Apply to each factor in (5.9). A final eigenspace is the orthogonal
+direct sum of tensor products of factor eigenspaces (respecting
+eigenvalues). Each NONZERO summand contains one of the indefinite
+six-cube spaces in the table. Diagonalising the other two hermitian
+forms, we see that its tensor form has BOTH signs: multiply any nonzero
+diagonal entry from the other factors by one positive and one negative
+entry from the six-cube factor. Union over summands is still
+indefinite, and its dimension is at least 21 (dimension of omega and
+omega^2 eigenspaces of the six-cube factor). QED for Lemma 5.3.
+
+Original: Section 5, Lemma 5.3 and its proof; Sakamoto 1974,
+Corollary 3.
+
+Exercise 6: The special unitary group G and the strong-approximation
+setup. What is G, and how do we split it as a product?
+
+Set V = Lambda_1 (x) Q, and let Theta_1 act on V. Since Theta_1 has
+finite order, its minimal polynomial is a product of DISTINCT
+cyclotomic polynomials (semisimplicity). Decompose V by primary
+decomposition into cyclotomic components:
+
+    V  =  direct sum_{d in D}  V_d,
+    E_d  =  Q(zeta_d),     F_d  =  Q(zeta_d + zeta_d^{-1}),
+    n_d  =  dim_{E_d} V_d,
+
+for some finite set D of cyclotomic indices d > 1 (d = 1 would be
+eigenvalue 1, excluded in Lemma 5.3; d = 2 is eigenvalue -1, also
+excluded).
+
+G = the algebraic Q-group of SELF-ISOMETRIES of (V, Theta_1, beta_1)
+with determinant 1 on every eigenspace over an algebraic closure. This
+is a Q-defined subgroup of GL(V).
+
+Lemma 5.3 implies all d in D are odd and divisible by 11 * 13. So
+E_d/F_d is a quadratic extension of a totally real field F_d, with
+complex conjugation sigma as its nontrivial involution.
+
+The trace construction of Section D (equation 4.5) gives an F_d-
+hermitian form h_d on V_d (E_d-valued, with the standard hermitian
+conventions). Commuting with Theta_1 on V_d means being E_d-LINEAR;
+preserving beta_1 means preserving the trace hermitian form h_d (by the
+perfect trace pairing); the determinant conditions on all conjugates
+of d amount to being in SU(h_d) (special unitary), not the full U(h_d).
+So
+
+    G  =  prod_{d in D}  Res_{F_d/Q}  SU(h_d).                  (5.11)
+
+Why G is as needed. For each d in D:
+    - SU(h_d) is connected, simply connected, absolutely almost
+      simple over F_d (Milne 2022, Theorem 24.44, Remark 24.46,
+      Summary 24.66). Dimension at least 2 (since n_d >= 21 with
+      dimension over E_d, corresponding to complex dimension
+      n_d >= 21 in the local structure).
+    - At each real place of F_d, the hermitian form h_d has real
+      signature (p, q) with p, q > 0 (because Lemma 5.3's indefinite
+      form is exactly the complexification at that real place;
+      n_d >= 21 ensures both parts occur), giving SU(p,q) noncompact.
+
+So G satisfies the strong-approximation hypotheses.
+
+Original: Section 5, "Special unitary group and strong approximation"
+subsection, especially equation for G.
+
+Exercise 7 (Proposition 5.4, integral equivariant isometry).
+
+Proposition 5.4. The triples H_1 and H_2 in (5.9) are ISOMETRIC over
+Z: there is a lattice isomorphism
+
+    U : Lambda_1 -> Lambda_2
+
+satisfying
+
+    U Theta_1  =  Theta_2 U,
+    U^t beta_2 U  =  beta_1.                                  (5.12)
+
+Proof. We have already described G and its decomposition (5.11).
+
+Step 1. Unpack the density statement. Apply strong approximation
+(Theorem 2.?, Section B) to each SU(h_d) over F_d, with all archimedean
+places omitted, obtaining density of SU(h_d)(F_d) in its finite adelic
+group (Platonov 1969, 1970; Rapinchuk 2014). Taking the product over d
+and using restriction of scalars gives
+
+    closure of G(Q)  in G(A_{Q,f})  =  G(A_{Q,f}).             (5.13)
+
+Here A_{Q,f} is the finite adele ring of Q, and G(A_{Q,f}) is the
+restricted product of G(Q_l) over finite primes l, restricted with
+respect to integral compact open subgroups at all but finitely many l.
+
+Step 2. Construct the adelic point a. For each prime l, set
+
+    a_l  :=  ( \widehat R^{-1} \widehat D_l )( l ),           (5.14)
+
+Viewing \widehat R^{-1} \widehat D_l as a Q_l-linear isometry of
+(Theta_1-pair) with (Theta_2-pair). Lemma 5.2 gives, on each
+eigenspace of Theta_1, determinant 1 for a_l. So
+
+    a_l  in  G(Q_l).
+
+And since \widehat R = \widehat D_l (i.e., D_l = R) for all but
+finitely many l (Prop 4.9), we have a_l = 1 for all but finitely many l.
+Thus a = (a_l) defines a WELL-DEFINED POINT of the finite adelic group
+G(A_{Q,f}).
+
+Step 3. Choose the lattice stabiliser. For each prime l set
+
+    Lambda_{j, l}  =  Lambda_j tensor Z_l,
+    K_l           :=  { g in G(Q_l) : g Lambda_{1, l} = Lambda_{1, l} },
+    K             :=  prod_l K_l.                             (5.15)
+
+Fact (standard). In a basis of the actual lattice Lambda_1,
+    K_l  =  G(Q_l)  intersect  GL_{rank Lambda_1}(Z_l),
+compact and open in G(Q_l). The point is that K_l is the Z_l-points
+of a schematic integral model of G inside GL(Lambda_1), and this model
+has G(A_{Q,f})-restricted-product-compatible integral subgroups at all
+but finitely many primes. Thus K is compact and open in G(A_{Q,f}).
+
+Side note: this integral model uses the FULL lattice Lambda_1 (no
+integral splitting of cyclotomic components is needed). Because we
+are taking stabilisers of the FULL lattice, not just of individual
+summands, the resulting K_l really are the integral points of the
+group G of determinant-1 eigenspace isometries.
+
+Step 4. Find a rational point. By the density (5.13), the NONEMPTY
+OPEN COSET
+
+    a K
+
+in G(A_{Q,f}) contains a rational point g in G(Q). Write g = a k with
+k in K, so for each prime l we have g = a_l k_l with k_l in K_l.
+
+Step 5. Conclude. For every prime l:
+
+    ( \widehat R g ) Lambda_{1, l}
+    =  \widehat R ( a_l k_l ) Lambda_{1, l}
+    =  \widehat R a_l Lambda_{1, l}     [ because k_l in K_l stabilises Lambda_1 (x) Z_l]
+    =  \widehat D_l Lambda_{1, l}       [ because \widehat R a_l = \widehat D_l]
+    =  Lambda_{2, l}.                   [ because \widehat D_l is an integral isometry]
+
+So \widehat R g is an isomorphism Lambda_1 (x) Q -> Lambda_2 (x) Q
+carrying Lambda_1 (x) Z_l to Lambda_2 (x) Z_l for EVERY prime l.
+
+Step 6. Assemble the integral lattice. A full integral lattice Lambda
+satisfies the "reconstruction" identity
+
+    Lambda  =  { x in Lambda tensor Q : x in Lambda tensor Z_l for all l }.  (5.16)
+
+(In a lattice basis, this is the elementary identity Q intersect
+intersection_l Z_l = Z, applied to each coordinate.)
+
+Apply this to \widehat R g and its inverse. Since \widehat R g carries
+Lambda_1 (x) Z_l into Lambda_2 (x) Z_l for every l, and the inverse carries
+Lambda_2 (x) Z_l into Lambda_1 (x) Z_l for every l, the identity (5.16)
+implies
+
+    \widehat R g (Lambda_1)  =  Lambda_2,                     (5.17)
+
+so \widehat R g is an integral isomorphism.
+
+Step 7. It is an equivariant isometry. Both \widehat R and g preserve
+the symmetric form beta_1 and Theta_1-equivariance, so their product
+does. Setting U := \widehat R g gives (5.12). QED.
+
+Original: Section 5, Proposition 5.4 and its proof.
+
+Exercise 8 (Remark: recover the integral Seifert-form isometry from the
+symmetric data).
+
+Remark 5.1. The proposition gives an equivariant isometry of the
+SYMMETRIC data. To recover from it an isometry of the ORIGINAL Seifert
+data, choose r in Z with 2r = 1 modulo a common odd exponent of the
+final operators. Put
+
+    S'_j  =  beta_j Theta_j^{-r}.                              (5.18)
+
+Then S'_j is the symmetrised form of the original Seifert data,
+compatible with the tensor-product convention (up to an overall tensor
+sign that is the SAME on both sides). The isomorphism U from Prop 5.4
+satisfies
+
+    U^t S'_2 U  =  U^t ( beta_2 Theta_2^{-r} ) ( ... )
+                =  (U^t beta_2 U) Theta_1^{-r}                 [ because U Theta_1 = Theta_2 U]
+                =  beta_1 Theta_1^{-r}
+                =  S'_1.                                        (5.19)
+
+The tensor-product sign convention can multiply both final forms by
+the same sign; this does not alter (5.19). So U is an integral
+Seifert-form isometry S'_1 ~= S'_2.
+
+Original: Section 5, Remark at the end (also labelled
+global:recover-seifert).
+
+Exercise 9: What does Section E accomplish?
+
+Section E has now given us ONE integral map U carrying the
+symmetric form B_1 to B_2 while intertwining the monodromy. Recovering
+the Seifert form from symmetric data (Remark 5.1), U is an integral
+Seifert-form isometry S_1 ~= S_2. This is the input for the topological
+step (Section F), where Levine's classification turns this integral
+Seifert equivalence into an ambiently-equivalent embedded-knot pair and
+then into an ambient homeomorphism of hypersurface germs.
+
+Original: Section 5, closing of the section.
+
+==============================================================
+SECTION F: TOPOLOGY -- FROM SEIFERT FORMS TO AMBIENT GERMS
+==============================================================
+
+Now we turn the algebraic isometry S_1 ~= S_2 (from Section E) into an
+ambient homeomorphism of hypersurface germs.
+
+Exercise 1: The final polynomials f_1 and f_2. How are they assembled,
+and how many variables does the ambient space have?
+
+For j in {1, 2}, put
+
+    f_j  =  q_j^[1] + q_j^[2] + q_j^[3] + q_j^[4]
+            +  z_1^3 + z_2^3 + ... + z_6^3
+            +  x^11 + y^13.                                    (6.1)
+
+The notation q_j^[a] means "an independent copy of q_j in its own set
+of variables" (four independent copies for a = 1, 2, 3, 4). The
+displayed cubes, x^11 and y^13 are also in their own new variables,
+shared between f_1 and f_2.
+
+Variable count. Each q_j has s = s_0 + 2 variables (Section C,
+Corollary 3.4). Four copies contribute 4 s = 4(s_0 + 2) = 4 s_0 + 8
+variables. Six cubes and x, y contribute 6 + 2 = 8 more. Total:
+
+    N  =  4(s_0 + 2) + 6 + 2  =  4 s_0 + 16.                   (6.2)
+
+Since s_0 is even (Section C), 4 s_0 is divisible by 8, so N = 0 mod
+8. Also N > 3.
+
+Weights. Give each factor of (6.1) its own positive weights; the
+resulting f_1, f_2 are weighted homogeneous for positive rational
+weights (each factor is; the direct sum inherits the weights).
+
+Milnor data. Let (L_j^f, T_j^f, S_j^f) be the Milnor lattice, chosen
+monodromy, and integral Seifert form of f_j. The convention (2.1) reads
+
+    T_j^f  =  (S_j^f)^{-1} (S_j^f)^t.
+
+All monodromy orders of interest are ODD (as we saw, of odd order
+divisible by 11 * 13 after including the extra factors).
+
+Choosing r. Pick a common odd multiple m_f of all these orders, and an
+integer r with 2 r = 1 mod m_f. Then the SYMMETRIC form associated with
+S_j^f (in even ambient dimension N) is
+
+    B_j^f  =  S_j^f (T_j^f)^r.
+
+The pair (B_j^f, T_j^f) recovers S_j^f by B_j^f (T_j^f)^{-r}.
+
+Comparison with Section E. The integral Thom-Sebastiani-Sakamoto
+formula identifies the operator-form data (L_j^f, T_j^f, B_j^f) with
+the H_j data of Proposition 5.4, up to the SAME overall tensor sign on
+both sides. Concretely, every factor in (6.1), grouped as four copies of
+q_j, the sum of six cubes, and x^11 + y^13, has even ambient dimension
+(4 s_0 for the four q_j copies together; 6 for the six cubes; 2 for
+x, y). So its symmetrization commutes with the indicated tensor
+products, and the Thom-Sebastiani signs (-1)^{ab} depend only on these
+common dimensions; hence they are the same on both sides and cancel.
+
+Consequently Prop 5.4 gives an integral isomorphism
+
+    Psi : L_1^f -> L_2^f
+
+with
+
+    Psi T_1^f  =  T_2^f Psi,
+    Psi^t B_2^f Psi  =  B_1^f.                               (6.3)
+
+Seifert form recovered. Compute:
+
+    Psi^t S_2^f Psi  =  Psi^t B_2^f (T_2^f)^{-r} Psi
+                     =  Psi^t B_2^f Psi (T_1^f)^{-r}
+                     =  B_1^f (T_1^f)^{-r}
+                     =  S_1^f.                                (6.4)
+
+(Using Psi T_1^f = T_2^f Psi, so Psi (T_1^f)^{-r} = (T_2^f)^{-r}
+Psi.) So Psi is a simultaneous integral isometry of Seifert forms
+and monodromies. This is exactly the input for the topological
+classification.
+
+Original: Section 6 "Final polynomials and their Seifert forms"
+subsection; equations (6.1)-(6.4).
+
+Exercise 2 (Lemma 6.1, analytic properties). State it and prove it.
+
+Lemma 6.1. The polynomials f_1, f_2 in (6.1) have isolated critical
+points at the origin, are reduced both as polynomials and as analytic
+germs, and their orders are
+
+    ord_0(f_1) = 2,     ord_0(f_2) = 3.                       (6.5)
+
+Moreover, the Hessian ranks are:
+
+    rank Hess_0(f_1) = 16 E,     rank Hess_0(f_2) = 0.        (6.6)
+
+Both ranks are EVEN.
+
+Proof of isolatedness. Let us argue by factoring. Lemma 3.2 (chain
+completion) shows that each chain singularity is isolated at the
+origin only (no critical point away from 0). Pure powers have the same
+"only the origin" property. A SUM IN DISJOINT VARIABLES has critical
+locus equal to the product of critical loci: the gradient equation
+partial_{z_j i} f (z) = 0 splits into independent equations for each
+factor, and each factor's "0" means the corresponding block of
+coordinates is 0. Since each factor's critical locus is {0}, their
+product is {0}.
+
+Away from the origin but inside one block. Consider a nonzero point in
+a specific factor's variables, other factors set to zero. By Lemma 3.2
+(and by the "no critical point of a pure power away from 0" fact for
+the pure power blocks), at least one partial derivative in that
+factor's block is nonzero. So this point is not critical. Combining
+these cases, the critical locus of each f_j is exactly {0}.
+
+Orders. Each q_j has order 2 or 3 (Corollary 3.4). Adding cubes and
+pure powers (5, 7, 11, 13) does not change the lowest-order term; four
+independent copies of q_j have the same order as q_j (there is nothing
+to cancel across disjoint variables). So ord_0(f_1) = ord_0(q_1) = 2,
+ord_0(f_2) = ord_0(q_2) = 3.
+
+Reducedness (analytic). Suppose an irreducible nonunit analytic germ a
+occurred with multiplicity at least 2 in f_j, so we can write f_j = a^2
+b near 0. Then every FIRST derivative of f_j vanishes identically on
+V(a) = {a = 0}. But a nonunit hypersurface germ in C^N has dimension
+N - 1 > 0 (positive-dimensional), contradicting the ISOLATED critical
+point of f_j at the origin. So a is not repeated. f_j is reduced as an
+analytic germ.
+
+Reducedness (polynomial). Suppose f_j = h^2 * w for nonconstant
+polynomials h, w. Then the gradient of f_j vanishes identically on the
+AFFINE variety V(h) = {h = 0}. V(h) in C^N has dimension N - 1 > 0,
+contradicting (from the isolation argument) the only critical point
+being 0. So f_j is reduced as a polynomial.
+
+Hessian ranks. Each q_j is built from 4E squares of independent
+variables and everything else of order >= 3. Four independent copies of
+q_1 contribute 4 * 4E = 16E squares. So the quadratic part of f_1 is a
+sum of 16E independent squares (in specially chosen coordinates) and
+0 squares in order >= 3 terms; hence rank Hess_0(f_1) = 16E, corank
+= N - 16E. The polynomial q_2 (and hence f_2) has order 3: no squares
+at all in the four copies. So Hess_0(f_2) = 0, corank = N. Both ranks are
+even (16E is divisible by 16; 0 is even). QED.
+
+Original: Section 6, Lemma 6.1 and its proof (also see Lemma 3.2 for
+the isolation argument reused).
+
+Exercise 3 (Lemma 6.2, the links are homotopy spheres and simple
+knots). State it and explain the key determinant step.
+
+Let
+
+    K_j  =  V(f_j) intersect S^(2N-1)_eps,
+    X_j  =  S^(2N-1)_eps \ K_j,     n = N - 1.
+
+So K_j is a smooth closed manifold of dimension 2n - 1 = 2N - 3.
+
+Lemma 6.2. Each K_j is a homotopy sphere, and each embedding
+K_j subset of S^(2n + 1)_eps is a SIMPLE knot (as defined in Section B).
+
+Key determinant step. Every eigenvalue of T_j^f has the form
+
+    xi . zeta_11^a . zeta_13^b,     1 <= a <= 10,  1 <= b <= 12,   (6.7)
+
+where zeta_11, zeta_13 are primitive 11th and 13th roots of unity, and
+xi has order dividing 3^A * 5 * 7 for some integer A >= 1. (This is
+from Lemma 5.3, transported across the tensor product data of (6.1).)
+
+The factors of orders 11 and 13 CANNOT cancel against each other or
+against xi (coprime). So every eigenvalue has order DIVISIBLE by both
+11 and 13.
+
+Consequences for the characteristic polynomial:
+    Every eigenvalue is a 143k-th root of unity for some integer k, so
+    the characteristic polynomial of T_j^f is a product of cyclotomic
+    polynomials Phi_d with every occurring d divisible by 143.
+
+    For d > 1, Phi_d(1) is:
+        p       if d = p^e is a power of the prime p,
+        1       if d has at least two distinct prime factors.
+    (Follows by induction from d = prod_{e | d, e > 1} Phi_e(1); see
+    Section 6, for the elementary inductive proof.)
+
+    So Phi_d(1) != 0 for all d > 1. All occurring d are > 1 (divisible
+    by 143).
+
+    Hence
+        det(1 - T_j^f)  =  prod_d  Phi_d(1)^{mult_d}  =  1.   (6.8)
+    (Each Phi_d(1) is 1 or a positive integer prime; but MULTIPLICITIES
+    multiply 1's, and primes p with p | d are already accounted for in
+    d's prime-power divisors, leaving Phi_d(1) = 1 for d with two+
+    distinct prime factors. A careful reading of the paper shows
+    the product equals 1; the paper's identity is exact.)
+
+In particular 1 - T_j^f is an invertible operator on middle homology
+of the Milnor fiber F_j.
+
+Wang sequence / conclusion. The geometric monodromy acts as
+(T_j^f)^{-1} in the convention (2.1). The identity
+
+    1 - (T_j^f)^{-1}  =  -(T_j^f)^{-1} (1 - T_j^f)              (6.9)
+
+shows the geometric monodromy's difference from identity is also
+invertible over Z. The WANG exact sequence of the Milnor fibration
+X_j -> S^1 then gives
+
+    H_i(X_j; Z)  =  Z  for i = 0, 1
+    H_i(X_j; Z)  =  0  for i >= 2.                             (6.10)
+
+The only possible additional groups are the kernel and cokernel of
+(1 - T_j^f) on H_n(F_j; Z), and (6.8) forces both to vanish.
+
+Alexander duality (Hatcher 2002, Corollary 3.45) then makes K_j an
+integrally-homology (2n - 1)-sphere. Since K_j is (n - 2)-connected
+(from the Milnor theorem), it is simply connected. Hurewicz and
+Whitehead (Hatcher 2002, Theorems 4.32, 4.33) then give that K_j is a
+HOMOTOPY sphere. Since 2n - 1 = 2N - 3 >= 5, Smale's high-dimensional
+Poincare theorem makes it homeomorphic to the standard sphere
+S^(2n-1).
+
+Simplicity. The homotopy exact sequence of the fibration
+int F_j -> X_j -> S^1, together with (n - 1)-connectivity of int F_j,
+gives
+
+    pi_1(X_j)  =  Z,     pi_i(X_j)  =  0  (2 <= i <= n - 1).  (6.11)
+
+This is precisely the simplicity condition for a (2n - 1)-knot.
+QED.
+
+Original: Section 6, Lemma 6.2 and its proof; Hatcher 2002, Corollary
+3.45, Theorems 4.32, 4.33.
+
+Exercise 4: How does Levine's classification + weighted radial
+coordinates give an ambient homeomorphism?
+
+Orientation. Orient the links K_1, K_2 and their Seifert surfaces by
+the Milnor fibration, using the SAME positive push-off convention on
+both sides. Equation (6.4) (= Psi^t S_2^f Psi = S_1^f) then gives:
+
+    The integral Seifert matrices of K_1 and K_2 are CONGRUENT
+    (equivalent over Z).
+
+Levine's theorem. Apply Section B's Levine statement with q = n >= 3.
+Both K_j are simple knots (Lemma 6.2). Their integral Seifert matrices
+are congruent. So they are AMBIENTLY ISOTOPIC in S^(2N-1)_eps. Smooth
+isotopy extension then gives an ORIENTATION-PRESERVING diffeomorphism
+of pairs
+
+    phi : (S^(2N-1)_eps, K_1)  ->  (S^(2N-1)_eps, K_2).         (6.12)
+
+(Levine's definition allows smooth homotopy spheres that are not
+standard smooth spheres; no standard-smooth-sphere assertion is
+needed.)
+
+Weighted radial coordinates. For a positive weight vector
+w = (w_1, ..., w_N), define
+
+    a_w(t, x)  =  ( t^(w_1) x_1, ..., t^(w_N) x_N ),  t > 0.   (6.13)
+
+For any nonzero x, the function t -> ||a_w(t, x)||^2 = sum_i t^(2 w_i)
+|x_i|^2 is strictly increasing on (0, inf) with limits 0 (as t -> 0)
+and inf (as t -> inf). So each weighted orbit
+t -> a_w(t, x) meets S^(2N-1)_eps in EXACTLY one point. Then
+
+    A_w : C(S^(2N-1)_eps)  ->  closed ball B^(2N)_eps,
+          [x, t]  ->  a_w(t, x),   0 <= t <= 1,                 (6.14)
+
+is a continuous bijection from the cone C(S^(2N-1)_eps) (where all
+(x, 0) are identified to one point) onto the closed ball. Continuity
+at the cone vertex follows from ||a_w(t, x)|| <= eps * t^{min_i w_i}
+(so as t -> 0 uniformly, image -> 0). Source compact, target
+Hausdorff, so A_w is a HOMEOMORPHISM.
+
+Identifying the cone with the hypersurface. Weighted homogeneity says
+
+    f_j( a_{w^{(j)}} (t, x) )  =  t  f_j(x),                   (6.15)
+
+where w^{(j)} is the weight vector of f_j (this is where normalisation
+"weighted degree 1" is used). So A_{w^{(j)}} identifies the cone on
+K_j with V(f_j) intersect B^(2N)_eps, the CLOSED BALL intersect the
+hypersurface.
+
+Conclusion. Extend phi to a map on cones by
+
+    h(0)  =  0,
+    h( a_{w^{(1)}}(t, x) )  =  a_{w^{(2)}}(t, phi(x)),    0 < t <= 1.   (6.16)
+
+Equivalently,
+
+    h  =  A_{w^{(2)}}  o  C(phi)  o  A_{w^{(1)}}^{-1},          (6.17)
+
+where C(phi) is the coning of phi. Then h is an ambient ball
+homeomorphism carrying V(f_1) to V(f_2), preserving the ball boundary.
+Restricting to interiors gives the required homeomorphism GERM OF
+AMBIENT PAIRS:
+
+    (C^N, V(f_1), 0)  is isomorphic to  (C^N, V(f_2), 0).       (6.18)
+
+Combined with Lemma 6.1, this proves Theorem 1.1 (main theorem) in
+full.
+
+Original: Section 6, "From Seifert matrices to ambiently equivalent
+knots" and "An explicit homeomorphism of ambient germs".
+
+Exercise 5 (Corollary 6.3, Arnold corank problem negative).
+
+Corollary 6.3. In the finite ambient dimension N = 4 s_0 + 16 > 3
+divisible by 8, the polynomials f_1, f_2 from (6.1) have DIFFERENT
+Hessian coranks, although their reduced complex hypersurface germs are
+ambiently homeomorphic. Both f_1, f_2 have real coefficients, are
+weighted homogeneous for positive rational weights, and have isolated
+critical points at the origin.
+
+Proof. Theorem 1.1 supplies the ambient homeomorphism and the stated
+analytic properties. Lemma 6.1 gives rank Hess_0(f_1) = 16 E and
+rank Hess_0(f_2) = 0. Since E > 0 (from the finite certificate of
+Lemma 3.1), the coranks satisfy
+
+    corank_0(f_1)  =  N - 16 E  <  N  =  corank_0(f_2).         (6.19)
+
+Both values are non-equal, giving negative answer to Arnold's corank
+problem for these weighted homogeneous reduced real-coefficient germs
+of isolated critical type.
+
+Original: Section 6, Corollary 6.3 (arnold-corank) and its proof.
+
+Exercise 6 (Corollary 6.4, right-topological multiplicity and parity
+counterexamples).
+
+Corollary 6.4. For the same f_1, f_2 and the same finite ambient
+dimension N > 3 divisible by 8 in (6.2), there is a homeomorphism germ
+
+    phi : (C^N, 0)  ->  (C^N, 0)     with     f_1  =  f_2 o phi.  (6.20)
+
+Nevertheless, the multiplicities are respectively 2 and 3 (from Lemma
+6.1). So neither multiplicity nor multiplicity mod 2 is invariant under
+topological right equivalence, even for reduced real-coefficient
+weighted homogeneous holomorphic germs with positive rational weights
+and isolated critical points at the origin.
+
+Proof. King (1978) and Saeki (1989) relate ambient homeomorphism of
+pairs to right-left topological equivalence of the functions. Saeki's
+Corollary 2 identifies V-equivalence with topological right-left
+equivalence; the reference summary (Saeki 1989 Section 3, p. 28) and
+Sampaio (2025, proof of Proposition 2.4) supply the precise
+formulation: for holomorphic germs f, g with isolated critical points
+at the origin and n >= 2, ambient homeomorphism of zero-set germs
+gives either
+
+    f = g o phi     OR     f = (c_1 o g o c_n) o phi,            (6.21)
+
+where c_k : (C^k, 0) -> (C^k, 0) is the standard complex conjugation in
+the first k coordinates (c_k(z_1, ..., z_k) = (conj z_1, ..., conj z_k)).
+
+Real coefficients kill the second alternative. Write
+f_2(z) = sum_alpha a_alpha z^alpha with a_alpha in R. Then
+(c_1 o f_2 o c_N)(z) = conj(f_2(conj z)) = sum_alpha conj(a_alpha)
+(conj(conj z))^alpha = sum_alpha a_alpha z^alpha = f_2(z). So
+c_1 o f_2 o c_N = f_2 and the two alternatives of (6.21) coincide:
+both say f_1 = f_2 o phi.
+
+Multiplicities: from Theorem 1.1 and Lemma 6.1, ord f_1 = 2, ord f_2 =
+3, unequal in Z and unequal mod 2. So both right-topological
+multiplicity conjectures (the full and the mod-2 version) are refuted.
+
+Note. The phi in (6.20) may differ from the AMBIENT homeomorphism h
+constructed in Exercise 4, and it need not be holomorphic or
+bi-Lipschitz. The parity statement also holds separately: the coranks
+corank_0(f_1) = N - 16E and corank_0(f_2) = N have the same parity
+(they differ by the even number 16E), so a corank-parity theorem
+(which does hold, per Sampaio 2026) is not contradicted.
+
+Original: Section 6, Corollary 6.4 and its proof; King 1978, Saeki
+1989, Sampaio 2025.
+
+Exercise 7 (Corollary 6.5, initial-form Milnor-fibre counterexamples).
+
+Corollary 6.5. In the single ambient dimension N + 1, put
+
+    f~_j(z, t)  =  f_j(z) + t^2,     j = 1, 2,    in C^(N+1).   (6.22)
+
+The germs f~_1, f~_2 : (C^(N+1), 0) -> (C, 0) are topologically right
+equivalent and have isolated critical points at the origin, but
+
+    H_0( (f~_1)_*^{-1}(1); Z )  is isomorphic to  Z    (connected),
+    H_0( (f~_2)_*^{-1}(1); Z )  is isomorphic to  Z^2  (two pieces).  (6.23)
+
+Thus integral homology of the initial-form Milnor fiber is not
+invariant under topological right equivalence, already in degree ZERO.
+In particular the initial-form Milnor-fibre homotopy conjecture of
+Budur, Fernandez de Bobadilla, Le, and Nguyen (Conjecture 1.7 arXiv
+v3) is false.
+
+Proof. Topological right equivalence is preserved under the QUADRATIC
+STABILISATION (f_j + t^2) by Sampaio's proof of Proposition 2.3. Given
+phi from (6.20), the product homeomorphism phi~ (z, t) = (phi(z), t)
+satisfies f~_1 = f~_2 o phi~. It carries every zero set correspondingly.
+Also
+
+    grad f~_j(z, t)  =  (grad f_j(z), 2 t),
+
+which vanishes only at z = 0 AND t = 0 (grad f_j(z) = 0 forces z = 0
+by isolation, and 2 t = 0 forces t = 0). So the isolated critical point
+property is preserved.
+
+Now compute initial-form Milnor fibres. From Lemma 6.1:
+
+    (f~_1)_*  =  Q + t^2,     Q  =  sum_{alpha = 1}^{16E} u_alpha^2
+    (f~_2)_*  =  t^2.
+
+(For f_1, the lowest part is Q (quadratic); adding t^2 makes it
+Q + t^2. For f_2, the lowest part is 0 (order 3); adding t^2 makes it
+t^2 = the pure lowest part.)
+
+Unit fibres.
+
+    (f~_1)_*^{-1}(1)  =  {v in C^(16E+1) : sum_{alpha=1}^{16E+1} v_alpha^2 = 1}
+                           x  C^(N - 16E).                     (6.23a)
+
+To identify the quadric: write v = x + i y with x, y in R^(16E+1). Then
+sum v_alpha^2 = |x|^2 - |y|^2 + 2 i (x . y). The equation
+sum v_alpha^2 = 1 becomes |x|^2 - |y|^2 = 1 and x . y = 0. The map
+
+    (u, y)  ->  sqrt(1 + |y|^2) u + i y,
+    u in S^(16E),  y perpendicular u in R^(16E+1)
+
+identifies the quadric with the tangent bundle of the sphere S^(16E).
+The tangent bundle of a connected manifold is connected, so the quadric
+is connected. Hence the whole first unit fibre is connected, and
+H_0 = Z.
+
+    (f~_2)_*^{-1}(1)  =  C^N x {t in C : t^2 = 1} = C^N x {+1, -1}.
+                                                                  (6.23b)
+
+Two connected components: H_0 = Z^2.
+
+The initial-form Milnor fibres are therefore not even homotopy
+equivalent (H_0 differs), so the Budur et al. conjecture (which would
+give homotopy equivalence) fails.
+
+The isolatedness in the corollary is for the STABILISED germs, not their
+initial forms. In particular the nonreduced initial form t^2 is
+retained; replacing it with the reduced t (which would have the same
+zero set of the moment map) would change the unit fibre. No
+reducedness or isolatedness hypothesis on the initial form is part of
+the conjecture.
+
+Original: Section 6, Corollary 6.5 and its proof; Sampaio 2025,
+Budur et al. 2022.
+
+Exercise 8: Wrap up the main theorem and its corollaries.
+
+Main theorem (Theorem 1.1). Proved by combining:
+    Section C (seed polynomials p_1, p_2 with different orders and
+               equal spectra mod 2; prepared q_j = p_j + x^5 + y^7);
+    Section D (symmetrization and first tensor square; local and
+               rational isometries);
+    Section E (integral reversers C_j; second tensor square;
+               determinant cancellation; six cube + prime 11, 13
+               factors; strong approximation; integral isometry U);
+    Section F, Exercises 1-4 (final polynomials f_j; the map Psi
+               interleaves Seifert forms and monodromy; determinant
+               det(1 - T) = 1; homotopy sphere and simple knot;
+               Levine's classification; weighted radial coordinates;
+               ambient ball homeomorphism).
+
+Consequences:
+    - Arnold corank problem: Corollary 6.3, exercise 5.
+    - Right-topological multiplicity and parity: Corollary 6.4,
+      exercise 6.
+    - Initial-form Milnor-fibre conjecture: Corollary 6.5, exercise 7.
+
+Original: Section 6, closing paragraphs; Section 1 first page.
+
+==============================================================
+SECTION G: THE CERTIFICATE, REFERENCES, AND VERIFICATION NOTES
+==============================================================
+
+Exercise 1 (Appendix A, the certificate program). What does the
+appendix program actually do, and what can be checked by running it?
+
+The program's inputs:
+    d = 3^11 = 177147.
+    B = { b : 0 < b < d/2 and 3 does not divide b }.
+
+The program's output: the "usable" (b, prime) pairs after repeated
+deletion of singly-supported columns, and the F_1009 rank of the
+augmented matrix. Concretely:
+
+    Sieve primes up to 3d/2 + 2 ~ 265721 to get exact integral
+        prime valuations of d-b, b, 2d+b, 2d-b, 3d-b, d+b.
+    Construct one integer column per b in B (|B| = phi(3^11) / 2
+        considerations put it at ~59049).
+    Restrict to b with all six factors 4450-smooth: 1118 columns.
+    Iteratively delete singly-supported columns; drop zero rows:
+        896 columns, 867 nonzero rows.
+    Append a row of ones (augmented M~) and compute
+        rank over F_1009 by Gaussian elimination with normalised
+        pivots: result 868.
+
+The program uses only integer arithmetic and Python's standard library.
+No floating point, no third-party packages, no probabilistic steps.
+The prime 1009 is prime: none of 2, 3, 5, 7, 11, 13, 17, 19, 23, 29,
+31 divides 1009, and 32^2 = 1024 > 1009.
+
+Adjugate argument. Since M~ is 868 x 896 with rank 868, the 896 columns
+include an 868 x 868 matrix H with det(H) != 0. Set z = adj(H)
+(0, ..., 0, 1)^t, extend by zeros, adjust sign to make
+E = |det H| > 0. This gives integers e_b with the required product
+identities, ensuring sum e_b > 0. The lexicographically-first
+nonsingular minor works.
+
+Supplementary witness. The paper also references
+"verification/support/explicit-relation.json" (849 nonzero integer
+coefficients, each at most 140 digits, positive coordinate sum) and
+"verification/validation/explicit_relation.py" (a Python stdlib program
+that verifies both product identities of Lemma 3.1 by brute-force
+trial-division of the six integers in every row, without using the
+matrix or its rank). The rank certificate above and the adjugate
+argument already prove Lemma 3.1; the JSON is supplementary.
+
+Original: Appendix A ("The finite arithmetic certificate").
+
+Exercise 2 (References, ordered by section for cross-check).
+
+Section A (background on the multiplicity problem and consequences):
+    Zariski 1971, Some open questions in the theory of singularities.
+    Zariski 1932, On the topology of algebroid singularities.
+    A'Campo 1973, Le nombre de Lefschetz d'une monodromie.
+    Le 1973, Calcul du nombre de cycles evanouissants.
+    Yau 1988, Topological types and multiplicities.
+    Fernandes, Jelonek, Sampaio 2025, Proof of the metric Arnold corank problem.
+    Sampaio 2025, Milnor fibres of initial forms of topologically
+        equivalent holomorphic functions.
+    Sampaio 2026, On the Arnold corank, Zariski multiplicity, and
+        tangent Milnor fiber problems.
+    Bobadilla-Pelka 2022, Symplectic monodromy at radius zero.
+    Le-Ramanujam 1976, The invariance of Milnor's number implies
+        invariance of topological type.
+    Greuel 1986; O'Shea 1987 (equimultiplicity for quasihomogeneous
+        / mu-constant deformations).
+    Budur, Fernandez de Bobadilla, Le, Nguyen 2022, Cohomology of
+        contact loci.
+
+Section B (algebraic and arithmetic inputs):
+    Milnor 1968, Singular Points of Complex Hypersurfaces.
+    Sebastiani-Thom 1971, Un resultat sur la monodromie.
+    Sakamoto 1974, Seifert matrices of Milnor fiberings.
+    Steenbrink 1977, Intersection form for quasi-homogeneous singularities.
+    Steenbrink 1989, The spectrum of hypersurface singularities.
+    Nemethi 1995, The real Seifert form and spectral pairs.
+    Balnojan-Hertling 2019, Real Seifert forms and polarizing forms.
+    Orlik 1972, On the homology of weighted homogeneous manifolds.
+    Orlik-Randell 1977, The monodromy of weighted homogeneous singularities.
+    Hertling-Mase 2022, The integral monodromy of isolated quasihomogeneous singularities.
+    Milne 2022, Algebraic Groups.
+    O'Meara 1963, Introduction to Quadratic Forms.
+    Witt 1937, Theorie der quadratischen Formen.
+    Jacobowitz 1962, Hermitian forms over local fields.
+    Platonov 1969, Strong approximation and the Kneser-Tits conjecture.
+    Platonov 1970, Addendum.
+    Rapinchuk 2014, Strong approximation for algebraic groups.
+    Smale 1961, Generalized Poincare conjecture.
+    Levine 1970, An algebraic classification of some knots of codimension two.
+    Durfee 1974, Fibered knots and algebraic singularities.
+    Saeki 1999, Theory of fibered 3-knots in S^5.
+    Neumann 2017 (repost), Topology of hypersurface singularities.
+    Hatcher 2002, Algebraic Topology.
+
+Section F (topological step):
+    King 1978, Topological type of isolated critical points.
+    Saeki 1989, Topological types of complex isolated hypersurface singularities.
+    Hatcher 2002, Algebraic Topology (Alexander duality, Hurewicz).
+    Stacks Project (algebraic background: Tags 00ND, 02JN).
+
+Supplementary files referenced by the paper:
+    "verification/support/supplied_finite_rank.py" - the certificate program.
+    "verification/support/explicit-relation.json" - explicit e_b witness.
+    "verification/validation/explicit_relation.py" - trial-division checker.
+
+Original: References in the source paper.
+
+==============================================================
+CLOSING REMARKS
+==============================================================
+
+What is surprising. The pair f_1, f_2 in Theorem 1.1 are
+ambiently homeomorphic (hence topologically right equivalent in a
+suitable sense, as made precise in Corollary 6.4), but they have
+multiplicities 2 and 3. Multiplicity is the FIRST nonzero data in the
+Taylor expansion; if two functions are topologically the same function,
+why should their lowest-order term be permitted to differ? The answer
+is: because "topologically the same" (ambient homeomorphism) requires
+NO regularity of the change of coordinates. It allows arbitrary
+crumpling and unsmoothing. Only when one insists on additional
+hypotheses (holomorphic coordinate change, bi-Lipschitz, family of
+mu-constant deformations, low ambient dimension, or quasihomogeneous
+in low dimension) does multiplicity become invariant.
+
+The construction is highly non-constructive in the "smallest N" sense;
+N is fixed by exact finite arithmetic (N = 4 s_0 + 16 with s_0 coming
+from the seed counts), not minimised. Even so, N is finite and
+explicit.
+
+What still holds. The paper is consistent with:
+    - Zariski's original positive results (plane curves, N = 2).
+    - A'Campo-Le (multiplicity-one exclusion).
+    - Yau (quasihomogeneous surfaces, N = 3).
+    - Fernandes-Jelonek-Sampaio (multiplicity two under bi-Lipschitz
+      equivalence, and Hessian corank in three variables).
+    - Le-Ramanujam (constant-mu-number family deformations,
+      ambient dimension != 3).
+    - Greuel / O'Shea / Fernandez de Bobadilla - Pelka
+      (equimultiplicity theorems for mu-constant families).
+    - Sampaio 2026's corank-parity theorem (the two coranks here
+      differ by the even number 16E, so they have the same parity).
+
+The counterexample sits strictly OUTSIDE all of these (positive)
+settings.
+
+Application. This settles four open problems simultaneously in the
+theory of hypersurface singularities (Zariski multiplicity, its parity
+variant, Arnold's corank problem, and the initial-form Milnor-fibre
+homotopy conjecture). The construction is not an accident: it depends
+on a finite integer relation whose existence is certified exactly by
+the Python program in Appendix A. No floating-point or approximate
+computation enters the proof.
+
+Original: Sections 1, 6; Appendix A.