             A characteristic-two counterexample to dévissage
                          for local deformation functors
                                                  Alec Kriebel∗

                                     12 October 2026 (research note)



                                                    Abstract
              We give a negative answer to the unrestricted ﬁbre-product question for local deformation
          functors posed by Byszewski. For the weakly ramiﬁed action of the additive group of F4 given
          by t 7→ t/(1 + at), and its subgroup F2 , restriction–induction cannot be formally smooth over
          any prorepresentable recipient, regardless of its tangent dimension. The proof combines the
          known involution deformation pathology with an exact calculation showing that restriction has
          injective tangent map. The variant assuming prorepresentability of the subgroup and quotient
          deformation functors is untouched. The involution pathology and additive action are established
          prior ingredients; the contribution is their use to obstruct the proposed factorization.


1        The unrestricted question and the counterexample
Let k be a perfect ﬁeld of characteristic p and C the category of local Artinian W (k)-algebras with
identiﬁed residue ﬁeld k. For a faithful local action ρ : H → Autk k[[t]], DH (A) is the set of lifts to
AutA A[[t]], modulo coordinate conjugation reducing to the identity. This is the unframed category
of [1, 3, 4]; nilpotent constant terms are permitted. The residual action gives a unique point over
k. For N / G and Q = G/N , restriction and induction give a natural map
                                                             Q
                                         (res, ind) : DG −→ DN × DQ ,
       Q
where DN is the actual ﬁxed subfunctor of DN . A prorepresentable functor has the form

                                          hR (A) = HomW (k),loc (R, A)

for a complete Noetherian local W (k)-algebra with residue k.
    The question in [1, p. 3039], [2, Question 6.4.4], and [3, Question 4.4] asks for a prorepresentable
                      Q
F , natural maps DN     → F ← DQ , and a formally smooth restriction–induction map into the
ﬁbre product. The reﬁned formulation prescribes T F = H 1 (Q, Θ] /ΘN ), with Θ = Derk k[[t]] and
Θ] = Derk (k[[t]]N ). Neither question assumes DN or DQ prorepresentable. Those extra hypotheses
are stated separately in [1, Theorem 3] and [3, Remark 4.5, Theorem 4.9].
Theorem 1. Put k = F4 = F2 [u]/(u2 + u + 1), G = (k, +), N = (F2 , +), and ρ(a)(t) = t/(1 + at).
                                                      Q
There are no prorepresentable F and natural maps f : DN → F , g : DQ → F for which restriction–
induction factors as a formally smooth map
                                                     Q
                                              DG −→ DN ×F DQ .
    ∗
        Independent researcher. ORCID: 0009-0001-9320-500X.


                                                        1
This remains impossible for every tangent dimension of F ; in particular it excludes the prescribed
one-dimensional tangent space in this example.
    The action is a standard weakly ramiﬁed additive action, not a new construction. Direct
substitution gives ρ(a)ρ(b) = ρ(a + b), and the t2 coeﬃcient distinguishes a. Every nonidentity
element has conductor one. The proof below uses actual deformation classes, without replacing
either component functor by a versal hull.


2    A quotient tangent invisible to every recipient
The N -quotient parameter and induced Q-action are
                         t2                                                   y
                   y=       ,       k[[t]]N = k[[y]],     σ(y) = ρ(u)(y) =       .
                        1+t                                                  1+y
Here t2 +yt+y = 0, ρ(1)(t) = t+y, and every series uniquely has the form A(y)+tB(y): the leading
terms of y m and ty m have even and odd t-orders, and successive leading-term subtraction converges.
Invariance forces yB(y) = 0. The same basis argument for Laurent series gives k((t))N = k((y)).
Finally u(u + 1) = 1 gives

                                            t2               t2       y
                      ρ(u)(y) =                          =       2
                                                                   =     .
                                  (1 + ut)(1 + (u + 1)t)   1+t+t     1+y
Thus DQ is the weakly ramiﬁed involution deformation functor.
Lemma 2. There is a nonzero v ∈ DQ (k[ε]/ε2 ) whose image under every natural map DQ → hR
is the residue basepoint.
Proof. Let B = k[ε]/ε3 , A = k[ε]/ε2 , and let j : B → B ﬁx k and send ε to ε + ε2 . These are
objects and maps of C via W (k) → k. Deﬁne
                                                               (   )
                                           y+ε                 1 ε
                                  σε (y) =     ,        Mε =       .
                                           1+y                 1 1

Its constant coeﬃcient is nilpotent and its linear coeﬃcient is a unit, so substitution deﬁnes a
continuous automorphism. Its matrix square is (1 + ε)I, so it is an involution lifting σ. Write
x = [σε ] ∈ DQ (B) and v for its reduction to A.
    The class
         (
              x)is ﬁxed by j. The identity-reducing coordinate change χ(y) = (1+ε)y+ε, represented
by C = 1+ε    ε
            0 1 , satisﬁes
                                        Mε+ε2 C = (1 + ε)CMε
in B. Scalar matrices do not change the fractional-linear substitution, so this gives the required
conjugacy. Function composition reverses the source’s ring-substitution order; the equivalence
is valid in either conjugating direction. This is the known calculation of [4, Proposition 4.1],
reproduced here for completeness.
    For a natural map η : DQ → hR , write ψ : R → B for ηB (x). Naturality gives j ◦ ψ = ψ. Since

                                            B j = k + kε2 ,

the composite R → B → A lands in k ⊂ A. It is consequently the residue map followed by k → A,
hence ηA (v) is the basepoint. This argument applies also to mixed-characteristic R and arbitrary
tangent dimension, with no linearity assumption on η.

                                                   2
    The class v is nonzero. Every identity-reducing coordinate change over A has the form y + εh(y)
with arbitrary h ∈ k[[y]], including its constant coeﬃcient. Its ﬁrst-order conjugacy perturbation
is, up to sign, h(σ(y)) − σ 0 (y)h(y), whose constant coeﬃcient is zero. But σε − σ = ε/(1 + y) has
constant perturbation coeﬃcient one. Thus no allowed conjugator trivializes v.

    The conclusion of Lemma 2 is also an elementary consequence of the previously published
failure to preserve injections in [4, Proposition 6.1, p. 899]. There the nonzero involution tangent
and the basepoint become equal after the injection k[e]/e2 ,→ k[w]/w3 , e 7→ w2 . Every hR preserves
injections of target rings. Naturality therefore makes their images equal already over k[e]/e2 . We
claim no new recipient-collapse phenomenon or new involution pathology; the direct cubic-ring
proof ﬁxes the particular tangent class used here.


3    The restriction tangent is injective
Over k, every algebraic derivation of k[[t]] is a coeﬃcient derivation: perfectness in characteristic
two gives h = a(t)2 +tb(t)2 , so d(h) = d(t)b(t)2 = d(t)h0 (t). Thus Θ = k[[t]]∂t , and these derivations
are continuous. There is no additional discontinuous tangent module in this example.
    Restriction to the N -ﬁxed ring identiﬁes
                                              ∼ M := yk[[y]]∂y .
                                           ΘN =

Indeed the unique extension of b(y)∂y to the separable ﬁeld k((t)) has coeﬃcient

                                    b(y)        (1 + t)2           dy      t2
                          a(t) =         = b(y)          ,            =          .
                                   dy/dt           t2              dt   (1 + t)2

Uniqueness makes it N -invariant. It preserves k[[t]] precisely when 2 ordy b − 2 ≥ 0, namely b ∈
yk[[y]]. Conversely an invariant derivation restricts and is recovered by this formula. The ﬁeld
polynomial has derivative y 6= 0, so separability and uniqueness hold.
    Conjugation by the quotient involution acts on coeﬃcient functions by
                                                       (      )
                                                            y
                              T (f )(y) = (1 + y)2 f           ,       T 2 = 1.
                                                           1+y
All group elements are involutions, so the inverse-conjugation convention gives the same operator.
Put z = y 2 /(1 + y). The preceding ﬁxed-ring basis argument gives k((y))Q = k((z)), and nonzero
invariant Laurent series have even y-valuation. For f ∈ yk[[y]], T f = f is equivalent to invariance
of f /y 2 . Its valuation is at least −1, hence by evenness at least zero. It follows that

                    M Q = y 2 k[[z]]∂y ,     (T − 1)(yh(z)) = y 2 h(z)      (h ∈ k[[z]]).

The reverse containment of (T − 1)M in M Q follows from (T − 1)2 = 0. The cyclic cocycle formula
therefore gives
                                                ker(1 + T )
                                 H 1 (Q, ΘN ) =             = 0.
                                                (T − 1)M
These are identities on complete formal series, not ﬁnite-jet calculations.
    The ﬁrst-order lifting calculation identiﬁes T DH = H 1 (H, Θ), and restriction with cohomolog-
ical restriction [3, Proposition 2.4]: write a lift as (id + εc(g))ρ(g); group relations give a cocycle,
and coordinate conjugation changes it by a coboundary. The vanishing above makes restriction
injective. Explicitly, for a cocycle whose restriction is zero in cohomology, subtract a coboundary

                                                       3
so it vanishes on N . The cocycle equation and commutativity of G make its value at u N -invariant;
the relation u + u = 0 gives (1 + T )c(u) = 0. Write c(u) = (T − 1)m with m ∈ ΘN and subtract this
coboundary too. It preserves the zero N -values and kills u, hence kills the entire cocycle. Thus

                                      ker(T DG −→ T DN ) = 0.
        Q
Since DN  (A) ⊂ DN (A) literally, restriction into the actual ﬁxed subfunctor also has zero tangent
kernel.
    For the prescribed space, Θ] = k[[y]]∂y and Θ] /ΘN =   ∼ k has trivial Q-action (constant coeﬃ-
                                      1      ]   N
cients are preserved by T ). Hence H (Q, Θ /Θ ) = k, as asserted in the theorem. Its dimension
is not needed for the contradiction.


4    Failure of formal smoothness and scope
Proof of Theorem 1. Suppose such F , f and g exist; identify F naturally with hR . At A = k[ε]/ε2 ,
                                                         Q
naturality from k sends the residual basepoint 0N ∈ DN     (A) under f to the residue basepoint
of F (A). Lemma 2 gives the same image for gA (v), with v 6= 0. Therefore (0N , v) belongs to
   Q
(DN  ×F DQ )(A). Its reduction is the unique residual point. The surjection A → k is small:
its kernel is annihilated by the maximal ideal of A. Formal smoothness would lift this pair to
w ∈ DG (A) with
                                  res(w) = 0N ,    ind(w) = v.
Restriction injectivity forces w = 0G . Naturality of induction then gives ind(w) = 0Q , contradicting
v 6= 0.

    The cubic ring is used only to constrain detection of a ﬁrst-order tangent by an arbitrary
prorepresentable recipient. The obstruction to smoothness already occurs on A → k. No extension
            Q
of f from DN   to DN is assumed, and no cohomological tangent formula for f or g is prescribed.
Nor is universality of a Klein-four deformation ring used.
    This answers the unrestricted original question negatively and does not contradict the condi-
tional obstruction-space theorems cited above. The involution component is in the exceptional
nonprorepresentable case. The variant imposing prorepresentability of DN and DQ remains unad-
dressed; changing to coordinate automorphisms ﬁxing the zero section would likewise change the
category and exclude the crucial nilpotent translation. The introductory and reﬁned formulations
are one question, not separate discoveries.

Priority, veriﬁcation and review status. The known additive action, involution conjugacy,
failure to preserve injections, and tangent-cohomology framework are credited above. The sup-
plement records independent formal-series and functor audits, ﬁnite exact checks and a bounded
priority search. No matching prior negative factorization was located in inspected primary bodies;
this does not prove exhaustive originality or that the question remained open until 2026. Finite com-
putations support explicit identities; the unrestricted formal series and all-natural-map statements
are proved in the text, without a formal proof-assistant certiﬁcate.
    AI tools were used extensively in solving, verifying, auditing sources, and preparing this paper
and supporting materials. This is an unrefereed research note. Independent adversarial reviews are
AI reviews; conventional human peer review and journal acceptance are not claimed.




                                                  4
References
[1] J. Byszewski, Dévissage for local deformation functors, Oberwolfach Reports 54/2008, 3038–
    3040. Primary report.

[2] J. Byszewski, Cohomological aspects of equivariant deformation theory, Ph.D. thesis, Utrecht
    University (2009), Question 6.4.4. Repository record.

[3] J. Byszewski, Deformation functors of local actions, arXiv:1112.0352v1 (2011).

[4] J. Byszewski and G. Cornelissen, Which weakly ramiﬁed group actions admit a universal formal
    deformation?, Ann. Inst. Fourier 59 (2009), 877–902. doi:10.5802/aif.2450, Propositions 4.1
    and 6.1.




                                                5
