Martin Escardo, 24-26 October 2023.
A type is called rigid if its only automorphism is the identity. In
HoTT/UF, we would instead say that its type of automorphisms is
contractible.
The type Ξ© is not rigid in boolean toposes, because in such toposes it
has precisely two automorphisms (the identity and negation).
But, in any topos, if there is an automorphism of Ξ© different from the
identity, then the topos is boolean.
\begin{code}
{-# OPTIONS --safe --without-K #-}
open import MLTT.Spartan
open import UF.ClassicalLogic
open import UF.Embeddings
open import UF.Equiv hiding (_β
_)
open import UF.Equiv-FunExt
open import UF.FunExt
open import UF.Subsingletons
open import UF.SubtypeClassifier hiding (Ξ©)
open import UF.SubtypeClassifier-Properties
module Higgs.Rigidity
{π€ : Universe}
(fe : Fun-Ext)
(pe : propext π€)
where
open import Higgs.InvolutionTheorem fe pe
\end{code}
From the existence of certain automorphisms of Ξ©, we conclude that
excluded middle holds.
\begin{code}
Ξ©-automorphism-that-maps-β€-to-β₯-gives-EM
: (Ξ£ π κ Aut Ξ© , β π β β€ οΌ β₯)
β EM π€
Ξ©-automorphism-that-maps-β€-to-β₯-gives-EM ((f , f-is-equiv) , e) = II
where
f-is-involutive : involutive f
f-is-involutive = automorphisms-of-Ξ©-are-involutive f f-is-equiv
I : (P : π€ Μ ) β is-prop P β Ξ£ Q κ π€ Μ , (P β Β¬ Q)
I P P-is-prop = f p holds , g , h
where
p : Ξ©
p = (P , P-is-prop)
g : P β Β¬ (f p holds)
g p-holds = equal-β₯-gives-fails (f p)
(f p οΌβ¨ ap f (holds-gives-equal-β€ pe fe p p-holds) β©
f β€ οΌβ¨ e β©
β₯ β)
h : Β¬ (f p holds) β P
h Ξ½ = equal-β€-gives-holds p
(p οΌβ¨ (f-is-involutive p)β»ΒΉ β©
f (f p) οΌβ¨ ap f (fails-gives-equal-β₯ pe fe (f p) Ξ½) β©
f β₯ οΌβ¨ ap f (e β»ΒΉ) β©
f (f β€) οΌβ¨ f-is-involutive β€ β©
β€ β)
II : EM π€
II = all-props-negative-gives-EM fe I
Ξ©-automorphism-swap-β : (π : Aut Ξ©)
β {p q : Ξ©}
β (β π β p οΌ q) β (β π β q οΌ p)
Ξ©-automorphism-swap-β π {p} {q} =
involution-swap-β β π β
(automorphisms-of-Ξ©-are-involutive β π β β π β-is-equiv)
(Ξ©-is-set fe pe)
\end{code}
A stronger version of the following is proved below.
\begin{code}
Ξ©-automorphism-apart-from-id-gives-EM
: (Ξ£ π κ Aut Ξ© , Ξ£ pβ κ Ξ© , β π β pβ β pβ)
β EM π€
Ξ©-automorphism-apart-from-id-gives-EM (π@(f , f-is-equiv) , pβ , Ξ½) = VIII
where
I : f β€ β β€
I e = VI
where
II : pβ β β€
II eβ = Ξ½ II'
where
II' : f pβ οΌ pβ
II' = transportβ»ΒΉ (Ξ» - β f - οΌ -) eβ e
III : pβ οΌ β₯
III = different-from-β€-gives-equal-β₯ fe pe pβ II
IV : f β₯ β β₯
IV eβ = Ξ½ IV'
where
IV' : f pβ οΌ pβ
IV' = transportβ»ΒΉ (Ξ» - β f - οΌ -) III eβ
V : f β₯ β β€
V eβ = β₯-is-not-β€
(β₯ οΌβ¨ (β Ξ©-automorphism-swap-β π β eβ)β»ΒΉ β©
f β€ οΌβ¨ e β©
β€ β)
VI : π
VI = no-truth-values-other-than-β₯-or-β€ fe pe (f β₯ , IV , V)
VII : f β€ οΌ β₯
VII = different-from-β€-gives-equal-β₯ fe pe (f β€) I
VIII : EM π€
VIII = Ξ©-automorphism-that-maps-β€-to-β₯-gives-EM (π , VII)
\end{code}
Notice that we can replace "Ξ£" by "β" in the above propositions, to
get the same conclusion EM π€, because the type EM π€ is a proposition.
Notice also that the converses of the above propositions hold.
We show that there can't be any automorphism of Ξ© distinct from the
identity unless excluded middle holds.
The fact eval-at-β€-is-lc stated and proved below, which is our main
lemma, is attributed to Denis Higgs in the literature [1] [2], without
any explicit citation I could find, with diagrammatic proofs in topos
theory rather than proofs in the internal language of a topos. Our
internal proofs don't necessarily follow the external diagrammatic
proofs.
[1] Peter T. Johnstone. Automorphisms of Ξ©. Algebra Universalis,
9 (1979) 1-7.
https://doi.org/10.1007/BF02488012
[2] Peter Freyd. Choice and well-ordering. Annals of Pure and Applied
Logic 35 (1987) 149-166.
https://doi.org/10.1016/0168-0072(87)90060-1
https://core.ac.uk/download/pdf/81927529.pdf
\begin{code}
private
fe' : FunExt
fe' π₯ π¦ = fe {π₯} {π¦}
eval-at-β€ : Aut Ξ© β Ξ©
eval-at-β€ π = β π β β€
eval-at-β€-is-lc : left-cancellable eval-at-β€
eval-at-β€-is-lc {π@(f , _)} {π@(g , _)} e = III
where
have-e : f β€ οΌ g β€
have-e = e
I : (p : Ξ©) β (f p οΌ β€) β (g p οΌ β€)
I p = (f p οΌ β€) ββ¨ Ξ©-automorphism-swap-β π β©
(f β€ οΌ p) ββ¨ transport-β (_οΌ p) e β©
(g β€ οΌ p) ββ¨ Ξ©-automorphism-swap-β π β©
(g p οΌ β€) β
II : f βΌ g
II p = Ξ©-ext' pe fe (I p)
III : π οΌ π
III = to-β-οΌ fe II
eval-at-β€-is-embedding : is-embedding eval-at-β€
eval-at-β€-is-embedding = lc-maps-into-sets-are-embeddings
eval-at-β€ eval-at-β€-is-lc
(Ξ©-is-set fe pe)
\end{code}
From this we conclude that there can't be any automorphism of Ξ©
distinct from the identity unless excluded middle holds. I don't think
this has been observed before in the literature, but it may have been
observed in the folklore.
\begin{code}
Ξ©-automorphism-distinct-from-ππ-gives-EM
: (Ξ£ π κ Aut Ξ© , π β ππ)
β EM π€
Ξ©-automorphism-distinct-from-ππ-gives-EM (π , Ξ½) = IV
where
f : Ξ© β Ξ©
f = β π β
I : f β€ οΌ β€ β π οΌ ππ
I = eval-at-β€-is-lc {π} {ππ}
II : f β€ β β€
II = contrapositive I Ξ½
III : f β€ οΌ β₯
III = different-from-β€-gives-equal-β₯ fe pe (f β€) II
IV : EM π€
IV = Ξ©-automorphism-that-maps-β€-to-β₯-gives-EM (π , III)
\end{code}
It follows that the type Ξ£ f κ Aut Ξ© , f β id is a proposition,
constructively. In boolean toposes it is a singleton, in non-boolean
toposes it is empty, and in all toposes it is a subsingleton. This is
because from any hypothetical element (f , Ξ½) of this type we conclude
that excluded middle holds, and hence Ξ© β π, and therefore f is
negation. So this is a constructive proof in which we deduce excluded
middle as an intermediate step. And once we conclude that this type is
a proposition, we see that it is equivalent to the type EM π€, which is
also a proposition, as these two propositions imply each other:
(Ξ£ f κ Aut Ξ© , f β id) β EM π€
and hence they are equal if we further assume propositional
extensionality (which follows from univalence).
TODO. Write down this argument in Agda.