Martin Ziegler
decision decision problem problem
•Logicians Tarski, Alonzo Church (PhD advisor)
•Kurt Gödel (1931): There exist arithmetical
statements which are true but cannot be proven so.
Alan M. Turing 1936
• • first first scientific scientific calculations calculations on digital on digital computers computers
• • What What are are its its fundamental fundamental limitations limitations ? ?
• • Uncountably Uncountably many many P P ⊆ ⊆
• • but but countably countably many many ' ' algorithms algorithms ' '
• • Undecidable Undecidable Halting Problem H : : No No algorithm algorithm B B can can always always correctly correctly answer answer the the following following question question
Given
Given 〈 〈 A, A, x〉 x 〉 , , does does algorithm algorithm A A terminate terminate on on input input x x ? ? Proof (by contradiction): Consider Consider algor algor . . B B ' ' that that , , on on input input A A , , executes executes B B on on 〈 〈 A,A A,A 〉 〉
1941
Halting
Halting Problem Problem H H
Proof
Proof ( ( by by contradiction contradiction ): ):
A A
x x B B + +
− −
A A
A A
B' B'
∞ ∞
How How does does B B ' ' behave behave on on B' B' ? ? answer
answer , , loops loops infinitely infinitely . .
and,
and, upon upon a positive a positive
simulator
simulator / / interpreter interpreter B B ? ? B' B' B' B'
B' B'
Martin Ziegler
Formalities & Tools
''Definition:Definition:' ' AlgorithmAlgorithm
A A
decidesdecides setsetL L ⊆ ⊆ { { 0 0 , , 1 1 }* }*
ifif•• on on inputsinputs
x x ∈ ∈ L L
printsprints 11 and and terminatesterminates,,•• on on inputsinputs
x x ∉ ∉ L L
printsprints 00 and and terminatesterminates..A A
semisemi--decidesdecides ifif terminatesterminates on onx x ∈ ∈ L L
,, elseelse divergediverge..all finite all finite
binary binary sequences sequences e.ge.g. "Turing . "Turing
machine machine""
Consider
Consider algor algor . . B B ' ' that that , on , on input input A A , , executes executes B B on on
〈 〈 A,A A,A 〉 〉 and, and, upon upon a positive a positive answer answer , , loops loops infinitely infinitely . .
countable countable!! Techniques
Techniques:: a) a) simulationsimulation c) c) dovetailingdovetailing
Theorem:
Theorem:
L L
decidabledecidable iffiff bothbothL L , , L L
CC semisemi-decidable-decidable InfiniteInfinite
L L ⊆ ⊆ { { 0 0 , , 1 1 }* }*
isis semisemi-decidable-decidable iffiffL L =range( =range( f f ) )
forfor somesome computablecomputable injectiveinjective
f f : : N N → → { { 0 0 , , 1 1 }* }*
b) b) diagonalizationdiagonalization egeg. . UU={ algorithms={ algorithms } } ×× { { inputsinputs }} Universes
Universes UU otherother thanthan
{ { 0 0 , , 1 1 }* }*
((e.ge.g. . NN):): encodeencode..Halting
Halting Problem Problem H H only only semi semi - - decidable decidable
Hilbert Hotel Hilbert Hotel
d) d) reductionreduction (in/(in/outputoutput translationtranslation))
Martin Ziegler
Some Undecidable Problems
''Definition:Definition:' ' AlgorithmAlgorithm
A A
decidesdecides setsetL L ⊆ ⊆ { { 0 0 , , 1 1 }* }*
ifif•• on on inputsinputs
x x ∈ ∈ L L
printsprints 11 and and terminatesterminates,,•• on on inputsinputs
x x ∉ ∉ L L
printsprints 00 and and terminatesterminates..Techniques
Techniques:: a) a) simulationsimulation c) c) dovetailingdovetailing
b) b) diagonalizationdiagonalization
For For
L,L' L,L' ⊆ ⊆ { { 0 0 , , 1 1 }* }*
writewriteL L ≼ ≼ L' L'
ifif therethere isis a computablea computablef f : : { { 0 0 , , 1 1 }* }* → → { { 0 0 , , 1 1 }* }*
such such thatthat∀ ∀ x x : : x x ∈ ∈ L L ⇔ ⇔ f f ( ( x x ) ) ∈ ∈ L L '. '.
a) a)
L' L'
decidabledecidable ⇒⇒ so soL L
. . b) b)L L ≼ ≼ L' L' ≼ ≼ L'' L''
⇒⇒L L ≼ ≼ L'' L''
Universes
Universes UU otherother thanthan
{ { 0 0 , , 1 1 }* }*
((e.ge.g. . NN):): encodeencode.. d) d) reductionreduction (in/(in/outputoutput translationtranslation)) HaltingHalting problemproblem: :
H H = { = { 〈 〈 A A , , x x 〉 〉 : : A A
terminatesterminates ononx x } }
Hilbert's
Hilbert's 10th:10th: TheThe followingfollowing setset isis undecidableundecidable::
{ { 〈 〈 p p 〉 〉 | | p p ∈ ∈
NN[ [ X X
11, , … … X X
nn], ], n n ∈ ∈
NN, , ∃ ∃ x x
11… … x x
nn∈ ∈ N N p p ( ( x x
11, , … … x x
nn)=0 } )=0 }
Word Problem
Word Problem forfor finitelyfinitely presentedpresented groupsgroups Mortality
Mortality Problem Problem forfor twotwo 2121××21 21 matricesmatrices Homeomorphy
Homeomorphy of 2 finite of 2 finite simplicialsimplicial complexescomplexes
Martin Ziegler
integer integer
Exercise Questions
Which
Which of of thethe followingfollowing areare unun--/semi/semi--//decidabledecidable?? a) a) GivenGiven an integer, an integer, isis itit a prime a prime numbernumber??
b) b) GivenGiven a finite a finite stringstring overover ++,,××,(,(,,)),,00,1,1,,
X X
11, , … … X X
n nisis itit syntacticallysyntactically correctcorrect??
c) c) GivenGiven a a BooleanBoolean formulaformula
ϕ ϕ ( ( X X
11, , … … X X
nn) )
,, doesdoes itit havehave a a satisfyingsatisfying assignmentassignment?? d) d) GivenGivenM M ∈ ∈
nn××nn and andb b ∈ ∈
nn, ,doesdoes therethere existsexists a a vectorvector
x x
s.t. s.t.M M · · x x ≤ ≤ b b
?? e) e) GivenGiven an an algorithmalgorithm AA, , inputinput xx, and integer , and integer NN,,doesdoes AA terminateterminate on on inputinput xx withinwithin NN stepssteps ??
f) Doesf) Does a givena given algorithmalgorithm terminateterminate on all on all inputsinputs?? g) g) DoesDoes givengiven algorithmalgorithm terminateterminate on on somesome inputinput??
realreal
Martin Ziegler
Computable Real Numbers
Theorem:
Theorem: For For r r ∈ ∈ , ,
t t he he following following are are equivalent equivalent : :
a) a) r r has a has a computable computable binary binary expansion expansion
b) b) There There is is an an algorithm algorithm printing printing , on , on input input n n ∈ ∈ , , some some a a ∈ ∈ with with | | r r - - a a /2 /2
n+1n+1| | ≤ ≤ 2 2
--nn. .
c) c) There There is is an an algorithm algorithm printing printing two two sequences
sequences ( ( q q
nn) ) ⊆ ⊆ and ( and ( ε ε
nn) ) with with | | r r - - q q
nn| | ≤ ≤ ε ε
nn→ → 0 0
H := { 〈 B,x 〉 : algorithm B terminates on input x } ⊆
There is an algorithm which, given n∈, prints bn∈{0,1} where r=∑n bn 2-n
b) ⇔ c) holds uniformly,
⇔ a) does not [Turing'37]
numerators+
denominators
Ernst
Ernst Specker Specker (1949): (c) (1949): (c) ⇔ ⇔ Halting Halting problem problem plus (d) plus (d) d) d) There There is is an an algorithm algorithm printing printing ( ( q q
nn) ) ⊆ ⊆ with with q q
nn→ → r r . .
interval arithmetic
⇔ r∈ [qn±εn]
Call Call r r ∈ ∈ computable computable if if
Martin Ziegler
Exercises: Computable Reals
r r ∈ ∈
computablecomputable iffiff an an algorithmalgorithm cancan printprint, , on on inputinputn n ∈ ∈
, , somesomea a
yetyet∈ ∈
naivelynaivelywithwith| | r r
computable-
computable- a a /2 /2
n+1n+1| | ≤ ≤ 2 2
-n-n ..a) a) EveryEvery rational has a rational has a computablecomputable binarybinary expansionexpansion b) Everyb) Every dyadicdyadic rational has tworational has two binarybinary expansionsexpansions c) Computablec) Computable binarybinary expansionexpansion ⇔ ⇔ computablecomputable realreal d) d) IfIf
a a , , b b
areare computablecomputable, , thenthen also alsoa+b a+b , , a a · · b b , , 1/ 1/ a a ( ( a a ≠ ≠ 0) 0)
e) Fix
e) Fix
p p ∈ ∈
[ [ X X ] ]
. . ThenThenp p
's's coefficientscoefficients areare computablecomputable⇔ ⇔ p p ( ( x x ) )
isis computablecomputable forfor all all computablecomputablex x
.. f) f) TheThe degreedegree of of everyeveryp p ∈ ∈
[ [ X X ] ]
isis computablecomputable..g) g) EveryEvery algebraicalgebraic numbernumber isis computablecomputable; and so ; and so isis
π π . .
h) h) IfIf
x x
isis computablecomputable, , thenthen so so areareexp( exp( x x ) ) , , sin( sin( x x ) ) , , log( log( x x ) )
j) For
j) For everyevery computablecomputable
x x
, ,sign( sign( x x ) )
isis computablecomputable.. k) k) Specker'sSpecker's sequencesequence( ( ∑ ∑
k>nk>n∈H∈H2 2
-n-n) )
kk isis computablecomputable,,itsits limitlimit isis uncomputableuncomputable
Martin Ziegler
Uniformity, Sequences and Equality Testing
In In numerics numerics , , don't don't test test for for ( ( in in - - )equality )equality ! !
Fact:
Fact: There exists a computable sequence ( There exists a computable sequence ( r r
mm) ) ⊆ ⊆ [0,1] [0,1]
such that {
such that { m m : : r r
mm≠ ≠ 0 } is the Halting problem 0 } is the Halting problem H H . .
H := { 〈 B,x 〉 : algorithm B terminates on input x } ⊆
Reminder:
Reminder: For For r r ∈ ∈ , , t t he he following following are are equivalent equivalent : : a) a) ∃ ∃ algorithm algorithm deciding deciding r r 's 's bin. bin. expansion expansion
b) b) ∃ ∃ algorithm algorithm printing printing on on input input n n some some a a ∈ ∈ with with | | r r - - a a /2 /2
n+1n+1| | ≤ ≤ 2 2
--nn. .
c) c) ∃ ∃ algorithm algorithm printing printing ( ( q q
nn),( ),( ε ε
nn) ) ⊆ ⊆ with with | | r r - - q q
nn| | ≤ ≤ ε ε
nn→ → 0 0 Call (r
m) ⊆ computable iff an algorithm can print, on input 〈 n,m 〉 ∈ , some a ∈ with |r
m-a/2
n+1|≤2
-n.
a)a)⇒⇒b)b)⇔c⇔c) ) computablecomputable transformation transformation on on algorithmsalgorithms b)b)⇒⇒aa) ') 'undecidableundecidable' '
casecase splitsplit on ron r∈∈
Martin Ziegler
x x ∈ ∈ computable computable ⇔ ⇔ | | x x - - a a
nn/2 /2
nn+1+1| | ≤ ≤ 2 2
--nnfor for recursive recursive ( ( a a
nn) ) ⊆ ⊆
f
Uniformly Computable Real Functions
A A computable computable function
function must must be be continuous continuous
x' x
Martin Ziegler
Computable Weierstrass Theorem
Theorem:
Theorem: For For f f :[0,1] :[0,1] → → the the following following are are equivalent equivalent : : a) a) There There is is an an algorithm algorithm converting converting any any seq seq . . q q
nn∈ ∈
n+1n+1with with | | x x - - q q
nn| | ≤ ≤ 2 2
--n ninto into p p
mm∈ ∈
m+1m+1with with | | f f ( ( x x ) ) - - p p
mm| | ≤ ≤ 2 2
-m-mb) b) There There is is an an algorithm algorithm printing printing a a sequence sequence (of (of degrees degrees
and and coefficient coefficient lists lists of) ( of) ( P P
nn) ) ⊆ ⊆ [X [X ] ] with with || || f f - - P P
nn|| || ≤ ≤ 2 2
-n-nc) c) The The real real sequence sequence f f ( ( q q ), ), q q ∈ ∈ ∩ ∩ [0,1], [0,1], is is computable computable
&
& f f admits admits a a computable computable modulus modulus of uniform of uniform continuity. continuity
Call Call ( ( r r
mm) ) ⊆ ⊆ computable computable iff iff an an algorithm algorithm can can print print , , on on input input n,m n,m ∈ ∈ , , some some q q ∈ ∈
n+1n+1with with | | r r
mm- - q q | | ≤ ≤ 2 2
--nn. .
:= :=
nnnn
, ,
nn:= := { { a a /2 /2
nn: : a a ∈ ∈ } }
| | x x - - y y | | ≤ ≤ 2 2
--µ(µ(m)m)⇒ ⇒ | | f f ( ( x x ) ) - - f f ( ( y y )| )| ≤ ≤ 2 2
--mmProof: Proof: a) a) ⇒ ⇒ c) c) ⇒ ⇒ b) b)
Martin Ziegler
uncomputable
uncomputable in in generalgeneral Exercises: Computable Real Functions
a) a) f f computable computable
⇒⇒same same for for any any restriction restriction
b) b) exp exp , , sin sin , , cos cos , ln(1+ , ln(1+ x x ) ) are are computable computable functions functions c) c) For For a a computable computable sequence sequence a a =( =( a a
nn), ),
the the power power series series x x → → ∑ ∑
nna a
nn· · x x
nnis is computable computable on on ( ( - - r,r r,r ) ) for for r r < < R R ( ( a a ) := 1/ ) := 1/ limsup limsup
nn| | a a
nn| |
1/n1/nd) d) Let Let f f ∈ ∈ C[0,1] C[0,1] be be computable computable . . Then Then so so are are
∫ ∫ f f : : x x → → ∫ ∫
00xxf f ( ( t t ) ) dt dt and and max( max( f f ): ): x x → → max{ max{ f f ( ( t t ): ): t t ≤ ≤ x x }. }.
e) e) If If ( ( x,m x,m ) ) →f → f
mm( ( x x ) ) computable computable with with | | f f
nn- - f f
mm| |
∞∞≤ ≤ 2 2
-n-n+2 +2
--mmthen
then
lim lim
nnf f
nn is computable.is computable.f) For
f) For computable computable a a ∈ ∈ , , f f :[0, :[0, a a ] ] → → , and , and
g To compute g To :[ :[ a a compute ,1] ,1] → → with with f f : : → → f f ( ( a a : : )= )= convert convert g g ( ( a a ) ) , , their their any any sequence sequence
joinjoin is is computable computable q q
nn∈ ∈
nn+1 +1with with | | x x - - q q
nn| | ≤ ≤ 2 2
--n ninto into p p
mm∈ ∈
m+1m+1with with | | f f ( ( x x ) ) - - p p
mm| | ≤ ≤ 2 2
--mmMartin Ziegler
Computable Urysohn
Proof
Proof : : Let Let f f ( ( x x ):= ):= ∑ ∑
mmrrmm rrmm++εεmm rrmm--εεmm
εεmm//22mm
Let Let ( ( r r
mm) )
mm
, , ( ( ε ε
mm) )
mm
⊆ ⊆ Q Q be be computable computable sequences sequences
Then Then there there is is a a computable computable f f :[0;1] :[0;1] → → [0;1] [0;1]
s.t.
s.t. f f
--11[0] = [0;1] [0] = [0;1] \ \
mm( ( r r
mm- - ε ε
mm, , r r
mm+ + ε ε
mm) ) . . max(0,
max(0, ε ε m m - - | | x x - - r r
mm|)/ |)/ 2 2
mmC C
∞∞CC∞∞ ''pulse' functionpulse' function
φφ((tt) = ) = exp(exp(--tt²²/1/1--tt²²))
||tt|<1|<1
Martin Ziegler
Specker'59: Uncomputable roots/argmin
approximating approximating a a root root vs vs . . approximate approximate root root
Lemma:
Lemma: There There are are computable computable sequences sequences
( ( r r
mm) )
mm
, , ( ( ε ε
mm) )
mm
⊆ ⊆ Q Q s.t. s.t. U U := :=
mm( ( r r
mm- - ε ε
mm, , r r
mm+ + ε ε
mm) )
contains
contains all all computable computable reals reals in [0;1] in [0;1]
and has
and has measure measure < < ½ ½ . .
Corollary
Corollary : : There There is is a a computable computable C C
∞∞f f :[0;1] :[0;1] → → [0;1] [0;1] s.t. s.t. f f
--11[0] [0] has has measure measure > > ½ ½
but but contains contains no no computable computable real real number number . . Let Let ( ( r r
mm) )
mm
, , ( ( ε ε
mm) )
mm
⊆ ⊆ Q Q be be computable computable sequences sequences
Then Then there there is is a a computable computable f f :[0;1] :[0;1] → → [0;1] [0;1]
s.t.
s.t. f f
--11[0] = [0;1] [0] = [0;1] \ \
mm( ( r r
mm- - ε ε
mm, , r r
mC C
m+ + ε ε
mm) ) . .
∞∞
Martin Ziegler
Singular Covering of Computable Reals
Lemma:
Lemma: There There are are computable computable sequences sequences
( ( r r
mm) )
mm
, , ( ( ε ε
mm) )
mm
⊆ ⊆ Q Q s.t. s.t. U U := :=
mm( ( r r
mm- - ε ε
mm, , r r
mm+ + ε ε
mm) )
contains
contains all all computable computable reals reals in [0;1] in [0;1]
and has
and has measure measure < < ½ ½ . . Proof
Proof : : Dove Dove - - tailing tailing w.r.t w.r.t . ( . ( M M , , t t ) ) : : If If Turing Turing machine machine # # M M within within t t
( ( but but not not t t - - 1 1 ) ) steps steps prints prints a a
11, , … … a a
MM+5+5s.t.
s.t. | | a a
kk/2 /2
kk+1+1- - a a
ℓℓ/2 /2
ℓℓ+1+1| | ≤ ≤ 2 2
--kk+2 +2
--ℓℓ∀1 ∀ 1 ≤ ≤ k k , , ℓ ℓ ≤ ≤ M M +5 +5
then then let let r r
〈〈M,t〉M,t〉:= := a a
MM+5+5/2 /2
M+6M+6and and ε ε
〈〈M,tM,t〉〉:= := 2 2
-M-M--55, , else else r r
〈〈M,tM,t〉〉:= := 0 0 and and ε ε
〈〈M,t〉M,t〉:= 2 := 2
--〈〈M,t〉M,t〉-3-3. .
Machine
Machine computescomputes
r r ∈ ∈ R R
iffiff prints seq. prints seq.
a a
nn⊆⊆ withwith| | a a
nn/2 /2
n+1 n+1- - a a
mm/2 /2
m+1 m+1| | ≤ ≤ 2 2
--n n+ + 2 2
--mm. .
Martin Ziegler
• • g g
kk( ( x x ):= ):= g g ( ( ( ( x x · · 2 2
ψψ((kk))- - 1) 1) · · 2 2
kk) ) / / 2 2
ψψ(k(k))h' h' := := ∑ ∑
kkg g
kk• • g g
kk( ( x x ):= ):= g g ( ( x x · · 2 2
ψψ((kk))- - 1 1 ) ) / / 2 2
ψψ(k(k))n n =1 =1 n n =2 =2
½½
¼¼
Fact Fact : : ∃ ∃ computable computable bijection bijection ψ ψ : : → → H H
• • g g and and ∫ ∫ g g computable computable
⅛⅛
½½
¼¼
⅛⅛ 11
00
Myhill'71:
Myhill'71: uncomputable uncomputable ∂∂∂∂ ∂∂∂∂ on on C C
11[0,1] [0,1]
yet yet h h := := ∫ ∫ h' h' ∈ ∈ C C
11[0;1] [0;1] computable computable . .
incomputable incomputable , ,
hat hat function function g g
• • g g
nn( ( x x ):= ):= g g ( ( x x · · 2 2
nn- - 1 1 ) ) /2 /2
nn• • ∫ ∫ g g
kk≤ ≤ 2 2
--kkcontinuous continuous , ,
∑ ∑ ∑
nn∈∈HHg g
nn∑
nng g
nne.g e.g . . H H ={2,3,5,...} ={2,3,5,...}
q.e.d
q.e.d . .
Martin Ziegler
The Case of the Wave Equation
Pour Pour - - El&Richards'81 construct a computable El&Richards'81 construct a computable ƒ∈ ƒ∈ C C
11( (
33) ) such that for
such that for g g :=0 the unique solution is :=0 the unique solution is in in computable at computable at t t =1 and =1 and x x =(0,0,0). =(0,0,0).
∂ ²/ ∂ t² u(x,t) = ∆ u(x,t), u(x,0)= ƒ (x), ∂ / ∂ t u(x,0)=g(x)
Church
Church - - Turing Turing Hypothesis Hypothesis ( ( Kleene Kleene ): ):
Everything
Everything that that can can be be computed computed by by a a Turing
Turing machine machine can can also also be be computed computed by by a a physical physical device device – – and and vice vice versa versa ! !
Myhill'71:
Myhill'71: computable computable h h ∈ ∈ C C 1 1 [0,1] [0,1]
with with uncomputable uncomputable h h '(1) '(1)
Martin Ziegler
The Case of the Wave Equation
Myhill'71:
Myhill'71: computable computable h h ∈ ∈ C C 1 1 [0,1] [0,1]
with with uncomputable uncomputable h h '(1) '(1)
Pour Pour - - El&Richards'81 construct a computable El&Richards'81 construct a computable ƒ∈ ƒ∈ C C
11( (
33) ) such that for
such that for g g :=0 the unique solution is :=0 the unique solution is in in computable. computable.
∂ ²/ ∂ t² u(x,t) = ∆ u(x,t), u(x,0)= ƒ (x), ∂ / ∂ t u(x,0)=g(x)
Kirchhoff's Kirchhoff's
formula
formula : :
Martin Ziegler
up to
up to permutationpermutation [Specker'67][Specker'67]
Example
Example fund. fund. theoremtheorem of of algebraalgebra:: Given
Given
a a
00, , … … a a
d-d-11∈ ∈
, , returnreturn rootsrootsx x
11, , … … x x
dd∈ ∈
of ofa a
00+a +a
11· · X+ X+ … … +a +a
d-d-11· · X X
d-d-11+X +X
dd∈ ∈ [ [ X X ] ]
incl. multiplicitiesincl. multiplicitiesTwo Effects in Real Computability a) a) Multivalued Multivalued ' ' functions functions ' '
b) b) Discrete Discrete ' ' advice advice ' '
Example
Example floorfloor functionfunction: : givengiven
x x ∈ ∈
, , returnreturn itsits least integerleast integer upperupper boundbound Given
Given
x x
, , returnreturn somesome integer integer upperupper bound: bound: computablecomputable!!
Example
Example matrixmatrix diagonalizationdiagonalization: : givengiven
A A ∈ ∈
d·d·(d(d--1)/21)/2,, returnreturn a a basisbasis of of eigenvectorseigenvectors
ThmThm:: ComputableComputable knowingknowing
| | σ σ ( ( A A )| )|
..ε·cos(1/ε) sin(1/ε) sin(1/ε) −cos(1/ε)
―
― discontinuousdiscontinuous::
―― discontinuousdiscontinuous..