Seite - 444 - in Differential Geometrical Theory of Statistics
Bild der Seite - 444 -
Text der Seite - 444 -
Entropy2016,18, 110
• Throughout theentiresetof28 languagesconsidered in [3], thefirst twoparametersareset to the
samevalue1,hence for thepurposeofcomparativeanalysiswithin this family,wecanregarda
code like theaboveasa twicespoiledcodeC=C′ 1 f1=(C′′ 2 f2) 1 f1whereboth f1 and f2
areconstantequal to1andC′′ ⊂F42 is thecodeobtainedfromtheabovebycancelingthefirst two
letters ineachcodeword.
• Conversely, we have C′′ = C′ 2 and C′ = C 1, in terms of the second spoiling operation
describedabove.
• Toillustrate the thirdspoilingoperation,onecansee, for instance, thatC(0,4)={ 1, 3},while
C(1,6)={ 2, 3}.
2.4. TheAsymptoticBound
Thespoilingoperationsoncodeswereusedin[13] toprovetheexistenceofanasymptoticbound in
thespaceofcodeparameters (δ,R), seealso [16,24,25] formoredetailedpropertiesof theasymptotic
bound.
LetVq⊂ [0,1]2∩Q2 denote thespaceofcodeparameters (δ,R)ofcodesC⊂Fnq andletUqbethe
setofall limitpointsofVq. ThesetUq is characterized in [13]as
Uq={(δ,R)∈ [0,1]2 |R≤αq(δ)}
foracontinuous,monotonicallydecreasingfunctionαq(δ) (theasymptoticbound).Moreover, code
parameters lying inUq arerealizedwith infinitemultiplicity,whilecodepoints inVq\(Vq∩Uq)have
finitemultiplicityandcorrespondto the isolatedcodes, see [13,16].
Codes lyingabove theasymptoticboundarecodeswhichhaveextremelygoodtransmissionrate
andrelativeminimumdistance,henceverydesirable fromthecodingtheoryperspective. The fact that
thecorrespondingcodeparametersarenot limitpointsofothercodeparametersandonlyhavefinite
multiplicityreflect the fact thatsuchcodesareverydifficult toreachorapproximate. Isolatedcodes
areknowntoarise fromalgebro-geometricconstructions, [14,15].
Relatively little is known about the asymptotic bound: the question of the computability of
the function αq(δ)was recently addressed in [25] and the relation toKolmogorov complexitywas
investigated in [16]. There are explicit upper and lower bounds for the function αq(δ), see [12],
includingthePlotkinbound
αq(δ)=0, for δ≥ q−1q ; (5)
thesingletonbound,which implies thatR=αq(δ) liesbelowthe lineR+δ=1; theHammingbound
αq(δ)≤1−Hq(δ2), (6)
whereHq(x) is theq-aryShannonentropy
x logq(q−1)−x logq(x)−(1−x) logq(1−x)
which is theusualShannonentropyforq=2,
H2(x)=−x log2(x)−(1−x) log2(1−x). (7)
Onealsohasa lowerboundgivenbytheGilbert–Varshamovbound
αq(δ)≥1−Hq(δ) (8)
444
Differential Geometrical Theory of Statistics
- Titel
- Differential Geometrical Theory of Statistics
- Autoren
- Frédéric Barbaresco
- Frank Nielsen
- Herausgeber
- MDPI
- Ort
- Basel
- Datum
- 2017
- Sprache
- englisch
- Lizenz
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03842-425-3
- Abmessungen
- 17.0 x 24.4 cm
- Seiten
- 476
- Schlagwörter
- Entropy, Coding Theory, Maximum entropy, Information geometry, Computational Information Geometry, Hessian Geometry, Divergence Geometry, Information topology, Cohomology, Shape Space, Statistical physics, Thermodynamics
- Kategorien
- Naturwissenschaften Physik