Gelfand-Tsetlin patterns are semistandard Young tableaux #
A Gelfand-Tsetlin pattern whose top row is a shape μ records a chain of shapes
∅ = ν⁰ ⊆ ν¹ ⊆ ⋯ ⊆ νⁿ = μ,
row j being the row-length sequence of νʲ; consecutive shapes differ by a horizontal strip
because consecutive rows interlace. Such a chain is the same thing as a semistandard Young
tableau of shape μ with entries in {0, …, n - 1}: the cells carrying the entry j are
exactly those of νʲ⁺¹ and not of νʲ. This file builds that bijection.
Both directions are counting maps, and both rest on the same principle: a downward closed
predicate on ℕ cuts an initial segment out of Finset.range N, so membership in the segment is
decided by comparison with its cardinality. Concretely, the tableau attached to a pattern P
has
T i c = #{j < n | λᵢ,ⱼ₊₁ ≤ c},
the number of rows of P that do not yet reach past c: for a cell (i, c) of μ this is one
less than the number of the first shape of the chain that contains it, so that the cells carrying
the entry j are exactly those of νʲ⁺¹ and not of νʲ. The pattern attached to a tableau T
has λᵢ,ⱼ the number of cells (i, c) of μ with T i c < j, that is, the length of row i of
the shape filled by the entries below j. The interlacing inequalities and the semistandardness
conditions are exchanged by these two formulas, and the two maps are mutually inverse.
Nonnegativity of the entries is not assumed on TauCeti.GTPattern, which admits the
determinant-twisted patterns; it is a consequence of the top row being a shape, since every
entry dominates a top-row entry (TauCeti.GTPattern.entry_nonneg). Only in this polynomial
regime is there a tableau to speak of.
Main definitions #
TauCeti.GTPattern.tableauEntryandTauCeti.GTPattern.toTableau: the tableau entryT i cread off a pattern, and the semistandard Young tableau it assembles.SemistandardYoungTableau.patternEntryandSemistandardYoungTableau.toGTPattern: the pattern read off a tableau.TauCeti.gtPatternEquivSSYT: the bijection, between the patterns withnrows and top row the shapeμand the semistandard Young tableaux of shapeμwith entries belown, that is,TauCeti.BoundedSSYT n μ.
Main results #
TauCeti.GTPattern.lt_tableauEntry_iffandSemistandardYoungTableau.lt_card_filter_rowLen_iff: the two counting maps are decided by the pattern entries, respectively by the tableau entries. Everything else is read off these two comparisons.TauCeti.card_gtPattern_topRow_eq_card_ssyt: the patterns with a given top row are as many as the semistandard tableaux of the corresponding shape with bounded entries, both counts being finite byTauCeti.GTPattern.finite_topRow_eqandTauCeti.finite_boundedSSYT.
Implementation notes #
With the bound ∀ i j, T i j < n on every entry of the tableau, the bijection fails in exactly one
case: for n = 0 and μ = ⊥ there is one (empty) pattern but no tableau at all, because the bound
also constrains the entries off μ, which vanish. The bound is therefore imposed on the cells of
μ, where it is the intended condition; whenever μ is nonempty the two readings agree, since a
nonempty shape forces n ≠ 0.
References #
- W. Fulton and J. Harris, Representation Theory: A First Course (1991), §15.3.
Initial segments cut out by a downward closed predicate #
Patterns whose top row is a shape #
The top row of a pattern whose top row is the shape μ is the row-length sequence of μ at
every index: past n both sides vanish, because μ has at most n rows.
From a pattern to a tableau #
The (i, c) entry of the tableau attached to a Gelfand-Tsetlin pattern: the number of rows
j < n of the pattern whose i-th entry is at most c. The counted rows form an initial
segment, so when some row v < n has c < λᵢ,ᵥ₊₁ the count is the least such v: the cell
(i, c) first appears in the shape νᵛ⁺¹ of the chain recorded by the pattern, one step further
along the chain than the value v carried by the cell. When no row does, every row is counted
and the entry is n.
Instances For
The tableau entry of a pattern, unfolded: the count of the rows j < n whose i-th entry is
at most c. The body of TauCeti.GTPattern.tableauEntry is not exposed, so this is how the
count is reached.
The tableau entry, read off the pattern: j < T i c exactly when j < n and the i-th
entry of row j + 1 is at most c. The i-th entries of a pattern increase weakly with the row
index on the informative cells and vanish on the rest, where the bound holds anyway since c is a
natural number, so the condition is downward closed in j and the count is an initial segment.
Tableau entries are bounded by the number of rows of the pattern: a cell strictly inside the top row is reached before the last shape of the chain.
Tableau entries increase strictly down a column, one step: the count m for row i is beaten
by the count for row i + 1, because the second interlacing inequality carries each of the m
rows witnessing the bound for row i one step up, while row 1 supplies a further witness.
Tableau entries increase strictly down a column. Only the rows i₁ ≤ i < i₂ that the chain of
one-step comparisons passes through need to be bounded.
The semistandard Young tableau of a Gelfand-Tsetlin pattern whose top row is the shape
μ: the cell (i, c) carries the entry j for which νʲ⁺¹ is the first shape in the chain that
contains it, one less than the number of that shape.
Instances For
The tableau of a pattern, unfolded: off μ the entry vanishes, and on μ it is the count
TauCeti.GTPattern.tableauEntry.
The defining comparison for the tableau of a pattern: the entry at a cell of μ is below j
exactly when the i-th entry of row j of the pattern runs past c.
From a tableau to a pattern #
The (i, j) entry of the pattern attached to a semistandard Young tableau, as a pattern with
n rows: for j ≤ n it is the number of cells of row i of μ whose entry is below j, that
is, the length of row i of the shape filled by the entries < j. Past the top row, n < j,
the entry is forced to 0, as a pattern with n rows carries no data there.
Instances For
The pattern entry of a tableau, unfolded: 0 past the top row, and otherwise the count of the
cells of row i of μ carrying an entry below j. The body of
SemistandardYoungTableau.patternEntry is not exposed, so this is how the count is
reached.
The pattern entry, read off the tableau: c is counted in row j of the i-th row
exactly when (i, c) is a cell of μ carrying an entry below j. The rows of a tableau
increase weakly, so the condition is downward closed in c.
The Gelfand-Tsetlin pattern of a semistandard Young tableau, with n rows: row j is
the row-length sequence of the shape filled by the entries < j. No bound on the entries is
needed to build the pattern; a bound is what makes its top row the whole of μ
(SemistandardYoungTableau.topRow_toGTPattern).
Instances For
The pattern of a tableau, unfolded.
The top row of the pattern of a tableau of shape μ is the row-length sequence of μ: every
entry lies below n, so the last shape of the chain is μ itself.
The defining comparison for the pattern of a tableau: the i-th entry of row j + 1 is at
most c exactly when the cell (i, c) carries an entry above j.
The bijection #
Gelfand-Tsetlin patterns are semistandard Young tableaux. For a shape μ with at most
n rows, the patterns with n rows and top row μ correspond to the semistandard Young tableaux
of shape μ whose entries lie in {0, …, n - 1}: row j of the pattern is the row-length
sequence of the shape filled by the entries below j, and the cell (i, c) of the tableau
carries the number of rows of the pattern whose i-th entry does not yet run past c, one less
than the index of the first row that does.
Together with the count of the patterns with a given top row, this is the tableau reading of the
Gelfand-Tsetlin basis of an irreducible polynomial representation of GL n.
Instances For
The bijection sends a pattern to the tableau TauCeti.GTPattern.toTableau it names.
The inverse bijection sends a tableau to the pattern
SemistandardYoungTableau.toGTPattern it names.
The pattern count is the tableau count: the patterns with top row the shape μ are as
many as the semistandard Young tableaux of shape μ with entries below n. This is the form in
which the bijection feeds the Gelfand-Tsetlin dimension count.