Hadamard three-lines Theorem #
In this file we present a proof of Hadamard's three-lines Theorem.
Main result #
norm_le_interp_of_mem_verticalClosedStrip: Hadamard three-line theorem: Iffis a bounded function, continuous onre ⁻¹' [l, u]and differentiable onre ⁻¹' (l, u), then forM(x) := sup ((norm ∘ f) '' (re ⁻¹' {x})), that isM(x)is the supremum of the absolute value offalong the vertical linesre z = x, we have that∀ z ∈ re ⁻¹' [l, u]the inequality‖f(z)‖ ≤ M(0) ^ (1 - ((z.re - l) / (u - l))) * M(1) ^ ((z.re - l) / (u - l))holds. This can be seen to be equivalent to the statement thatlog M(re z)is a convex function on[0, 1].norm_le_interp_of_mem_verticalClosedStrip': Variant of the above lemma in simpler terms. In particular, it makes no mention of the helper functions defined in this file.
Main definitions #
Complex.HadamardThreeLines.verticalStrip: The vertical strip defined by :re ⁻¹' Ioo a bComplex.HadamardThreeLines.verticalClosedStrip: The vertical strip defined by :re ⁻¹' Icc a bComplex.HadamardThreeLines.sSupNormIm: The supremum function on vertical lines defined by :sSup {|f(z)| : z.re = x}Complex.HadamardThreeLines.interpStrip: The interpolation between thesSupNormImon the edges of the vertical stripre⁻¹ [0, 1].Complex.HadamardThreeLines.interpStrip: The interpolation between thesSupNormImon the edges of any vertical strip.Complex.HadamardThreeLines.invInterpStrip: Inverse of the interpolation between thesSupNormImon the edges of the vertical stripre⁻¹ [0, 1].Complex.HadamardThreeLines.F: Function defined byftimesinvInterpStrip. Convenient form for proofs.
Note #
The proof follows from Phragmén-Lindelöf when both frontiers are not everywhere zero.
We then use a limit argument to cover the case when either of the sides are 0.
The vertical strip in the complex plane containing all z ∈ ℂ such that z.re ∈ Ioo a b.
Equations
Instances For
The vertical strip in the complex plane containing all z ∈ ℂ such that z.re ∈ Icc a b.
Equations
Instances For
The supremum of the norm of f on imaginary lines. (Fixed real part)
This is also known as the function M
Equations
- Complex.HadamardThreeLines.sSupNormIm f x = sSup (norm ∘ f '' (Complex.re ⁻¹' {x}))
Instances For
The inverse of the interpolation of sSupNormIm on the two boundaries.
In other words, this is the inverse of the right side of the target inequality:
|f(z)| ≤ |M(0) ^ (1-z)| * |M(1) ^ z|.
Shifting this by a positive epsilon allows us to prove the case when either of the boundaries is zero.
Equations
- Complex.HadamardThreeLines.invInterpStrip f z ε = (↑ε + ↑(Complex.HadamardThreeLines.sSupNormIm f 0)) ^ (z - 1) * (↑ε + ↑(Complex.HadamardThreeLines.sSupNormIm f 1)) ^ (-z)
Instances For
A function useful for the proofs steps. We will aim to show that it is bounded by 1.
Equations
- Complex.HadamardThreeLines.F f ε z = Complex.HadamardThreeLines.invInterpStrip f z ε • f z
Instances For
sSup of norm is nonneg applied to the image of f on the vertical line re z = x
sSup of norm translated by ε > 0 is positive applied to the image of f on the
vertical line re z = x
Useful rewrite for the absolute value of invInterpStrip
The function invInterpStrip is diffContOnCl.
If f is bounded on the unit vertical strip, then f is bounded by sSupNormIm there.
Alternative version of norm_le_sSupNormIm with a strict inequality and a positive ε.
When the function f is bounded above on a vertical strip, then so is F.
The interpolation of sSupNormIm on the two boundaries.
In other words, this is the right side of the target inequality:
|f(z)| ≤ |M(0) ^ (1-z)| * |M(1) ^ z|.
Note that if sSupNormIm f 0 = 0 ∨ sSupNormIm f 1 = 0 then the power is not continuous
since 0 ^ 0 = 1. Hence the use of ite.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Rewrite for InterpStrip when 0 < sSupNormIm f 0 and 0 < sSupNormIm f 1.
Rewrite for InterpStrip when 0 = sSupNormIm f 0 or 0 = sSupNormIm f 1.
Rewrite for InterpStrip on the open vertical strip.
The correct generalization of interpStrip to produce bounds in the general case.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The transformation on ℂ that is used for scale maps the closed strip re ⁻¹' [l, u]
to the closed strip re ⁻¹' [0, 1].
The norm of the function scale f l u is bounded above on the closed strip re⁻¹' [0, 1].
The supremum of the norm of scale f l u on the line z.re = 0 is the same as the supremum
of f on the line z.re = l.
The supremum of the norm of scale f l u on the line z.re = 1 is the same as
the supremum of f on the line z.re = u.
A technical lemma relating the bounds given by the three lines lemma on a general strip to the bounds for its scaled version on the strip ``re ⁻¹' [0, 1]`.
Hadamard three-line theorem on re ⁻¹' [0, 1]: If f is a bounded function, continuous on the
closed strip re ⁻¹' [0, 1] and differentiable on open strip re ⁻¹' (0, 1), then for
M(x) := sup ((norm ∘ f) '' (re ⁻¹' {x})) we have that for all z in the closed strip
re ⁻¹' [0, 1] the inequality ‖f(z)‖ ≤ M(0) ^ (1 - z.re) * M(1) ^ z.re holds.
Hadamard three-line theorem on re ⁻¹' [0, 1] (Variant in simpler terms): Let f be a
bounded function, continuous on the closed strip re ⁻¹' [0, 1] and differentiable on open strip
re ⁻¹' (0, 1). If, for all z.re = 0, ‖f z‖ ≤ a for some a ∈ ℝ and, similarly, for all
z.re = 1, ‖f z‖ ≤ b for some b ∈ ℝ then for all z in the closed strip
re ⁻¹' [0, 1] the inequality ‖f(z)‖ ≤ a ^ (1 - z.re) * b ^ z.re holds.
The transformation on ℂ that is used for scale maps the strip re ⁻¹' (l, u)
to the strip re ⁻¹' (0, 1).
If z is on the closed strip re ⁻¹' [l, u], then (z - l) / (u - l) is on the closed strip
re ⁻¹' [0, 1].
The function scale f l u is diffContOnCl.
Hadamard three-line theorem: If f is a bounded function, continuous on the
closed strip re ⁻¹' [l, u] and differentiable on open strip re ⁻¹' (l, u), then for
M(x) := sup ((norm ∘ f) '' (re ⁻¹' {x})) we have that for all z in the closed strip
re ⁻¹' [a,b] the inequality
‖f(z)‖ ≤ M(0) ^ (1 - ((z.re - l) / (u - l))) * M(1) ^ ((z.re - l) / (u - l))
holds.
Hadamard three-line theorem (Variant in simpler terms): Let f be a
bounded function, continuous on the closed strip re ⁻¹' [l, u] and differentiable on open strip
re ⁻¹' (l, u). If, for all z.re = l, ‖f z‖ ≤ a for some a ∈ ℝ and, similarly, for all
z.re = u, ‖f z‖ ≤ b for some b ∈ ℝ then for all z in the closed strip
re ⁻¹' [l, u] the inequality
‖f(z)‖ ≤ a ^ (1 - (z.re - l) / (u - l)) * b ^ ((z.re - l) / (u - l))
holds.