Basic inequality for the Lasso Created 2026-09-24 Updated 2026-11-05
Suppose and minimizes the Lasso objective . With , comparing its value with the value at gives
Indeed, expand and cancel . This deterministic inequality needs neither a normal distribution nor a full-rank design matrix.[ref]
The noise term is at most by Holder inequality. Controlling that term probabilistically leads to a Lasso cone condition; a Compatibility condition for the Lasso then converts coefficient error into prediction error. The stronger prediction inequality from Lasso stationarity uses subgradient optimality and has a different leading constant.
Block mixed norm 2026-10-06
The block mixed norm sums the norm of each block, for . Its dual norm is , where for and for . Apply Holder inequality blockwise and choose a norming block to attain equality. Absolute values are essential for signed components.
The displayed spikes have norm one. Pairing with and applying Holder inequality bounds the pairing by , which tends to zero. At exponent one, however, cannot converge weakly in : every interval away from zero forces any limit to vanish there, while the constant test function preserves integral one. Their failure of uniform integrability records exactly the obstruction in the Dunford-Pettis theorem.
Cubic gradient nonlinearity Created 2026-09-28 Updated 2026-11-05
On a fixed bounded smooth three-dimensional domain, the nonlinear map
sends into . The Sobolev embedding theorem gives , and Holder inequality yields
The domain regularity is part of the embedding hypotheses.
For real , write . The difference identity
and the same embeddings give
Thus its local Lipschitz continuity constant on a ball of radius is . Composing this small nonlinear bound with a bounded linear elliptic inverse gives a contraction mapping on a sufficiently small ball, which is the mechanism of small-data existence for a nonlinear elliptic Dirichlet problem.
Let and use Nodewise Lasso with and nonzero design columns. Form with entry equal to one and other entries , and put row of equal to . The nodewise-Lasso residual identity and the Karush-Kuhn-Tucker conditions show that the diagonal of equals one and its row- off-diagonal entries have magnitude at most . For the Debiased Lasso,
The displayed bound follows row by row from Holder inequality. If the noise is conditionally multivariate normal with covariance , and the nodewise fits and tuning parameters depend only on , the first term is conditionally normal with covariance .
For and in an Lp space, the derivative with respect to the real scalar has the displayed form; the integrand is zero where . The pointwise chain rule gives the derivative, while the Holder inequality makes integrable and permits the dominated convergence theorem. This variational derivative characterizes nearest points to closed convex sets and constructs the Lp duality density of a bounded linear functional.
If satisfy , then
It follows by induction from the two-factor Holder inequality.
Group Lasso 2026-09-28
The group Lasso partitions a coefficient vector into groups and replaces the coordinatewise penalty with a weighted sum of group norms,
It tends to retain or discard an entire group at once. Its dual norm is
so the corresponding Holder inequality is .
H1 L3 interpolation inequality in three dimensions Created 2026-09-28 Updated 2026-11-05
On a bounded Lipschitz domain , the Sobolev inequality gives . The Holder inequality gives
The regular-domain hypothesis supplies the -to- Sobolev embedding theorem; boundedness of an arbitrary open set alone does not provide that embedding for all functions. On , the homogeneous Sobolev inequality instead gives for .
For , nonnegative consumption with state-price budget satisfies this bound under CRRA utility. Apply Holder inequality on probability-times-time measure to and , with conjugate exponents and . If the price integral is finite, equality forces full budget use and . Financing this process is a separate feasibility issue, so the equality pattern alone does not assert attainability in an incomplete market.
For a zero-starting continuous local martingale , the stochastic exponential satisfies
The Holder inequality and the supermartingale property of the first exponential bound its stopped th moment by . The subtraction in the coefficient is , not a square root of .
Lavrentiev phenomenon 2026-10-31
A Lavrentiev gap is a strict change in an infimum when the admissible regularity class is restricted, despite retaining the same endpoint values. For example with has value zero at the absolutely continuous function . Every Lipschitz competitor has a uniform positive lower energy bound. Indeed its ratio to starts at zero, so let be its first crossing of . Before that point, . Holder inequality gives
Thus . This proves a gap directly. The example's integrand lacks a uniform superlinear lower bound along its zero set; it illustrates the admissible-class issue rather than by itself furnishing a counterexample under stronger coercivity assumptions.
For a random variable with finite fourth moment, apply Holder inequality with conjugate exponents and to . This gives . Raising to the power proves the displayed Lp norm interpolation inequality.
Lp inclusion on a finite measure space Created 2026-09-24 Updated 2026-10-14
On a finite measure space, for . For , Hölder's inequality gives
with the same formula interpreted as when .
Lp interpolation inequality Created 2026-10-06 Updated 2026-10-14
If , and , then the displayed bound follows by applying the Holder inequality to . Interpret the endpoint cases using the essential supremum. In particular convergence in Lp space and boundedness in a larger-exponent Lp space imply convergence at intermediate exponents.
For in the finite moment-generating function domain and integrable , Jensen inequality gives the lower bound and almost surely gives the upper bound. The Holder inequality makes convex with . Thus for , and the effective bandwidth is nondecreasing. It need not be strictly increasing: deterministic traffic gives a constant bandwidth. Infinite peaks or exponential moments require extended-value conventions.
For a cube of side length in , , and a continuously differentiable function , its average satisfies
Averaging the fundamental theorem of calculus along a line segment from to and changing variables gives . The Holder inequality applies because , and the kernel norm is . Approximation extends the estimate to the continuous representative of a Sobolev space function. On it yields a Hölder continuous function of exponent representing every class.
Morrey's inequality 2026-10-06
For , every Sobolev space element has a Hölder continuous function representative of exponent , with the displayed bound. It also satisfies . Averaging the fundamental theorem of calculus along a line segment over a ball gives
The Holder inequality bounds this by , because . To compare two ball averages, translate a ball along the segment between their centres and use the same fundamental theorem of calculus along a line segment. These bounds give the displayed estimate. Density of smooth functions in a Sobolev space supplies the representative for nonsmooth . For , the corresponding conclusion is Lipschitz continuity.
Suppose , , and . Set . The part of where is at most . On the complementary level set, the Holder inequality bounds the integral by , proving the bound. For the bound becomes . The lower-exponent bound prevents spreading into low amplitudes, while the higher-exponent bound prevents concentration into arbitrarily narrow spikes.
Norming vector for Holder inequality Created 2026-09-24 Updated 2026-11-05
Let , let , and take nonzero . For the complex pairing , put
Since , direct summation gives and . This vector attains equality in the Holder inequality. For the bilinear pairing , conjugate the displayed phase instead.
At , the norming vector lies in and is simply the phase of , with the same pairing convention. At , an norming vector need not exist. For , the supremum norm is1 but no coordinate has modulus1. Every of norm1 then satisfies , though coordinate unit vectors approach the bound. Endpoint duality therefore cannot be handled by blindly inserting into the finite-exponent formula.