The Setting
Finitely many nonzero vectors and widths cut out a symmetric polyhedron, the intersection of the slabs . The soft slab law replaces its indicator by a Gibbs weight. Each slab contributes a factor , with a high even power of , so that the factor is close to inside the slab and to outside, and a Gaussian factor with a fixed makes the weight integrable. Its total mass is a smoothed volume of the polyhedron.
Let the widths move geometrically from at to at , as support values do in the logarithmic Brunn-Minkowski inequality. Is the smoothed volume log-concave in ? The answer passes through a general Gibbs family, whose potential depends on through exponential weights with real rates on fixed nonnegative polynomials, and a bound of a variance by a mean along that family, which for slab features is the soft variance inequality.
The Gibbs Family
Fix , a finite set , polynomials on and real numbers , the rates, for ; the Gibbs family determined by these data is the family of probability measures , indexed by a real parameter , defined below. For an integer and a real parameter , the variation of order is the polynomial (1)✦ the th derivative in of . The regularized potential of the family is the function on given by (2)✦ and we write and for its first two derivatives in .
The Gibbs density of the family at is the function on ✦, and its partition function is the function of given by (3)✦ the Gibbs law of the family at is the probability measure on . The Gibbs mean and the Gibbs variance of a real function on are (4) The first and second partition derivatives are the integrals of the first and second -derivatives of the Gibbs density: (5)
Slab Features
For a vector , a width and an integer , the feature polynomial of and is the th power of the normalized linear form .✦ The feature family of vectors and widths , , is the family of their feature polynomials, (6)✦ For two widths , the decay rate of exponent is a real number and the interpolated width at a positive one, given by (7)
The soft slab law of nonzero vectors and widths , with an even exponent , is the law of the Gibbs family of their feature family: with .
The soft variance inequality is the following claim. Let , let be a finite nonempty set, let be an even integer and , let , , be nonzero vectors, widths and real numbers, and let be the soft slab law of the and . Then (8)✦
The Claim
Theorem 1. Let and let be a finite set.
Let , , be polynomials on that are nonnegative everywhere, and let be real numbers. For every real , (9)✦ If moreover for every real , then for all and (10)✦ and for every (11)✦
Let be vectors, widths and an integer; let be the feature family of the and the , and the decay rates of exponent . For every integer , every real and every , (12)✦
Assume the soft variance inequality. Let and be nonempty, and let and be as in (b), with nonzero vectors and an even exponent . For every real , (13)✦
The left side of (9) is the expression of through and , and (13) is the hypothesis of (10) and (11) for the feature family with the decay rates.
What Is Checked
The claim is stated over the Gibbs family of the regularized potential (2), with its variations, partition function, partition derivatives, Gibbs mean and Gibbs variance, over the feature family, the decay rates and the interpolated widths, and, in (13), under the soft variance inequality as a hypothesis.