Exponential square root graphical model

⚠️ Curated Limitations & Negative Results

• Strict negativity on nonnegative directions is sufficient, not an exhaustive characterization of boundary parameters. For example, in one dimension Phi=0 and theta<0 still gives a finite integral 2/theta^2. This example does not classify all multivariate boundary cases.
• An indefinite matrix can satisfy the restricted negative-quadratic condition. A positive-definite matrix cannot; the paper’s speculation to that effect is not adopted. Its condition includes coordinate axes and therefore requires negative diagonals in the strict regime.
• The displayed conditional-partition formula on PDF p. 7 fails its independent exponential special case: at eta1=-1 and eta2=0 it gives a partition of 1+sqrt(pi)/2, whereas the integral of exp(-t) on t>=0 is 1. The record retains the defining integral and the explicit valid tail argument rather than that printed closed form.
• Outside the independent special case, node conditionals need not be ordinary exponential distributions and their means are not generally reciprocals of simple linear rates. This model does not by itself define a temporal waiting-time process.
• The graph is finite and Phi is symmetric; an off-diagonal matrix entry contributes twice in the quadratic form. Separately estimated node regressions need not agree on one symmetric joint parameterization.
• The node conditionals usually have two sufficient statistics and differ from the original univariate base family. A base-family name is not a claim about every conditional or marginal distribution, and an edge parameter is not a universal marginal-correlation or causal-effect statement.
• The printed radial formulas omit the Jacobian needed to interpret x=z*v as a change of continuous coordinates. In p dimensions the L1-radius transformation contributes z^(p-1); a bare ray kernel is not generally the radius distribution. Discrete directions require a counting-measure treatment. The normalization arguments stated here use direct joint domination instead.
• The paper proposes nodewise l1-penalized fitting, numerical conditional partitions and annealed importance sampling. Model existence does not establish finite-data graph recovery, sampler mixing, numerical accuracy or optimizer convergence; no author code, empirical fitting or benchmark was reproduced.
• Reported synthetic and airport-delay experiments are contextual. The paper’s synthetic description prints a positive diagonal while calling the matrix negative definite, an inconsistent sign convention that must be resolved before reproducing that generator. Airport associations and fitted likelihoods do not establish causal effects or held-out predictive performance.