Total TV Error Bound (Conditional)

Computes the combined conditional TV bound using the triangle inequality: $$d_{TV}(\mathcal{L}(S_J | \alpha), \text{Poisson}(\alpha c_J)) \le B_{\text{Pois}} + B_{\text{lin}}$$

Usage

compute_total_tv_bound(J, alpha, cJ = log(J))

Arguments

J

Integer; sample size.

alpha

Numeric; concentration parameter (vectorized).

cJ

Numeric; scaling constant (default: log(J)).

Value

Numeric; upper bound on total TV error (capped at 1).

Details

The result is capped at 1 since TV distance is bounded by 1.

In the submitted manuscript this is Appendix D, Corollary cor:combined-conditional (Equations D14–D16). The total conditional TV error decomposes as:

1. Poissonization error: \(S_J | \alpha\) vs \(\text{Poisson}(\lambda_J(\alpha))\)

2. Linearization error: \(\text{Poisson}(\lambda_J(\alpha))\) vs \(\text{Poisson}(\alpha c_J)\)

See also

compute_poissonization_bound, compute_linearization_bound, expected_tv_bound

Examples

if (FALSE) { # \dontrun{
# Total bound at alpha = E[alpha] under Gamma(2, 1)
compute_total_tv_bound(J = 50, alpha = 2)

# Vectorized
compute_total_tv_bound(J = 50, alpha = c(0.5, 1, 2, 5))

} # }