Noiseless Non-Adaptive Group Testing

• Noiseless non-adaptive group testing is a method for identifying a small set of defectives from a large population using fixed pooled tests that return a Boolean OR outcome.
• It leverages random and structured pooling designs, such as Bernoulli and constant-column methods, to achieve order-optimal test counts and efficient recovery algorithms.
• The approach is underpinned by rigorous information-theoretic limits and sharp phase transitions, ensuring high-probability and uniform recovery in practical scenarios.

Noiseless non-adaptive group testing is the problem of identifying a small subset of defective items within a large population using tests that pool arbitrary groups, where each test returns a noiseless Boolean OR of the defectivity indicators, and all test designs are fixed in advance. Recovery is required with high probability (for-each/probabilistic guarantee) or uniformly (combinatorial/disjunct-matrix) over the choice of the defective set. The foundational results establish both information-theoretic limits and order-optimal, computationally efficient algorithms. This area now features refined results both for standard settings and under practical combinatorial constraints.

1. Problem Formulation and Limits

Given $n$ items with exactly $k$ defectives ($x\in\{0,1\}^n$), a noiseless non-adaptive Boolean OR test outcome is modeled as $y = M \lor x$ for a test matrix $M \in \{0,1\}^{T \times n}$. The goal is to design $M$ (subject to possible combinatorial constraints), minimize $T$ (number of tests), and provide a decoding algorithm such that the recovered set $\hat{S}$ equals the true defective set $S$ with probability $1 - o(1)$ as $n\to\infty$.

The information-theoretic lower bound is $T \ge (1-o(1))k\log_2(n/k)$ for $k = o(n)$, derived from a basic entropy argument via Fano's inequality (Chan et al., 2011, Cheng et al., 2021). For more general sparsity regimes, the phase transition is characterized by $$T^*(n, k) = \max\{k\log_2(n/k),\, k\log_2 k/\ln 2\}$$ tests, and any scheme with $T < (1-o(1)) T^*(n, k)$ cannot succeed with $P[\hat{S} = S] \to 1$ (Bay et al., 2020). In dense settings where $k \gg n/\log n$, individual testing ($T = n$) is rate-optimal (Bay et al., 2020).

2. Test Designs and Pooling Schemes

Random designs dominate the regime of non-adaptive noiseless group testing (Chan et al., 2011, Chan et al., 2012):

• Bernoulli (CBP/COMP) design: Each $M_{t, i}\sim\mathrm{Bernoulli}(p)$, typically with $p=1/k$. Each test contains $\sim n/k$ items.
• Constant (or near-constant)-column design: Each item appears in a fixed number $L$ of tests, sampled (uniformly, with or without replacement). This improves concentration and leads to sharper performance (Bay et al., 2020).
• Sparse pooling graph ensembles: Bipartite $(\ell,r)$-regular designs enforce fixed item-degree $\ell$ and test-degree $r$ (Wadayama, 2013, Yacoub et al., 27 Jul 2025, Vem et al., 2017). These are widely used both for test efficiency and algorithmic analysis, including left-and-right-regular ensembles for optimal peeling-type decoding (Vem et al., 2017).

Refined constructions, such as multi-level and recursively split pooling (e.g., binary-splitting with hashing (Price et al., 2020)), combine algebraic and combinatorial techniques to achieve both test-optimality and runtime optimality.

3. Decoding Algorithms

Canonical decoders for noiseless non-adaptive group testing are as follows:

Decoder	Output Error Types	Complexity and Notes
COMP	0 false negatives	$O(Tn)$, possible false positives
DD	0 false positives	$O(Tn)$, possible false negatives
SCOMP	Balanced errors	$O(Tn)$, sequential refinement
LP Decoding	Both zero (w.h.p.)	$O(Tn + \mathrm{LP})$
Peeling (SAFFRON)	Both zero (w.h.p.)	$O(K\log(N/K))$ on sparse graphs

• COMP (Combinatorial Orthogonal Matching Pursuit) marks as non-defective any item appearing in a negative test; the rest are declared defective (Chan et al., 2011, Chan et al., 2012). It achieves $T \geq e k \ln n$ tests for error $n^{-\delta}$ (Chan et al., 2011).
• DD (Definite Defectives) identifies items that are guaranteed to be defective by "singleton" positive tests after COMP preprocessing (Bay et al., 2020, Gebhard et al., 2020). For all $(n, k)$ in the sparse regime, it achieves order-optimal $T = C_{n,k} k\log n$ up to sharp constants (Bay et al., 2020).
• SCOMP (Sequential COMP) and W-SCOMP (Weighted SCOMP): SCOMP greedily adds items explaining unexplained positive tests for balanced error; W-SCOMP reweighs candidate scores according to test ambiguity for better SNR, reducing test counts (up to 10%) with negligible decoding cost increase (Franco-Vivo, 12 Jan 2026).
• LP decoding: Relaxes the indicator constraints to $[0,1]^n$ and solves a set covering LP; achieves $T = O(e^2)\,k \log n$ (Chan et al., 2012).
• Peeling (sparse-graph coding / regular-SAFFRON): Iterative decoding proceeds via "binning" and recursively removes discovered singletons/doubletons, with optimal $O(K\log(N/K))$ scaling (Vem et al., 2017).

Modern high-rate approaches use hashing or binary-splitting with randomization for test and memory optimality with logarithmic runtime (Price et al., 2020).

4. Performance Guarantees and Sharp Thresholds

Order-optimal recovery guarantees are available for nearly all practical decoders:

• With a Bernoulli or constant-weight design, COMP/CBP achieves all-$k$ recovery for $T \geq e k \ln n$ (Chan et al., 2011). LP and DD improve constants further and, for sparse regimes ($k = o(n)$), $T \sim k \log n$ tests guarantee $P[\hat S = S] \to 1$ (Bay et al., 2020, Chan et al., 2012).
• Exact sharp constants $C_{k, n}$ in $T = C_{k, n} k \log n$ are formally identified for all sparsity regimes; for $k \gg n/\log n$, $T = n$ is necessary and sufficient (Bay et al., 2020).
• For partial/approximate recovery (fraction $(1-\epsilon)$ of $k$), sparse-graph code constructions using left-and-right-regular ensembles achieve $m = c_\epsilon K \log(N/K)$ tests, and for exact recovery, $m = c_2 K \log K \log(N/K)$ (Vem et al., 2017).
• The binary-splitting approach attains $O(k \log n)$ tests and $O(k \log n)$ time, improving previous $O(k^2 \log k \log n)$ runtime bounds while maintaining test optimality (Price et al., 2020).
• With degree-constrained or pool-size-constrained designs, SCOMP and DD equipped with regular pooling graphs are shown to be information-theoretically optimal up to constant factors. Explicit phase transitions are identified for feasible $(l, r, p)$ (Gebhard et al., 2020, Vem et al., 2017, Yacoub et al., 27 Jul 2025, Wadayama, 2013).

5. Graph-based Designs and Ensemble Analysis

Sparse pooling graphs provide an analytic bridge between design and error rates. Ensemble-based generating function techniques enumerate false alarm and misdetection probabilities of COMP and DD for general prescribed $(l,r)$ degree distributions, yielding precise phase-transition phenomena (Yacoub et al., 27 Jul 2025, Wadayama, 2013):

• For a fixed test-to-item ratio $\xi = m/n$ and prevalence $\delta = d/n$, the average false alarm rate (under COMP) and misdetection rate (under DD) display sharp threshold behavior as functions of $(\xi, \delta)$, separating regions of vanishing error from persistent error.
• Edge-type enumeration, via tailored generating functions, allows exact evaluation of ensemble error rates and guides the practical design of test matrices for prescribed performance.

6. Algorithmic and Information-Theoretic Insights

• Randomization is essential for "for-each" recovery with high probability; deterministic constructions with the same efficiency remain an open problem except for combinatorial/disjunct-matrix lower bounds, which require $\Theta(k^2 \log n)$ tests and are computationally intractable for large $n$ (Vem et al., 2017, Wadayama, 2013).
• Binary splitting and dyadic partitioning translate adaptivity into non-adaptation via randomized hashing or recursive grouping, efficiently eliminating non-defective items while maintaining a tractable memory and decoding burden (Price et al., 2020).
• All known practical designs approach the information-theoretic lower bound within a universal constant, with the best possible constant-factor separation for random Bernoulli pooling and the COMP algorithm being $e$ (Chan et al., 2011, Chan et al., 2012).

7. Extensions, Trade-offs, and Current Research Directions

• Extensions to noise-robust non-adaptive group testing are natural but typically require more tests and different decoder design; current optimality guarantees are strictly for the noiseless model (Chan et al., 2011, Chan et al., 2012).
• Practical constraints such as bounded item degree ($\Delta$) or fixed pool size ($\Gamma$) fundamentally alter recovery thresholds and induce phase transitions and adaptivity gaps (Gebhard et al., 2020).
• Advanced post-optimal algorithms, including trellis-based exact MAP coordinate detection, enable systematic exploration of the full ROC trade-off surface, but have exponential complexity in the number of positive tests and are thus used on moderate-size instances or as subroutines (Liva et al., 2021).
• Weighted decoders such as W-SCOMP achieve strictly improved test efficiency (5–10% savings empirically) at negligible additional computational burden, confirming the value of information-theoretic SNR analysis in the design of group testing decoders (Franco-Vivo, 12 Jan 2026).

The current frontier for noiseless non-adaptive group testing research encompasses further reductions to optimal constant factors, derandomization, robust design under noise or adversarial error, and joint test-design/decoding co-optimization all while maintaining phase transition sharpness and computational tractability (Price et al., 2020, Bay et al., 2020, Franco-Vivo, 12 Jan 2026, Gebhard et al., 2020).