Log-APX-hardness: Definition & Implications

• Log-APX-hardness is a complexity notion defining NP optimization problems that resist sub-logarithmic approximations unless major complexity collapses occur.
• L-reductions and gap amplification techniques translate constant PCP gaps into logarithmic separations, underpinning the formal proof framework.
• Canonical problems such as minimum set cover and dominating set illustrate these hardness bounds, setting practical limits for polynomial-time approximation algorithms.

Log-APX-hardness is a central notion in the classification of the approximability of NP optimization problems, capturing a region of computationally hard search problems that are provably inapproximable below logarithmic-factor ratios in polynomial time—unless one admits a collapse in fundamental complexity classes. This concept provides both a threshold for the effectiveness of approximation algorithms and a framework for inapproximability reductions via approximation-preserving transformations. Log-APX-hard problems appear in combinatorial optimization, graph algorithms, and sequence assembly, and their study is intertwined with the development of the PCP theorem and hardness amplification techniques.

1. Formal Definition and Hierarchical Placement

For a minimization problem $P$ in the class of NP-optimization problems (NPO), log-APX is rigorously defined using approximation ratios and algorithmic guarantees. A problem $P$ is in log-APX if there exists a polynomial-time algorithm and a constant $c > 0$ such that for every instance $I$ of size $n$,

$$\frac{\text{ALG}(I)}{\text{OPT}(I)} \leq c \cdot \log n,$$

where $\text{OPT}(I)$ denotes the optimal cost and $\text{ALG}(I)$ the cost attained by the algorithm. Log-APX-hardness is the property that, unless P=NP or a similar widely believed collapse in complexity-theoretic assumptions occurs, there does not exist any polynomial-time algorithm achieving an approximation ratio of $o(\log n)$ for the given problem (Lee et al., 2021, Jha et al., 2019).

The placement of log-APX in the approximability landscape is as follows, with strict inclusions assuming P$\neq$NP:

FPTAS	PTAS	APX	log-APX	poly-APX
$1+\varepsilon$	Arbitrarily close	$O(1)$	$O(\log n)$	$n^{O(1)}$

2. Hardness via L-Reductions and Gap Amplification

Proving log-APX-hardness fundamentally relies on L-reductions, a reduction framework tailored to preserve approximation ratios. An L-reduction from minimization problem $P_1$ to $P_2$ requires:

• Polynomial-time constructible functions $f, g$ mapping instances and solutions between $P_1$ and $P_2$,
• Constants $a, b > 0$ such that for any $x$ and $y$,

$$\text{OPT}_2(f(x)) \leq a \cdot \text{OPT}_1(x), \quad |\text{VAL}_1(g(y)) - \text{OPT}_1(x)| \leq b \cdot |\text{VAL}_2(y) - \text{OPT}_2(f(x))|.$$

If $P_2$ admits an $O(\log n)$-approximation, so does $P_1$, and inapproximability results for $P_1$ transfer to $P_2$ up to constant factors (Lee et al., 2021, Yu, 2016).

Gap amplification, often using repeated PCP-based reductions, is employed to generate logarithmic separation between YES- and NO-instances. For example, successive amplification steps can derive the $O(\log n)$ gap for the minimum set cover problem, converting a constant-factor PCP gap into a logarithmic separation via iterative reduction (Lee et al., 2021).

3. Canonical Log-APX-Hard Problems

Several problems are known to be log-APX-hard due to the above machinery:

• Minimum Set Cover: Classical log-APX-complete. Unless P=NP, no $(1-\varepsilon)\ln n$-approximation exists, and greedy/LP algorithms match this guarantee (Lee et al., 2021).
• Minimum Dominating Set: Also log-APX-hard via L-reductions and similar gap-amplification arguments (Jha et al., 2019).
• Metric Dimension: On maximum degree-3 graphs, provably log-APX-hard via polynomial-time reductions from bipartite dominating set. No $o(\log n)$-approximation exists unless P=NP (Hartung et al., 2012).
• Shortest Common Superstring with Negative Strings (SCSN): Proven log-APX-hard and log-APX-complete via L-reduction from set cover; no $o(\log n)$-approximation unless P=NP (Yu, 2016).
• Minimum Neighborhood Total Dominating Set: No algorithm achieves $(1-\varepsilon)\ln n$-approximation unless NP $\subseteq$ DTIME$(n^{O(\log\log n)})$. The reduction ensures the log-APX-hardness threshold via set cover/dominating set gap preservation (Jha et al., 2019).

4. Tightness of Logarithmic Thresholds and Algorithmic Barriers

The log-APX threshold is both upper and lower bound for many canonical problems. For minimum set cover, the harmonic-number lemma and Chvátal's greedy algorithm yield an $H_n \leq \ln n + 1$ factor, matching the hardness resulting from PCP-based gap reductions (Lee et al., 2021).

Attempts to overcome this barrier via other algorithmic paradigms—primal–dual approaches, local search, semidefinite programming—have not succeeded in breaking the $\Theta(\ln n)$ gap for these problems. The LP relaxation for set cover has an integrality gap of $H_n$. Strong SDP hierarchies also fail to improve this bound, and even access to an APX-oracle does not yield sub-logarithmic approximations without collapsing the polynomial hierarchy (Lee et al., 2021).

5. Extensions and Variants: Sublogarithmic and Polylogarithmic Hardness

Log-APX-hardness captures the setting where polynomial-time algorithms are confined to logarithmic-factor approximations. For certain problems, such as node-disjoint and edge-disjoint paths in planar graphs of bounded degree, the hardness threshold can be even higher: $2^{\Omega(\sqrt{\log n})}$, which excludes the existence of any polylogarithmic-approximation algorithm under standard assumptions (unless NP$\subseteq$DTIME$(n^{O(\log n)})$) (Chuzhoy et al., 2016). These instances demonstrate exponential separation and place such problems "firmly in log-APX-hard," and beyond, into a domain where only quasi-polynomial or worse approximations are feasible.

A summary of the main log-APX-hard and related separation thresholds is given below:

Problem Class	Approximation Threshold	Hardness Basis
Minimum Set Cover	$\Theta(\ln n)$	PCP + gap amplification
Min Dominating Set, Min-NTDS	$\Theta(\ln n)$	L-reduction from set cover
Metric Dimension on degree-3 graphs	$\Theta(\ln n)$	Reduction from bipartite dom. set
SCS with Negative Strings	$\Theta(\log n)$	L-reduction from set cover
NDP/EDP on special planar graphs	$2^{\Omega(\sqrt{\log n})}$	PCP + multi-level gap amplification

6. Significance in Complexity Theory and Open Problems

The class log-APX and its hardness threshold demarcate the boundary where polynomial-time algorithms yield approximations better than polynomial factors yet cannot achieve constant-ratio guarantees. The landscape is shaped largely by the power of PCP theorems and their gap-amplification consequences, and the L-reduction technique provides a conduit for transferring these results broadly among combinatorial optimization problems.

Current research in hardness of approximation explores tightening thresholds, developing finer reductions, and examining whether canonical problems (such as set cover) might admit uniform improvements under refined assumptions. Distinguishing the combinatorial features inherent to log-APX-hard problems and uncovering approaches to circumvent logarithmic barriers in practice remain open directions (Lee et al., 2021).

7. Representative Algorithms and Matching Bounds

For each log-APX-hard problem, natural greedy or LP-based algorithms exist that achieve $O(\log n)$ approximation. For example, the greedy set cover algorithm selects at each step the set covering the largest number of uncovered elements, ensuring an $H_n$-factor guarantee. For Min-NTDS, a greedy-style algorithm attains $2(\ln(\Delta+1) + 1)$ ratio, with $\Delta\leq n-1$, and hence also $O(\log n)$ (Jha et al., 2019). Algorithmic improvements beyond these factors are precluded by log-APX-hardness, as any such progress would contradict standard complexity assumptions.

---

In summary, log-APX-hardness crystallizes a fundamental bottleneck in polynomial-time approximability, exemplified by set cover, dominating set, metric dimension, and several other problems. It combines sophisticated reductions, gap-amplification, and the boundaries set by the PCP theorem to delineate the landscape of hard-to-approximate problems, and serves as a reference point for classifying future developments in approximation algorithms and complexity theory (Lee et al., 2021, Jha et al., 2019, Hartung et al., 2012, Yu, 2016, Chuzhoy et al., 2016).