Fractional Programming Transformations for Wireless Resource Allocation

79 concepts72 papers2 proposals

In modern wireless network design, engineers rarely optimize for a single raw metric like maximum throughput or minimum power consumption. Instead, the most critical performance indicators are balanced ratios.

For instance, maximizing Energy Efficiency (EE) means maximizing the ratio of data rate to power consumption (bits per Joule). Maximizing Data Processing Efficiency (DPE) balances computational throughput against latency and energy overhead. In status-updating internet-of-things (IoT) networks, minimizing the Age of Information (AoI) requires minimizing the ratio of average data age to average transmission time.

Mathematically, these objectives take the form of fractional functions:

$$\max_{x \in \mathcal{X}} \frac{f(x)}{g(x)}$$

Even when the numerator $f(x)$ is concave and the denominator $g(x)$ is convex, their ratio is fundamentally non-convex. Standard gradient descent methods easily get trapped in local optima or fail entirely due to the non-monotonic nature of the fractional surface.

Fractional programming (FP) transformations solve this by converting intractable ratio objectives into equivalent subtractive or parametric forms. This decoupling allows engineers to solve complex ratio problems as a sequence of highly tractable, standard convex subproblems.

The Core Mechanism: From Ratios to Subtraction

To understand why fractional programming works, consider the optimization of a single ratio where $f(x) \ge 0$ and $g(x) > 0$ over a feasible set $\mathcal{X}$. Let $q^*$ denote the maximum achievable value of this ratio:

$$q^* = \max_{x \in \mathcal{X}} \frac{f(x)}{g(x)}$$

By definition, for any feasible choice of $x$, the ratio cannot exceed $q^*$:

$$\frac{f(x)}{g(x)} \le q^* \implies f(x) - q^* g(x) \le 0$$

Crucially, the equality $f(x) - q^* g(x) = 0$ holds if and only if $x$ is the exact global optimizer $x^*$. This simple algebraic rearrangement yields a profound optimization insight: maximizing a ratio is perfectly equivalent to finding the root of an auxiliary subtractive function.

As reviewed by Ya-Feng Liu et al. in their comprehensive survey of resource allocation methods, Optimization problems and methods for resource allocation in wireless communications, this linearization bypasses the fractional non-convexity. If $f(x)$ is concave and $g(x)$ is convex, the transformed objective $f(x) - q g(x)$ is strictly concave for any non-negative parameter $q$.

Why It Works: The Monotonicity Loop

To find the optimal ratio $q^*$, we define a parametric function representing the maximum value of the subtractive problem for a given guess $q$:

$$F(q) = \max_{x \in \mathcal{X}} \{ f(x) - q g(x) \}$$

The function $F(q)$ possesses a beautiful geometric property: it is strictly decreasing and continuous in $q$.

  • If our guess is too low ($q < q^*$), then $F(q) > 0$.
  • If our guess is too high ($q > q^*$), then $F(q) < 0$.
  • If our guess is exactly right ($q = q^*$), then $F(q) = 0$.
  F(q)
   ^
   |  \
   |   \
   |    \
---+-----\--------> q
   |      \  q*
   |       \

This strict monotonicity guarantees that finding the root $F(q) = 0$ is incredibly stable. Dinkelbach’s method exploits this by iteratively updating the parameter $q$. In each iteration $k$, the algorithm solves the concave subtractive subproblem to find $x_k$, and then updates the next guess as the actual ratio achieved by that point:

$$q_{k+1} = \frac{f(x_k)}{g(x_k)}$$

This simple loop exhibits superlinear (often quadratic) convergence, transforming an otherwise intractable fractional problem into a fast, predictable routine.

The authors of Coordinated Beamforming for Energy Efficient Transmission in Multicell Multiuser Systems leveraged this exact mechanism to optimize multi-cell cellular networks, converting non-convex energy efficiency ratios into linear or convex subproblems. Similarly, Mushu Li et al. applied Dinkelbach's method to joint trajectory design and computation scheduling in Energy-Efficient UAV-Assisted Mobile Edge Computing: Resource Allocation and Trajectory Optimization, maximizing the "bits-per-Joule" efficiency of drones moving through dynamic environments.

The framework is flexible enough to handle expectations in stochastic systems as well. For instance, Ahmed Arafa et al. utilized a subtractive auxiliary transformation in Online Timely Status Updates with Erasures for Energy Harvesting Sensors to minimize the ratio of expected age over expected time, deriving optimal threshold-based transmission policies for energy-harvesting sensors.

Handling Multi-User Interference and Sum-of-Ratios

While a single ratio is easily managed by Dinkelbach's method, practical wireless networks often feature multiple independent users sharing the same spectrum. This introduces two major complications: multi-user interference (making the numerator non-concave) and sum-of-ratios objectives:

$$\max_{x} \sum_{i} \frac{f_i(x)}{g_i(x)}$$

A single parameter $q$ cannot decouple a sum of multiple fractions. To resolve this, researchers introduce vector-valued auxiliary variables or parametric weights to decouple each individual ratio.

For example, Jianqing Liu et al. used a multi-parameter subtractive transformation in Optimizing IoT Energy Efficiency on Edge (EEE): a Cross-layer Design in a Cognitive Mesh Network to handle a non-convex weighted sum-of-ratios objective across cognitive radio routers. In the context of the decentralized web, Jun Zhao et al. applied an equivalent subtractive transformation in Data Processing Efficiency Aware User Association and Resource Allocation in Blockchain Enabled Metaverse over Wireless Communications to simplify multi-user sum-of-ratios problems governing NFT transaction latency and energy consumption.

When multi-user interference makes the data-rate expressions non-concave, fractional programming is frequently paired with Successive Convex Approximation (SCA) or Difference-of-Convex (DC) programming. The core idea is to approximate the non-concave numerator by its concave lower bound at a local point, apply the fractional transformation, solve the subproblem, and update the local bound.

Ming Zeng et al. demonstrated this hybrid approach in Energy-Efficient Joint User-RB Association and Power Allocation for Uplink Hybrid NOMA-OMA, combining Dinkelbach's transform with graph-based matching to solve pseudo-concave resource allocation problems. The authors extended this paradigm to handle complex multi-user interference terms in millimeter-wave systems in Energy-Efficient Power Allocation in Uplink mmWave Massive MIMO with NOMA, using DC programming to linearize the interference terms inside the Dinkelbach loop.

Alternative Transformations: Charnes-Cooper and Quadratic Forms

Dinkelbach's method is inherently iterative, requiring the system to solve an optimization problem at every step. However, certain fractional problems can be solved in a single shot using direct variable transformations.

The Charnes-Cooper transformation is a classic example. Instead of iteratively guessing the root, it maps a fractional program directly into a concave problem by changing variables. By defining:

$$y = \frac{x}{g(x)} \quad \text{and} \quad t = \frac{1}{g(x)}$$

the ratio objective is linearized into $t \cdot f(y/t)$, subject to the convex constraint $t \cdot g(y/t) = 1$. Panayotis Mertikopoulos et al. leveraged this exact mapping in Learning to be green: robust energy efficiency maximization in dynamic MIMO-OFDM systems, allowing a dynamic MIMO system to compute its optimal energy-efficient transmission policy "on the fly" via robust online learning without needing to predict future channel states.

For highly structured or non-smooth fractional objectives, researchers have developed advanced quadratic transforms. Ganzhao Yuan introduced FADMM in ADMM for Structured Fractional Minimization, a framework that integrates the Alternating Direction Method of Multipliers (ADMM) with parametric and quadratic transforms. By converting ratio-based objectives into quadratic forms, the complex global structure decomposes into small, parallelizable subproblems, achieving state-of-the-art convergence rates for structured fractional minimization.

The Ultimate Shortcut: The Lambert W Function

In specific single-user or decoupled scenarios, the subtractive subproblem resulting from a fractional transformation can be solved analytically, bypassing numerical optimization entirely.

When the numerator is a logarithmic rate function (e.g., Shannon capacity $\log(1 + \gamma p)$) and the denominator is a linear power function ($p + P_c$), the first-order optimality condition of the subtractive form $f(x) - q g(x) = 0$ reduces to a transcendental equation of the form:

$$x e^x = y$$

The exact analytical solution to this equation is given by the Lambert W function.

Lokman Sboui demonstrated this elegant shortcut in A New Relation Between Energy Efficiency and Spectral Efficiency in Wireless Communications Systems, deriving a closed-form expression for optimal power allocation that maps energy efficiency directly to spectral efficiency without any iterative numerical overhead.

This closed-form tractability is incredibly valuable for real-time systems. In No-Pain No-Gain: DRL Assisted Optimization in Energy-Constrained CR-NOMA Networks, Zhiguo Ding et al. used the Lambert W function to instantaneously calculate optimal power and time-sharing ratios within a deep reinforcement learning loop. Similarly, Chang Liu et al. utilized the Lambert W function in Optimization for the Metaverse over Mobile Edge Computing with Play to Earn to compute optimal uplink transmit power in closed form, minimizing communication delays for edge-computed Metaverse applications without expensive iterative searches.

Key Foundational Work

Coordinated Beamforming for Energy Efficient Transmission in Multicell Multiuser Systems (2013) — contributed 1 concept elements to this direction.
Max-Min Energy Efficient Beamforming for Multicell Multiuser Joint Transmission Systems (2013) — contributed 1 concept elements to this direction.
History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 73 papers
  2. Version 1 · 2026-10-04 · 87 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_fractional_programming_transformation_fr_30c99704,
  title = {fractional programming transformation (fractional programming)},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/fractional-programming-transformation-fractional-programming/rdr_30c99704629f4e98b9bc8aeef84851f4},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 2, frozen text):

@misc{lacuna_fractional_programming_transformation_fr_30c99704_v2,
  title = {fractional programming transformation (fractional programming)},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/fractional-programming-transformation-fractional-programming/rdr_30c99704629f4e98b9bc8aeef84851f4/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake