Paper: arXiv 2305.09046

Authors: James Chok, Geoffrey M. Vasil

Abstract

We propose a new iteration scheme, the Cauchy-Simplex, to optimize convex problems over the probability simplex \({“w\in\mathbb{R”}^n\ |\ \sum_i w_i=1\ \textrm{“and”}\ w_i\geq0}\). Specifically, we map the simplex to the positive quadrant of a unit sphere, envisage gradient descent in latent variables, and map the result back in a way that only depends on the simplex variable. Moreover, proving rigorous convergence results in this formulation leads inherently to tools from information theory (e.g., cross-entropy and KL divergence). Each iteration of the Cauchy-Simplex consists of simple operations, making it well-suited for high-dimensional problems. In continuous time, we prove that \(f(x_T)-f(x^) = {“O”}(1/T)\) for differentiable real-valued convex functions, where \(T\) is the number of time steps and \(w^\) is the optimal solution. Numerical experiments of projection onto convex hulls show faster convergence than similar algorithms. Finally, we apply our algorithm to online learning problems and prove the convergence of the average regret for (1) Prediction with expert advice and (2) Universal Portfolios.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 5.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents a novel optimization algorithm with heavy mathematical proofs, including convergence rates and connections to information theory, indicating high mathematical complexity. While it includes numerical experiments and applications to online learning, the evidence is primarily theoretical and experimental, lacking the extensive backtesting and data processing typical of high empirical rigor studies.

Research Flowchart

  flowchart TD
  A["Research Goal: Optimize convex functions<br>over the probability simplex"] --> B["Methodology: Cauchy-Simplex Scheme"]
  
  B --> C["Mapping:<br>Simplex → Positive unit sphere quadrant"]
  C --> D["Computation:<br>Gradient descent in latent variables"]
  D --> E["Mapping:<br>Back to probability simplex"]
  
  E --> F{"Computational Process"}
  F --> G["Iterative optimization<br>constrained to simplex"]
  F --> H["Continuous time analysis<br>O(1/T) convergence rate"]
  
  G --> I["Key Findings"]
  H --> I
  
  I --> J["Theoretical: Convergence proof using<br>information theory metrics"]
  I --> K["Experimental: Faster convergence<br>in convex hull projection"]
  I --> L["Application: Online learning<br>Universal Portfolios & Expert Advice"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a new iterative algorithm called the Cauchy‑Simplex to optimize convex problems over the probability simplex.

Source sentence
We propose a new iteration scheme, the Cauchy‑Simplex, to optimize convex problems over the probability simplex {w∈ℝⁿ | ∑ᵢ

Assumes 4

  • The market variability parameter is known for each dataset. market
    Source sentence
    √(2 log(N)/T), respectively, where a is the market variability parameter. We assume that the market variability parameter is given for each dataset.
  • Transaction costs are not included in the performance analysis. execution
    Source sentence
    this experiment does not reflect real‑world performance, as the market variability parameter is assumed to be known, transaction costs are not factored into our analysis
  • The no‑junk‑bond assumption is made, which tends to overestimate performance. model
    Source sentence
    the no‑junk‑bond assumption tends to overestimate performance
  • The objective functions are convex and differentiable. model
    Source sentence
    In continuous time, we prove that f(x_T)‑f(x^*) = O(1/T) for differentiable real‑valued convex functions

Data

  • TSE market-prices public
    Jan 4, 1994 to Dec 31, 1998 · 88 stocks traded on the Toronto Stock Exchange
    Source sentence
    TSE is a collection of 88 stocks traded on the Toronto Stock Exchange from Jan 4, 1994, to Dec 31, 1998.
  • NYSE market-prices public
    Source sentence
    For the NYSE, DJIA, and SP500 datasets, CS slightly outperforms EGD in both the APY and Sharpe ratio
  • DJIA market-prices public
    Source sentence
    For the NYSE, DJIA, and SP500 datasets, CS slightly outperforms EGD in both the APY and Sharpe ratio
  • SP500 market-prices public
    Source sentence
    For the NYSE, DJIA, and SP500 datasets, CS slightly outperforms EGD in both the APY and Sharpe ratio

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
transaction costs are not factored into our analysis

Finds 3

  • The Cauchy‑Simplex algorithm achieves a convergence rate of O(1/T) for differentiable convex functions. theoretical
    Source sentence
    In continuous time, we prove that f(x_T)‑f(x^*) = O(1/T) for differentiable real‑valued convex functions
  • On NYSE, DJIA, and SP500 datasets, Cauchy‑Simplex slightly outperforms the EGD scheme in both APY and Sharpe ratio. performance
    Source sentence
    For the NYSE, DJIA, and SP500 datasets, CS slightly outperforms EGD in both the APY and Sharpe ratio
  • On the TSE dataset, the Buy‑and‑Hold strategy outperforms both Cauchy‑Simplex and EGD. performance
    Source sentence
    the B&H strategy outperforms both CS and EGD on the TSE

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

https://github.com/infamoussoap/ConvexHull — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.