Toward a salmon conjecture

Basic Information

• Paper ID: 1009.6181
• Title: Toward a salmon conjecture
• Authors: Daniel J. Bates, Luke Oeding
• Classification: math.AG (Algebraic Geometry)
• Publication Date: February 2, 2011 (arXiv version)
• Paper Link: https://arxiv.org/abs/1009.6181

Abstract

Using computational results from the numerical algebraic geometry software package Bertini, the authors prove (at high numerical precision) that a specific set of degree 6 and 9 polynomials cut out the secant variety $\sigma_4(\mathbb{P}^2 \times \mathbb{P}^2 \times \mathbb{P}^3)$. Combined with arguments provided by Landsberg and Manivel (with proofs corrected by Friedland), this implies that for a broader range of secant varieties (including $\sigma_4(\mathbb{P}^3 \times \mathbb{P}^3 \times \mathbb{P}^3)$), there exist set-theoretic defining equations of degrees 5, 6, and 9. This has special significance for the "salmon conjecture" regarding ideal-theoretic defining equations proposed by E. Allman.

Research Background and Motivation

Core Problem

The central problem addressed in this paper is determining the defining ideal of the secant variety $\sigma_4(\mathbb{P}^3 \times \mathbb{P}^3 \times \mathbb{P}^3)$, which is the "salmon conjecture" problem posed by E. Allman in 2007.

Significance

1. Theoretical Importance: Secant varieties are classical objects of study in algebraic geometry, and determining their defining equations is crucial for understanding the border rank of tensors
2. Applied Value: In phylogenetics, this secant variety is associated with mixture models of evolutionary independence
3. Universality: Allman and Rhodes proved that finding polynomial invariants for this small evolutionary tree will provide all polynomial invariants for any binary evolutionary tree's statistical model

Limitations of Existing Methods

• Friedland recently proved that sets of degree 5, 9, and 16 polynomials can set-theoretically define $\sigma_4(\mathbb{P}^3 \times \mathbb{P}^3 \times \mathbb{P}^3)$, but using higher degree equations
• The ideal-theoretic version of the problem remains unsolved
• Lack of effective membership testing methods

Core Contributions

1. Provided lower-degree set-theoretic defining equations: Proved that $\sigma_4(\mathbb{P}^3 \times \mathbb{P}^3 \times \mathbb{P}^3)$ can be set-theoretically defined by degree 5, 6, and 9 equations, using lower degrees than Friedland's result
2. Numerically verified key conjecture: Used the Bertini software package to verify Conjecture 3.8, showing that the zero set of degree 6 equations is precisely the union of the target secant variety and a specific subspace variety
3. Constructed concrete polynomial bases: Provided construction methods for concrete polynomial bases of relevant Schur modules
4. Proposed ideal-theoretic version of conjecture: Based on computational results, proposed the "salmon conjecture" suggesting the ideal-theoretic version also holds

Detailed Methodology

Problem Definition

Determine the set-theoretic and ideal-theoretic defining equations of the secant variety $\sigma_4(\mathbb{P}^a \times \mathbb{P}^b \times \mathbb{P}^c)$, particularly for the case $a=b=c=3$.

Core Methodological Framework

1. Representation-Theoretic Approach

Utilizing the symmetry of $GL(A) \times GL(B) \times GL(C)$, decompose the space of homogeneous polynomials as a direct sum of irreducible modules: $S^d(A^* \otimes B^* \otimes C^*) = \bigoplus_{|\pi_1|=|\pi_2|=|\pi_3|=d} (S^{\pi_1}A^* \otimes S^{\pi_2}B^* \otimes S^{\pi_3}C^*)^{\oplus m_{\pi_1,\pi_2,\pi_3}}$

2. Identification of Key Modules

• Degree 6 module: $M_6 = S^{(2,2,2)}A^* \otimes S^{(2,2,2)}B^* \otimes S^{(3,1,1,1)}C^*$
• Degree 5 module: $M_5 = S^{(3,1,1)}A^* \otimes S^{(2,1,1,1)}B^* \otimes S^{(2,1,1,1)}C^* \oplus \cdots$
• Degree 9 module: $M_9 = S^{(3,3,3)}A^* \otimes S^{(3,3,3)}B^* \otimes S^{(3,3,3)}C^*$

3. Inheritance Principle

Apply the Landsberg-Manivel inheritance principle to extend equations from $\sigma_4(\mathbb{P}^2 \times \mathbb{P}^2 \times \mathbb{P}^3)$ to higher-dimensional cases.

Technical Innovations

1. Numerical Algebraic Geometry Methods

First application of the Bertini software package's numerical methods to verify theoretical conjectures in algebraic geometry, demonstrating the power of computational methods in pure mathematics research.

2. Polynomial Construction Algorithm

Provides concrete algorithms for constructing polynomials in Schur modules from Young tableau fillings:

• Construction of highest weight vectors
• Application of Young symmetrizers
• Reordering isomorphism maps
• Symmetrization to obtain homogeneous polynomials

3. Integration of Geometric Arguments and Numerical Verification

Combines classical geometric arguments with modern numerical computation to provide strong numerical evidence for theoretical conjectures.

Experimental Setup

Computational Environment

• Hardware: 8 processors (two 2.66 GHz quad-core Xeon 5410s)
• Software: Bertini numerical algebraic geometry software package
• Computation Time: Approximately 2 weeks
• Precision Control: Tracking and final tolerances set to $10^{-10}$ or smaller

Key Computations

Computation 4.1: Verification that the zero set of $M_6$ has exactly two irreducible components:

• Component of dimension 31 and degree 345 (corresponding to $\sigma_4(\mathbb{P}^2 \times \mathbb{P}^2 \times \mathbb{P}^3)$)
• Component of dimension 29 and degree 84 (corresponding to $\text{Sub}_{3,3,3}$)

Reliability Controls

• Adaptive precision numerical methods employed
• Tracking over 200,000 paths with no path failures or crossings
• Monodromy and trace test procedures error-free
• Multiple verifications and error controls

Experimental Results

Main Results

Theorem 3.10: For $a,b,c \geq 3$, at high numerical precision, the set-theoretic zero set of $\sigma_4(\mathbb{P}^{a-1} \times \mathbb{P}^{b-1} \times \mathbb{P}^{c-1})$ is defined by the following equations:

1. Strassen commutativity conditions ($M_5$): 1728 degree 5 equations
2. Equations inherited from $\sigma_4(\mathbb{P}^2 \times \mathbb{P}^2 \times \mathbb{P}^3)$:
   • $M_6$: 1000 degree 6 equations
   • $M_9$: 8000 degree 9 equations
3. Modules in $S^5(A^* \otimes B^* \otimes C^*)$ containing $\bigwedge^5$

Numerical Verification Results

Computation 4.1 successfully verified Conjecture 3.8, confirming: $V(S^{(2,2,2)}\mathbb{C}^3 \otimes S^{(2,2,2)}\mathbb{C}^3 \otimes S^{(3,1,1,1)}\mathbb{C}^4) = \sigma_4(\mathbb{P}^2 \times \mathbb{P}^2 \times \mathbb{P}^3) \cup \text{Sub}_{3,3,3}$

Precision achieved to 10 decimal places with no computational errors.

Concrete Polynomial Construction

Successfully constructed concrete polynomial bases for equations of various degrees:

• Degree 6: 10 polynomials with 936 or 576 terms respectively
• Degree 5: 3 classes of representative polynomials based on Young tableau fillings
• Degree 9: Basis for 20-dimensional space with polynomials having 9,216 to 43,668 terms

Related Work

Classical Background

• Strassen (1983): First provided degree 9 defining equations for $\sigma_4(\mathbb{P}^2 \times \mathbb{P}^2 \times \mathbb{P}^2)$
• Landsberg-Manivel: Developed representation-theoretic methods and inheritance principles for secant variety equations
• Landsberg-Weyman: Proved normality and rational singularity properties of subspace varieties

Recent Progress

• Friedland (2010): Corrected errors in Landsberg-Manivel proofs, provided set-theoretic definitions using degree 5, 9, and 16 equations
• Allman-Rhodes: Established connections between secant varieties and statistical models in phylogenetics

Unique Contributions of This Paper

First application of numerical methods to solve pure algebraic geometry problems, providing lower-degree equation sets than existing results.

Conclusions and Discussion

Main Conclusions

1. Resolution of set-theoretic problem: Solved the set-theoretic version of Allman's salmon conjecture within numerical precision
2. Optimized equation degrees: Provided defining equations with lower degrees than Friedland's result
3. Practical Value: Provided more efficient methods for model membership testing in phylogenetics

Limitations

1. Constraints of numerical methods: Results depend on numerical computation; while highly precise, they are not rigorous mathematical proofs
2. Ideal-theoretic version unresolved: Remains conjectural, requiring further theoretical work
3. Computational complexity: Large-scale numerical computation requires significant computational resources

Future Directions

1. Ideal-theoretic proofs: Seek direct proofs that degree 6 equations do not lie in the ideal generated by degree 5 equations
2. Algorithm optimization: Improve computational efficiency by exploiting additional symmetries
3. Application extensions: Apply methods to other secant variety problems

In-Depth Evaluation

Strengths

1. Methodological Innovation: Pioneering application of numerical algebraic geometry to pure mathematical theory problems
2. Result Optimization: Achieved lower-degree defining equations compared to existing results
3. Computational Rigor: Employed multiple verifications and high-precision controls
4. Theory Integration: Skillfully combined representation theory, algebraic geometry, and numerical computation

Weaknesses

1. Proof Completeness: Relies on numerical methods, lacking completely rigorous mathematical proof
2. Computational Cost: Requires substantial computational resources, limiting reproducibility
3. Generality: Primarily addresses specific secant varieties with limited generalization

Impact

1. Interdisciplinary Bridging: Demonstrates the importance of computational methods in theoretical mathematics
2. Problem Advancement: Significantly advances the resolution of the salmon conjecture
3. Methodological Precedent: Provides an exemplar for numerical verification in similar problems

Applicable Scenarios

• Research on defining equations of complex algebraic varieties
• Verification of statistical models in phylogenetics
• Computational studies of tensor border rank problems
• Algebraic geometry problems requiring large-scale symbolic computation

References

This paper cites important literature from multiple fields including algebraic geometry, representation theory, numerical algebraic geometry, and phylogenetics, particularly:

• Strassen's classical work on tensor rank
• Landsberg-Manivel's series of studies on secant varieties
• Technical literature on the Bertini software package
• Allman-Rhodes' work on phylogenetic applications