Skip to the content.

Parameter Reference

Validated Parameters for Coopetition-Gym Environments

This document provides a comprehensive reference for all parameters used in Coopetition-Gym, including validated values, acceptable ranges, and calibration guidance.


Quick Reference Card

For practitioners who need parameter values quickly:

# Value Function Parameters (TR-1)
THETA = 20.0          # Logarithmic scale
BETA = 0.75           # Power exponent
GAMMA = 0.65          # Complementarity

# Trust Dynamics Parameters (TR-2)
LAMBDA_PLUS = 0.10    # Trust building rate
LAMBDA_MINUS = 0.30   # Trust erosion rate
MU_R = 0.60           # Reputation damage severity
DELTA_R = 0.03        # Reputation decay rate
XI = 0.50             # Interdependence amplification
KAPPA = 1.0           # Signal sensitivity

# Initial Conditions
T_INIT = 0.50         # Initial trust level
R_INIT = 0.00         # Initial reputation damage

Pillar 1: Interdependence & Complementarity Parameters

Value Function Parameters

Parameter Symbol Validated Value Range Source
Scale Factor θ 20.0 [15, 30] TR-1 §7.3, S-LCD validation
Complementarity γ 0.65 [0.50, 0.80] TR-1 §7.2, multi-criteria

Usage:

\[f_i(a_i) = \theta \cdot \ln(1 + a_i)\] \[V(\mathbf{a} \mid \gamma) = \sum_{i=1}^{N} f_i(a_i) + \gamma \cdot \left(\prod_{i=1}^{N} a_i\right)^{1/N}\]

Validation Performance: 58/60 (96.7%) on Samsung-Sony S-LCD case study

When to Use: Manufacturing joint ventures, technology partnerships, scenarios where initial capabilities are highly valuable but incremental improvements have declining impact.

Power Specification (Alternative)

Parameter Symbol Validated Value Range Source
Exponent β 0.75 [0.65, 0.85] TR-1 §7.1, 22,000+ trials
Complementarity γ 0.65 [0.50, 0.80] TR-1 §7.2

Usage:

\[f_i(a_i) = a_i^{\beta}\] \[V(\mathbf{a} \mid \gamma) = \sum_{i=1}^{N} f_i(a_i) + \gamma \cdot \left(\prod_{i=1}^{N} a_i\right)^{1/N}\]

Validation Performance: 46/60 (76.7%) on Samsung-Sony S-LCD case study

When to Use: General scenarios, platform ecosystems, when cooperation magnitudes may be larger.

Comparison

Criterion Logarithmic (θ=20) Power (β=0.75) Winner
S-LCD Validation 58/60 46/60 Logarithmic
Cooperation Prediction 41% increase 166% increase Logarithmic (realistic)
Bounded Returns Yes No Logarithmic
Mathematical Tractability Moderate High Power

Interdependence Parameters

Parameter Symbol Typical Values Range Description
Dependency Weight $w_d$ Context-specific [0, 1] Goal importance (normalized)
Criticality Factor $\text{crit}$ Calculated [0, 1] 1/n for n alternatives
Bargaining Share $\alpha_i$ 0.50 (symmetric) [0, 1] Must sum to 1.0

Interdependence Matrix Guidance:

Relationship Type $D_{ij}$ Range Example
No dependency 0.00 Competitors in separate markets
Weak dependency 0.10 - 0.30 Multiple alternative suppliers
Moderate dependency 0.30 - 0.60 Preferred but substitutable partner
Strong dependency 0.60 - 0.85 Critical supplier, few alternatives
Complete dependency 0.85 - 1.00 Sole provider of essential resource

Pillar 2: Trust Dynamics Parameters

Trust Evolution Parameters

Parameter Symbol Validated Value Range Source
Trust Building Rate $\lambda^+$ 0.10 [0.05, 0.15] TR-2 §7.2
Trust Erosion Rate $\lambda^-$ 0.30 [0.20, 0.45] TR-2 §7.2
Negativity Ratio $\lambda^-/\lambda^+$ 3.0 [2.5, 4.0] Behavioral economics

The 3:1 Ratio: Empirically grounded in behavioral economics research showing trust erodes approximately 3× faster than it builds. This captures:

Calibration Guidance:

Context $\lambda^+$ $\lambda^-$ Ratio Rationale
High-trust culture 0.12 0.30 2.5 Faster trust building
Standard business 0.10 0.30 3.0 Default validated values
Low-trust environment 0.08 0.35 4.4 Slower building, faster erosion
Post-crisis recovery 0.06 0.25 4.2 Difficult trust rebuilding

Reputation Parameters

Parameter Symbol Validated Value Range Source
Damage Severity $\mu_R$ 0.60 [0.45, 0.75] TR-2 §7.3
Decay Rate $\delta_R$ 0.03 [0.01, 0.05] TR-2 §7.3

Interpretation:

Hysteresis Effect:

\[\Large \Theta = 1 - R \quad \text{(Trust Ceiling)}\]
Stage $T$ $R$ $\Theta$ Status
Initial 0.50 0.00 1.00 Full recovery possible
After violation 0.35 0.40 0.60 Ceiling at 60%
After recovery 0.58 0.40 0.60 Cannot exceed ceiling

Amplification Parameters

Parameter Symbol Validated Value Range Source
Interdep. Amplification $\xi$ 0.50 [0.30, 0.70] TR-2 §7.4
Signal Sensitivity $\kappa$ 1.0 [0.5, 2.0] TR-2 §6.1

Interdependence Amplification Effect:

\[\Large \text{Erosion} = \lambda^- \cdot |s| \cdot T \cdot (1 + \xi \cdot D_{ij})\]
Dependency Level $D_{ij}$ Erosion Factor
Low dependency 0.2 1.10
High dependency 0.8 1.40

Result: 27% faster trust erosion in high-dependency relationships

Initial Conditions

Parameter Symbol Recommended Range Description
Initial Trust $T^0_{ij}$ 0.50 [0.0, 1.0] Starting trust level
Initial Reputation $R^0_{ij}$ 0.00 [0.0, 1.0] Starting reputation damage
Baseline Action $a^{\text{baseline}}$ Context $[0, e]$ Expected cooperation level

Initial Trust Guidance:

Relationship History $T^0$ $R^0$ Scenario
First interaction 0.50 0.00 Neutral starting point
Positive reputation 0.70 0.00 Known reliable partner
Prior relationship 0.60-0.80 0.00 Successful past collaboration
Recovery scenario 0.30 0.40 Post-violation situation
Hostile history 0.20 0.60 Prior conflicts

Environment-Specific Parameters

TrustDilemma-v0

default_params = {
    'n_agents': 2,
    'max_steps': 100,
    'endowment': 100.0,
    'theta': 20.0,
    'gamma': 0.65,
    'lambda_plus': 0.10,
    'lambda_minus': 0.30,
    'mu_R': 0.60,
    'delta_R': 0.03,
    'xi': 0.50,
    'kappa': 1.0,
    'T_init': 0.50,
    'R_init': 0.00,
}

PlatformEcosystem-v0

default_params = {
    'n_agents': 5,  # 1 platform + 4 developers
    'max_steps': 100,
    'endowment': [200.0, 100.0, 100.0, 100.0, 100.0],  # Platform has more
    'theta': 20.0,
    'gamma': 0.70,  # Higher complementarity in ecosystems
    'lambda_plus': 0.10,
    'lambda_minus': 0.30,
    'mu_R': 0.55,   # Slightly lower damage (more forgiving)
    'delta_R': 0.04, # Slightly faster decay
    'xi': 0.40,     # Lower amplification
    'kappa': 1.0,
}

SLCD-v0 (Validated Case Study)

# Parameters calibrated to Samsung-Sony S-LCD (2004-2011)
validated_params = {
    'n_agents': 2,
    'max_steps': 8,  # 8 years
    'endowment': [100.0, 100.0],
    'theta': 20.0,           # Validated
    'gamma': 0.65,           # Validated
    'alpha': [0.50, 0.50],   # Equal bargaining power
    'D_matrix': [[0.0, 0.45],
                 [0.40, 0.0]],  # Moderate mutual dependency
    'lambda_plus': 0.10,
    'lambda_minus': 0.30,
    'T_init': 0.50,
    'R_init': 0.00,
}

RenaultNissan-v0 (Validated Case Study)

# Parameters calibrated to Renault-Nissan Alliance (1999-2025)
validated_params = {
    'n_agents': 2,
    'max_steps': 26,  # 26 years
    'phases': 5,      # Crisis, Recovery, Growth, Ghosn, Post-Ghosn
    'theta': 20.0,
    'gamma': 0.60,
    'lambda_plus': 0.08,   # Slower trust building (cross-cultural)
    'lambda_minus': 0.32,  # Standard erosion
    'mu_R': 0.65,          # Higher damage (visible scandals)
    'delta_R': 0.02,       # Slower forgetting
    'T_init': 0.30,        # Started in crisis
    'R_init': 0.20,        # Some initial reputation damage
}

Parameter Sensitivity Analysis

High-Impact Parameters

Parameters with largest effect on environment dynamics:

Parameter Sensitivity Impact
$\lambda^-/\lambda^+$ ratio Very High Determines trust recovery feasibility
$\gamma$ (complementarity) High Controls cooperative incentive strength
$D_{ij}$ (interdependence) High Shapes utility landscape
$\mu_R$ (damage severity) Medium-High Determines hysteresis strength

Low-Impact Parameters

Parameters that can be approximated without significant effect:

Parameter Sensitivity Notes
$\kappa$ (signal sensitivity) Low 1.0 works for most scenarios
$\delta_R$ (reputation decay) Low Affects long-horizon only
$\theta$ vs $\beta$ (specification) Low within spec Both work, $\theta$ slightly better validated

Sensitivity Recommendations

# For robustness testing, vary these parameters:
sensitivity_ranges = {
    'lambda_ratio': [2.5, 3.0, 3.5, 4.0],  # Primary sensitivity
    'gamma': [0.55, 0.65, 0.75],            # Secondary sensitivity
    'mu_R': [0.50, 0.60, 0.70],             # Tertiary sensitivity
}

# Keep these fixed at validated values:
fixed_params = {
    'theta': 20.0,
    'kappa': 1.0,
    'delta_R': 0.03,
}

Calibration Workflow

For New Scenarios

  1. Start with defaults: Use recommended values from this document
  2. Identify analogous case: Find most similar validated environment
  3. Adjust key parameters: Focus on high-impact parameters
  4. Validate qualitatively: Check behavior matches domain expectations
  5. Run sensitivity analysis: Test robustness across parameter ranges

For Research Extensions

  1. Define hypothesis: What parameter relationship are you testing?
  2. Design sweep: Systematic variation of parameters
  3. Measure outcomes: Return, trust, cooperation patterns
  4. Report ranges: Document which parameter ranges support findings

Example: Custom Manufacturing Partnership

# Step 1: Start with SLCD as base (validated manufacturing JV)
params = slcd_params.copy()

# Step 2: Adjust for specific context
params['D_matrix'] = [[0.0, 0.60],   # Higher mutual dependency
                      [0.55, 0.0]]
params['gamma'] = 0.70               # Higher complementarity (specialized assets)
params['T_init'] = 0.65              # Prior positive relationship

# Step 3: Run and validate
env = coopetition_gym.make("TrustDilemma-v0", **params)
# ... run episodes and verify reasonable dynamics

Benchmark-Derived Insights

From 760 experiments across 20 algorithms:

Parameter Combinations That Work Well

Configuration Result Insight
$\lambda^-/\lambda^+ = 3.0$, $T_{\text{init}} = 0.50$ Trust-Return r=0.552 Standard validated setup
High $\gamma$ + High $D$ Cooperation emergence Strong incentives align behavior
Moderate all parameters Robust performance Avoids edge case instabilities

Parameter Combinations That Fail

Configuration Result Insight
$\lambda^-/\lambda^+ > 5.0$ Trust collapse Recovery becomes impossible
$\gamma < 0.3$ Defection dominance Insufficient cooperative incentive
$D_{ij} \to 0$ for all pairs Pure competition No structural incentive for cooperation

Citation

For parameter validation methodology:

@article{pant2025tr1,
  title={Computational Foundations for Strategic Coopetition: Formalizing Interdependence and Complementarity},
  author={Pant, Vik and Yu, Eric},
  journal={arXiv preprint arXiv:2510.18802},
  year={2025},
  note={Section 7: Validation methodology; 22,000+ trials}
}

@article{pant2025tr2,
  title={Computational Foundations for Strategic Coopetition: Formalizing Trust and Reputation Dynamics},
  author={Pant, Vik and Yu, Eric},
  journal={arXiv preprint arXiv:2510.24909},
  year={2025},
  note={Section 7: Parameter validation; 78,125 configurations}
}

@article{pant2026tr3,
  title={Computational Foundations for Strategic Coopetition: Formalizing Collective Action and Loyalty},
  author={Pant, Vik and Yu, Eric},
  journal={arXiv preprint arXiv:2601.16237},
  year={2026},
  note={Loyalty parameters: $\phi_B$=0.8, $\phi_C$=0.3; Apache validation 52/60}
}

@article{pant2026tr4,
  title={Computational Foundations for Strategic Coopetition: Formalizing Sequential Interaction and Reciprocity},
  author={Pant, Vik and Yu, Eric},
  journal={arXiv preprint arXiv:2604.01240},
  year={2026},
  note={Reciprocity parameters: $\rho_0$, $\eta$, $\kappa$, $k$; Apple App Store validation 48/55}
}

Technical Reports