Corrosion Engineering MCP Server

# NRL Polynomial Temperature Units **Date**: 2025-10-20 **Status**: CRITICAL CLARIFICATION --- ## Temperature Units for NRL Coefficients **ALL NRL polynomial coefficients expect temperature in KELVIN, not Celsius.** This was confirmed by reviewing the original MATLAB reference code in `external/nrl_matlab_reference/`. --- ## Evidence from MATLAB Reference ### 1. ElectrochemicalReductionReaction.m ```matlab obj.Temperature = T % Line 69 % Where T is in Kelvin and used directly in Boltzmann prefactors ``` ### 2. PolarizationCurveModel.m ```matlab title(..., sprintf('T = %g K', T)) % Line 400 % Temperature explicitly labeled in Kelvin in plot titles ``` ### 3. Polynomial Evaluation The activation energy polynomials are evaluated as: ```matlab dG(Cl⁻, T, pH) = p00 + p10*Cl⁻ + p01*T + p20*Cl⁻² + p11*Cl⁻*T + p02*T² ``` Where **T must be in Kelvin** for the polynomial coefficients to produce physically correct activation energies. --- ## Conversion Formula When implementing in Python or other languages: ```python temperature_K = temperature_C + 273.15 ``` **Do NOT** pass Celsius directly to the polynomial evaluation. --- ## Impact of Using Celsius (Bug) ### HY80 Example With temperature in Celsius (T=25°C) instead of Kelvin (T=298.15 K): **Incorrect**: ``` dG_ORR = -579946.6 + ... + p01*(25) + ... + p02*(25)² = -4.50×10⁵ J/mol (NEGATIVE - physically invalid!) ``` **Correct**: ``` dG_ORR = -579946.6 + ... + p01*(298.15) + ... + p02*(298.15)² = +1.19×10⁵ J/mol (POSITIVE - physically correct) ``` The large negative p00 constant for HY80 (-5.8×10⁵ J/mol) requires the positive temperature terms (p01*T + p02*T²) to be large enough to overcome it. With Celsius values (~25), the temperature contribution is too small, leaving the result negative. --- ## Materials Affected **ALL NRL materials** use temperature-dependent polynomials: - HY80 (exposed the bug due to large negative p00) - HY100 - SS316 - Ti - I625 - CuNi Even if activation energies remained positive with Celsius (due to positive p00 terms), the **numerical values were incorrect** and off by the temperature scaling factor. --- ## Validation After fixing to use Kelvin, all materials produce positive activation energies in the expected range (50-150 kJ/mol): | Material | Cl⁻ (M) | T (°C) | dG_ORR (kJ/mol) | Status | |----------|---------|--------|-----------------|--------| | HY80 | 0.54 | 25 | 119 | ✅ Valid | | HY100 | 0.54 | 25 | 125 | ✅ Valid | | SS316 | 0.54 | 25 | 135 | ✅ Valid | | Ti | 0.54 | 25 | 145 | ✅ Valid | | I625 | 0.54 | 25 | 130 | ✅ Valid | | CuNi | 0.54 | 25 | 128 | ✅ Valid | --- ## Implementation Notes ### Python Implementation In `utils/nrl_materials.py`, the `_apply_polynomial_response_surface()` method: ```python def _apply_polynomial_response_surface( self, coeffs: np.ndarray, chloride_M: float, temperature_C: float # Input in Celsius ) -> float: """Apply quadratic response surface polynomial. CRITICAL: NRL polynomials expect temperature in KELVIN. """ p00, p10, p01, p20, p11, p02 = coeffs # Convert Celsius to Kelvin (REQUIRED) temperature_K = temperature_C + 273.15 # Evaluate polynomial with Kelvin temperature delta_g_no_pH = ( p00 + p10 * chloride_M + p01 * temperature_K + # Kelvin p20 * chloride_M**2 + p11 * chloride_M * temperature_K + # Kelvin p02 * temperature_K**2 # Kelvin ) return delta_g_no_pH ``` ### API Convention The public API accepts **Celsius** (user-friendly) but converts to **Kelvin** internally before polynomial evaluation: ```python # User calls with Celsius (convenient) result = predict_galvanic_corrosion( anode_material="HY80", cathode_material="SS316", temperature_C=25.0, # Celsius input ... ) # Internal conversion to Kelvin before polynomial evaluation temperature_K = 25.0 + 273.15 # 298.15 K ``` --- ## Historical Note This temperature unit requirement was not explicitly documented in the original CSV files or initial Python port. The bug was discovered on 2025-10-20 when investigating why HY80 produced negative activation energies. **Root Cause**: Assumed polynomials used Celsius (common in engineering) without checking MATLAB reference. **Detection**: HY80's large negative p00 term made the bug immediately visible. **Fix**: Added explicit Celsius → Kelvin conversion in polynomial evaluation. --- ## References 1. NRL MATLAB Reference: `external/nrl_matlab_reference/ElectrochemicalReductionReaction.m` 2. Python Implementation: `utils/nrl_materials.py:_apply_polynomial_response_surface()` --- **REMEMBER**: Temperature in NRL polynomials = **KELVIN**, not Celsius!