Atomizer/studies/M1_Mirror

M1 Mirror - Telescope Primary Mirror Optimization

Project Type: Telescope Primary Mirror Support Structure Optimization Material: Zerodur Class 2 Glass Ceramic Analysis: NX Nastran SOL 101 (Linear Static, 4 Subcases) Status: Active Optimization Campaign

---

1. Project Overview

The M1 Mirror project optimizes the support structure of a telescope primary mirror to minimize wavefront error (WFE) across operational elevation angles while managing mass and manufacturing constraints.

1.1 What We're Optimizing

The mirror itself is a fixed optical prescription. We optimize how it's supported:

• Whiffle-tree axial support (18 contact points) - distributes gravity load evenly
• Lateral support (3 contact points) - prevents sideways motion at different elevations
• Mirror blank geometry - backface angle, rib structure affecting mass

1.2 Why This Matters

At different telescope elevation angles, gravity acts on the mirror differently:

• 90° (zenith): Gravity purely axial - whiffle-tree handles load
• 20-60° (operational): Mixed axial + lateral gravity component
• Polishing (horizontal): Different deformation than operational use

Poor support design causes:

• Tracking errors (WFE change between elevations)
• Manufacturing difficulty (optician must polish out gravity sag)
• Excessive mass (over-designed structure)

---

2. Optical Prescription

Source: Validated optical design specification (2025-12-22)

Parameter	Value	Tolerance	Units
Radius of Curvature (R)	2890	± 3	mm
Conic Constant (K)	-0.987	± 0.001	-
Clear Aperture (Ø)	1202.00	-	mm
Central Bore (Ø)	271.56	-	mm
Focal Length	1445	-	mm
f-number (f/D)	f/1.20	-	-
Surface Type	Near-Parabolic	-	-

Notes:

• Surface is concave (front surface)
• Conic constant K = -0.987 indicates near-parabolic (K = -1 is pure parabola)
• Focal length derived: f = R/2 = 1445 mm
• FEA mesh (1300 mm) includes lateral support nodes beyond the 1202 mm optical clear aperture

2.1 Usage in OPD Extractor

For rigorous WFE analysis (especially lateral support optimization), use explicit focal length:

from optimization_engine.extractors import ZernikeOPDExtractor

extractor = ZernikeOPDExtractor(
    op2_file,
    focal_length=1445.0,  # mm - from optical prescription R/2
    concave=True
)

2.2 Mesh vs Optical Aperture

Geometry	Mesh Value	Optical Value	Reason
Diameter	1300 mm	1202 mm	Mesh includes lateral support structure
Central hole	135 mm	271.56 mm	Mesh underestimates bore size

When analyzing WFE, filter nodes to clear aperture (r < 601 mm) or use the optical aperture radius in post-processing.

---

3. Physical System

3.1 Mirror Blank

Property	Value	Notes
Material	Zerodur Class 2	E=91 GPa, ρ=2.53 g/cm³, CTE≈0
Mass Target	< 105 kg	Hard constraint
Current Best	~94-95 kg	Achieved in V8/V9

3.2 Support Structure

System	Points	Purpose
Whiffle-tree (axial)	18	Distribute gravity load uniformly
Lateral supports	3	Prevent lateral motion at elevation
Vertical support skeleton	-	Structural frame

3.3 Loading

Subcase	Elevation	Gravity Direction	Purpose
1	90°	Pure -Z (axial)	Manufacturing/polishing reference
2	20°	Mixed	Reference for tracking
3	40°	Mixed	Operational tracking
4	60°	Mixed	Operational tracking

---

4. Optimization Objectives

4.1 WFE Metrics (Zernike-based)

Objective	Formula	Weight	Target	Description
40-20 Tracking	RMS_filt(Z₄₀ - Z₂₀)	5.0	4 nm	WFE change from 20° to 40°
60-20 Tracking	RMS_filt(Z₆₀ - Z₂₀)	5.0	10 nm	WFE change from 20° to 60°
Manufacturing	RMS_J1-J3(Z₉₀ - Z₂₀)	3.0	20 nm	Optician workload

Where:

• RMS_filt = RMS after removing J1-J4 (piston, tip, tilt, defocus)
• RMS_J1-J3 = RMS after removing only J1-J3 (keeps defocus for optician)

4.2 Mass

Objective	Weight	Target
M1_Blank mass	1.0	Minimize (< 105 kg hard limit)

4.3 Weighted Sum Formula

WS = 5×(40-20 nm) + 5×(60-20 nm) + 3×(MFG nm) + 1×(mass kg)

---

5. Design Variables

5.1 Whiffle Tree Parameters

Variable	Range	Units	Description
whiffle_min	30-72	mm	Minimum whiffle extent
whiffle_outer_to_vertical	70-85	deg	Outer arm to vertical angle
whiffle_triangle_closeness	65-120	mm	Triangle spacing
blank_backface_angle	4.2-4.5	deg	Mirror blank taper (affects mass)

5.2 Lateral Support Parameters

Variable	Range	Units	Description
lateral_inner_angle	25-32	deg	Inner lateral angle
lateral_outer_angle	9-17	deg	Outer lateral angle
lateral_outer_pivot	9-12	deg	Outer pivot position
lateral_inner_pivot	5-12	deg	Inner pivot position
lateral_middle_pivot	15-27	deg	Middle pivot position
lateral_closeness	7-12	deg	Lateral support closeness

5.3 Structural Parameters

Variable	Range	Units	Description
inner_circular_rib_dia	480-620	mm	Inner rib diameter
Pocket_Radius	Fixed 10.05	mm	Lightweighting pocket size

---

6. Optimization Campaign History

6.1 Campaign Phases

Phase 1: Initial Exploration (V11-V15)
─────────────────────────────────────
V11 (GNN + TuRBO)  →  V12 (GNN extended)  →  V13 (TPE validation)
    107 trials          ~5000 surrogate        291 trials
    WS = 129.33         WS = 129.33            WS = 129.33
                              ↓
V14 (TPE intensive)  →  V15 (NSGA-II)
    785 trials             126 trials
    WS = 121.72 ✓          WS = 121.72 (confirmed)

Phase 2: Cost Reduction (V6-V9)
─────────────────────────────────
V6 (TPE)  →  V7 (CMA-ES whiffle)  →  V8 (CMA-ES lateral)  →  V9 (CMA-ES combined)
281.82        268.75 (-4.6%)          266.02 (-5.6%)          In progress

6.2 Best Results Summary

Campaign	Best WS	40-20 nm	60-20 nm	MFG nm	Mass kg
V11-V15 (adaptive)	121.72	5.99	13.10	26.28	91.02
V6-V9 (cost reduction)	266.02*	6.03	12.81	25.73	94.66

*Different weighting scheme - not directly comparable

---

7. Sub-Studies Index

Study	Focus	Algorithm	Trials	Best WS	Status
m1_mirror_adaptive_V11	Initial GNN + TuRBO	GNN+TuRBO	107	129.33	Complete
m1_mirror_adaptive_V12	Extended surrogate	GNN+TuRBO	~5000	129.33	Complete
m1_mirror_adaptive_V13	TPE validation	TPE	291	129.33	Complete
m1_mirror_adaptive_V14	Intensive TPE	TPE	785	121.72	Complete
m1_mirror_adaptive_V15	NSGA-II Pareto	NSGA-II	126	121.72	Complete
m1_mirror_cost_reduction	Mass reduction baseline	TPE	150	-	Complete
m1_mirror_cost_reduction_V7	Whiffle refinement	CMA-ES	200	268.75	Complete
m1_mirror_cost_reduction_V8	Lateral optimization (Z-only)	CMA-ES	200	266.02	Complete
m1_mirror_cost_reduction_V9	Combined fine-tuning	CMA-ES	200	-	Complete
m1_mirror_cost_reduction_V10	Lateral + OPD (corrupted)	CMA-ES	-	-	Abandoned
m1_mirror_cost_reduction_V11	Lateral + OPD + extract_relative	CMA-ES	200	284.19	Complete
m1_mirror_cost_reduction_V12	Combined Whiffle + Lateral (10 vars)	CMA-ES	200	TBD	Active
m1_mirror_surrogate_turbo	GNN surrogate + turbo optimization	GNN+FEA	~1000	TBD	Setup Complete

Flat Back Studies (Cost Reduction Option C)

Study	Focus	Algorithm	Trials	Best WS	Status
m1_mirror_cost_reduction_flat_back_V3	Initial exploration	TPE	~297	-	Complete
m1_mirror_cost_reduction_flat_back_V4	Continued exploration	TPE	~19	-	Complete
m1_mirror_cost_reduction_flat_back_V5	MLP Surrogate + L-BFGS	Surrogate	45	290.18	Complete (Failed)
m1_mirror_cost_reduction_flat_back_V6	Pure TPE baseline	TPE	196	225.41	Complete
m1_mirror_cost_reduction_flat_back_V7	Self-Aware Turbo	SAT	200	TBD	Ready

Flat Back Notes:

• blank_backface_angle = 0 (flat backface eliminates custom jig during machining)
• Mass constraint: ≤ 120 kg
• Weighted sum: 6*wfe_40_20 + 5*wfe_60_20 + 3*mfg_90
• V5 failure: MLP surrogate led L-BFGS to "fake optima" (OOD extrapolation)
• V7 fix: EnsembleSurrogate with uncertainty quantification + OOD detection

---

8. Technical Notes

8.1 Zernike Analysis

• Standard method: Uses Z-displacement only at original (x,y) coordinates
• OPD method: Accounts for lateral (X,Y) displacement - critical for lateral support optimization
• extract_relative() method (V11+): Correct relative WFE computation:
  1. Compute WFE at each node for both subcases
  2. Compute node-by-node difference: WFE_rel[i] = WFE_target[i] - WFE_ref[i]
  3. Fit Zernike to the difference field
  4. Compute RMS of filtered difference

Important

: abs(RMS_target - RMS_ref) is WRONG. Always use extract_relative() for relative metrics.

See: docs/physics/ZERNIKE_OPD_METHOD.md

8.2 WFE Convention

WFE = 2 × surface_error    (reflection doubles path length)

All WFE values in nanometers (nm).

8.3 Coordinate System

• Z-axis: Optical axis (positive toward sky)
• Origin: Mirror vertex
• Concave mirror: Surface curves toward -Z

---

9. File Structure

studies/M1_Mirror/
├── README.md                          # This file (master documentation)
├── m1_mirror_adaptive_V11/            # Sub-studies...
├── m1_mirror_adaptive_V12/
├── ...
├── m1_mirror_cost_reduction_V9/
│   ├── 1_setup/
│   │   ├── optimization_config.json   # Study configuration
│   │   └── model/                     # NX model files
│   ├── 2_iterations/                  # FEA iteration folders
│   ├── 3_results/                     # Optuna DB, logs, archives
│   ├── run_optimization.py            # Main entry point
│   └── README.md                      # Study-specific docs

---

10. Quick Reference

Run Optimization

cd studies/M1_Mirror/m1_mirror_cost_reduction_V9
conda activate atomizer
python run_optimization.py --start

Check Progress

python -c "
import optuna
study = optuna.load_study(study_name='m1_mirror_cost_reduction_V9',
                          storage='sqlite:///3_results/study.db')
print(f'Trials: {len(study.trials)}')
print(f'Best: {study.best_value:.2f}')
"

Generate Insights

python -m optimization_engine.insights generate . --type zernike_wfe
python -m optimization_engine.insights generate . --type zernike_opd_comparison

---

11. Related Documentation

Document	Description
M1_MIRROR_CAMPAIGN_SUMMARY.md	V11-V15 campaign analysis
docs/physics/ZERNIKE_FUNDAMENTALS.md	Zernike analysis basics
docs/physics/ZERNIKE_OPD_METHOD.md	OPD method for lateral supports
.claude/skills/modules/extractors-catalog.md	Extractor quick reference

---

M1 Mirror Optimization Project Atomizer Framework Last Updated: 2026-01-20