@zigrivers/scaffold 3.14.0 → 3.16.0

package/content/knowledge/research/research-sim-engine-patterns.md

@@ -0,0 +1,448 @@
+---
+name: research-sim-engine-patterns
+description: Simulation engine integration patterns including wrapping solvers as callable experiments, configuration management, mesh handling, batch job submission, and result parsing
+topics: [research, simulation, engine-integration, openfoam, fenics, simpy, solver-configuration, mesh-management, batch-jobs, result-parsing]
+---
+
+Simulation engines (OpenFOAM, FEniCS, SimPy, COMSOL, Ansys) are typically standalone tools with their own input/output formats, solver configurations, and execution models. Wrapping them as callable experiments requires a uniform interface that abstracts engine-specific details while preserving access to solver parameters that matter for optimization. The key design challenge is creating a thin adapter layer that makes any simulation engine look like a function from parameters to results, without hiding failure modes or losing important solver diagnostics.
+
+## Summary
+
+Wrap simulation engines behind a `SimulationExperiment` interface that accepts a parameter dictionary and returns structured results including convergence status, wall-clock time, and domain-specific outputs. Manage solver configuration as declarative parameter objects that can be serialized for reproducibility. Handle mesh generation as a separate cacheable step with independence checks. Submit batch jobs through an abstraction that works locally or on HPC clusters. Parse results from engine-specific output files into a normalized format for the experiment tracker.
+
+## Deep Guidance
+
+### Simulation Experiment Interface
+
+Define a uniform interface that any simulation engine adapter must implement:
+
+```python
+# src/simulation/interface.py
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from enum import Enum
+from pathlib import Path
+from typing import Any
+
+class SimulationStatus(Enum):
+    CONVERGED = "converged"
+    DIVERGED = "diverged"
+    MAX_ITERATIONS = "max_iterations"
+    TIMEOUT = "timeout"
+    ERROR = "error"
+
+@dataclass
+class SimulationResult:
+    """Normalized result from any simulation engine."""
+    status: SimulationStatus
+    outputs: dict[str, float]  # Named scalar outputs (drag_coeff, stress_max, etc.)
+    fields: dict[str, Path]  # Paths to field data files (pressure, velocity, etc.)
+    residuals: list[float]  # Final residual history
+    wall_time_seconds: float
+    iterations: int
+    metadata: dict[str, Any] = field(default_factory=dict)
+
+    @property
+    def is_valid(self) -> bool:
+        return self.status == SimulationStatus.CONVERGED
+
+class SimulationExperiment(ABC):
+    """Interface for wrapping any simulation engine as a callable experiment."""
+
+    @abstractmethod
+    def setup(self, params: dict[str, Any]) -> Path:
+        """Generate input files for the simulation. Returns case directory."""
+
+    @abstractmethod
+    def run(self, case_dir: Path, timeout_seconds: int | None = None) -> SimulationResult:
+        """Execute the simulation and return parsed results."""
+
+    @abstractmethod
+    def validate_params(self, params: dict[str, Any]) -> list[str]:
+        """Check parameter validity before running. Returns list of issues."""
+
+    def __call__(self, params: dict[str, Any], timeout: int | None = None) -> SimulationResult:
+        """Run the full pipeline: validate -> setup -> run."""
+        issues = self.validate_params(params)
+        if issues:
+            raise ValueError(f"Invalid parameters: {issues}")
+        case_dir = self.setup(params)
+        return self.run(case_dir, timeout)
+```
+
+### OpenFOAM Adapter
+
+OpenFOAM uses directory-based case structures with text dictionaries for configuration:
+
+```python
+# src/simulation/engines/openfoam.py
+import subprocess
+import time
+from pathlib import Path
+from typing import Any
+
+from src.simulation.interface import SimulationExperiment, SimulationResult, SimulationStatus
+
+class OpenFOAMExperiment(SimulationExperiment):
+    """Wraps OpenFOAM as a callable simulation experiment."""
+
+    def __init__(self, template_dir: Path, solver: str = "simpleFoam"):
+        self.template_dir = template_dir
+        self.solver = solver
+
+    def validate_params(self, params: dict[str, Any]) -> list[str]:
+        issues = []
+        if "inlet_velocity" in params and params["inlet_velocity"] <= 0:
+            issues.append("inlet_velocity must be positive")
+        if "turbulence_model" in params:
+            valid_models = ["kEpsilon", "kOmegaSST", "SpalartAllmaras"]
+            if params["turbulence_model"] not in valid_models:
+                issues.append(f"turbulence_model must be one of {valid_models}")
+        return issues
+
+    def setup(self, params: dict[str, Any]) -> Path:
+        """Generate OpenFOAM case from template with parameter substitution."""
+        import shutil
+        import hashlib
+        import json
+
+        # Create unique case directory based on parameters
+        param_hash = hashlib.md5(json.dumps(params, sort_keys=True).encode()).hexdigest()[:8]
+        case_dir = Path(f"runs/openfoam_{param_hash}")
+        if case_dir.exists():
+            shutil.rmtree(case_dir)
+        shutil.copytree(self.template_dir, case_dir)
+
+        # Substitute parameters into OpenFOAM dictionaries
+        self._write_transport_properties(case_dir, params)
+        self._write_boundary_conditions(case_dir, params)
+        self._write_control_dict(case_dir, params)
+
+        return case_dir
+
+    def run(self, case_dir: Path, timeout_seconds: int | None = None) -> SimulationResult:
+        """Execute OpenFOAM solver and parse results."""
+        start = time.time()
+
+        # Run mesh generation if needed
+        self._run_mesh(case_dir)
+
+        # Run solver
+        result = subprocess.run(
+            [self.solver, "-case", str(case_dir)],
+            capture_output=True, text=True,
+            timeout=timeout_seconds,
+        )
+        wall_time = time.time() - start
+
+        # Parse results
+        return self._parse_results(case_dir, result, wall_time)
+
+    def _parse_results(self, case_dir: Path, proc, wall_time: float) -> SimulationResult:
+        """Parse OpenFOAM log and postProcessing output."""
+        residuals = self._extract_residuals(proc.stdout)
+        status = self._determine_status(residuals, proc.returncode)
+        outputs = self._read_force_coefficients(case_dir)
+        fields = self._find_field_files(case_dir)
+
+        return SimulationResult(
+            status=status,
+            outputs=outputs,
+            fields=fields,
+            residuals=residuals,
+            wall_time_seconds=wall_time,
+            iterations=len(residuals),
+            metadata={"solver": self.solver, "case_dir": str(case_dir)},
+        )
+
+    def _extract_residuals(self, log: str) -> list[float]:
+        """Extract residual values from solver log."""
+        import re
+        pattern = r"Solving for Ux.*Final residual = ([0-9.e+-]+)"
+        return [float(m) for m in re.findall(pattern, log)]
+
+    def _determine_status(self, residuals: list[float], returncode: int) -> SimulationStatus:
+        if returncode != 0:
+            return SimulationStatus.ERROR
+        if not residuals:
+            return SimulationStatus.ERROR
+        if residuals[-1] < 1e-6:
+            return SimulationStatus.CONVERGED
+        if residuals[-1] > residuals[0] * 100:
+            return SimulationStatus.DIVERGED
+        return SimulationStatus.MAX_ITERATIONS
+
+    def _run_mesh(self, case_dir: Path) -> None:
+        subprocess.run(["blockMesh", "-case", str(case_dir)], check=True, capture_output=True)
+
+    def _write_transport_properties(self, case_dir: Path, params: dict) -> None:
+        """Write constant/transportProperties with parameter values."""
+        props_file = case_dir / "constant" / "transportProperties"
+        nu = params.get("kinematic_viscosity", 1e-6)
+        props_file.write_text(
+            f"FoamFile {{ version 2.0; class dictionary; object transportProperties; }}\n"
+            f"nu [0 2 -1 0 0 0 0] {nu};\n"
+        )
+
+    def _write_boundary_conditions(self, case_dir: Path, params: dict) -> None:
+        """Write 0/ directory boundary condition files."""
+        # Implementation substitutes inlet velocity, turbulence quantities, etc.
+        pass
+
+    def _write_control_dict(self, case_dir: Path, params: dict) -> None:
+        """Write system/controlDict with iteration limits and write intervals."""
+        pass
+
+    def _read_force_coefficients(self, case_dir: Path) -> dict[str, float]:
+        """Read postProcessing/forceCoeffs output."""
+        coeffs_file = case_dir / "postProcessing" / "forceCoeffs" / "0" / "coefficient.dat"
+        if not coeffs_file.exists():
+            return {}
+        lines = coeffs_file.read_text().strip().split("\n")
+        last_line = lines[-1].split()
+        return {"Cd": float(last_line[1]), "Cl": float(last_line[2])}
+
+    def _find_field_files(self, case_dir: Path) -> dict[str, Path]:
+        """Find the latest time directory with field outputs."""
+        time_dirs = sorted(
+            [d for d in case_dir.iterdir() if d.is_dir() and d.name.replace(".", "").isdigit()],
+            key=lambda d: float(d.name),
+        )
+        if not time_dirs:
+            return {}
+        latest = time_dirs[-1]
+        return {f.stem: f for f in latest.iterdir() if f.is_file()}
+```
+
+### FEniCS Adapter
+
+FEniCS uses Python-native problem definitions with mesh objects and variational forms:
+
+```python
+# src/simulation/engines/fenics_adapter.py
+from pathlib import Path
+from typing import Any
+import numpy as np
+
+from src.simulation.interface import SimulationExperiment, SimulationResult, SimulationStatus
+
+class FEniCSExperiment(SimulationExperiment):
+    """Wraps FEniCS as a callable experiment for PDE solving."""
+
+    def __init__(self, mesh_path: Path, problem_class: type):
+        self.mesh_path = mesh_path
+        self.problem_class = problem_class
+
+    def validate_params(self, params: dict[str, Any]) -> list[str]:
+        issues = []
+        if "mesh_refinement" in params and params["mesh_refinement"] < 1:
+            issues.append("mesh_refinement must be >= 1")
+        if "youngs_modulus" in params and params["youngs_modulus"] <= 0:
+            issues.append("youngs_modulus must be positive")
+        return issues
+
+    def setup(self, params: dict[str, Any]) -> Path:
+        """Prepare mesh and problem configuration."""
+        import json
+        case_dir = Path(f"runs/fenics_{hash(frozenset(params.items())) % 10**8:08d}")
+        case_dir.mkdir(parents=True, exist_ok=True)
+        (case_dir / "params.json").write_text(json.dumps(params))
+        return case_dir
+
+    def run(self, case_dir: Path, timeout_seconds: int | None = None) -> SimulationResult:
+        """Solve the PDE problem with given parameters."""
+        import json
+        import time
+
+        params = json.loads((case_dir / "params.json").read_text())
+        start = time.time()
+
+        try:
+            problem = self.problem_class(self.mesh_path, params)
+            solution, residuals = problem.solve()
+            wall_time = time.time() - start
+
+            outputs = problem.extract_quantities(solution)
+            solution_path = case_dir / "solution.xdmf"
+            problem.save_solution(solution, solution_path)
+
+            status = (
+                SimulationStatus.CONVERGED
+                if residuals[-1] < problem.tolerance
+                else SimulationStatus.MAX_ITERATIONS
+            )
+
+            return SimulationResult(
+                status=status,
+                outputs=outputs,
+                fields={"solution": solution_path},
+                residuals=residuals,
+                wall_time_seconds=wall_time,
+                iterations=len(residuals),
+            )
+        except Exception as e:
+            return SimulationResult(
+                status=SimulationStatus.ERROR,
+                outputs={},
+                fields={},
+                residuals=[],
+                wall_time_seconds=time.time() - start,
+                iterations=0,
+                metadata={"error": str(e)},
+            )
+```
+
+### Solver Configuration as Parameters
+
+Treat solver settings as first-class parameters that the optimizer can tune:
+
+```python
+# src/simulation/config.py
+from dataclasses import dataclass
+from typing import Any
+
+@dataclass
+class SolverConfig:
+    """Declarative solver configuration -- serializable and reproducible."""
+    solver_type: str  # "GAMG", "PCG", "BiCGStab"
+    preconditioner: str  # "DIC", "DILU", "none"
+    tolerance: float = 1e-6
+    relative_tolerance: float = 0.01
+    max_iterations: int = 1000
+    relaxation_factor: float = 0.7
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "solver": self.solver_type,
+            "preconditioner": self.preconditioner,
+            "tolerance": self.tolerance,
+            "relTol": self.relative_tolerance,
+            "maxIter": self.max_iterations,
+            "relaxationFactor": self.relaxation_factor,
+        }
+
+def solver_param_space() -> dict[str, Any]:
+    """Define solver parameters as an optimizable space."""
+    return {
+        "solver_type": {"type": "categorical", "choices": ["GAMG", "PCG", "BiCGStab"]},
+        "preconditioner": {"type": "categorical", "choices": ["DIC", "DILU", "none"]},
+        "tolerance": {"type": "continuous", "low": 1e-8, "high": 1e-4, "log": True},
+        "relaxation_factor": {"type": "continuous", "low": 0.3, "high": 0.95},
+        "max_iterations": {"type": "discrete", "low": 100, "high": 5000, "step": 100},
+    }
+```
+
+### Batch Job Submission
+
+Abstract job submission to work locally or on HPC clusters:
+
+```python
+# src/simulation/batch.py
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from enum import Enum
+from pathlib import Path
+
+class JobStatus(Enum):
+    PENDING = "pending"
+    RUNNING = "running"
+    COMPLETED = "completed"
+    FAILED = "failed"
+    TIMEOUT = "timeout"
+
+@dataclass
+class JobSpec:
+    """Specification for a simulation job."""
+    case_dir: Path
+    command: list[str]
+    num_cores: int = 1
+    wall_time_hours: float = 1.0
+    memory_gb: float = 4.0
+    partition: str = "default"
+
+class JobSubmitter(ABC):
+    """Abstract job submission interface."""
+
+    @abstractmethod
+    def submit(self, spec: JobSpec) -> str:
+        """Submit job, return job ID."""
+
+    @abstractmethod
+    def status(self, job_id: str) -> JobStatus:
+        """Check job status."""
+
+    @abstractmethod
+    def wait(self, job_id: str, poll_interval: float = 30.0) -> JobStatus:
+        """Block until job completes."""
+
+class LocalSubmitter(JobSubmitter):
+    """Run jobs locally as subprocesses."""
+
+    def __init__(self):
+        self._processes: dict[str, Any] = {}
+        self._counter = 0
+
+    def submit(self, spec: JobSpec) -> str:
+        import subprocess
+        self._counter += 1
+        job_id = f"local_{self._counter}"
+        proc = subprocess.Popen(
+            spec.command,
+            cwd=spec.case_dir,
+            stdout=open(spec.case_dir / "stdout.log", "w"),
+            stderr=open(spec.case_dir / "stderr.log", "w"),
+        )
+        self._processes[job_id] = proc
+        return job_id
+
+    def status(self, job_id: str) -> JobStatus:
+        proc = self._processes[job_id]
+        if proc.poll() is None:
+            return JobStatus.RUNNING
+        return JobStatus.COMPLETED if proc.returncode == 0 else JobStatus.FAILED
+
+    def wait(self, job_id: str, poll_interval: float = 30.0) -> JobStatus:
+        self._processes[job_id].wait()
+        return self.status(job_id)
+```
+
+### Result Parsing
+
+Parse engine-specific output into normalized formats for the experiment tracker:
+
+```python
+# src/simulation/parsers.py
+from pathlib import Path
+from typing import Any
+import re
+
+def parse_openfoam_log(log_path: Path) -> dict[str, Any]:
+    """Extract convergence data from OpenFOAM solver log."""
+    text = log_path.read_text()
+    residual_pattern = r"Time = (\d+)\n.*?Solving for (\w+).*?Final residual = ([0-9.e+-]+)"
+    matches = re.findall(residual_pattern, text, re.DOTALL)
+
+    residuals_by_field: dict[str, list[float]] = {}
+    for time_step, field_name, residual in matches:
+        residuals_by_field.setdefault(field_name, []).append(float(residual))
+
+    execution_time_match = re.search(r"ExecutionTime = ([0-9.]+) s", text)
+    exec_time = float(execution_time_match.group(1)) if execution_time_match else 0.0
+
+    return {
+        "residuals_by_field": residuals_by_field,
+        "execution_time": exec_time,
+        "num_iterations": len(matches) // max(len(residuals_by_field), 1),
+    }
+
+def parse_csv_results(results_path: Path, output_columns: list[str]) -> dict[str, float]:
+    """Parse final row of CSV results file for scalar outputs."""
+    import csv
+    with open(results_path) as f:
+        reader = csv.DictReader(f)
+        rows = list(reader)
+    if not rows:
+        return {}
+    last_row = rows[-1]
+    return {col: float(last_row[col]) for col in output_columns if col in last_row}
+```