[Iter 4] Code modification in PDE_D_MichaelisMenten.py

[Automated commit by Claude]

Author: allierc

src/ParticleGraph/generators/PDE_D_MichaelisMenten.py

@@ -0,0 +1,234 @@
+import torch
+import torch_geometric as pyg
+import torch_geometric.utils as pyg_utils
+from ParticleGraph.utils import to_numpy
+
+class PDE_D_MichaelisMenten(pyg.nn.MessagePassing):
+    """
+    Michaelis-Menten saturating kinetics particle model.
+
+    Consumption and production rates depend on local field concentration
+    via Michaelis-Menten saturation: rate_eff = base_rate * |C1| / (Km + |C1|).
+
+    Literature:
+    - Michaelis, L. & Menten, M. L. (1913) Biochem Z 49:333-369
+      "Die Kinetik der Invertinwirkung"
+
+    Physics:
+    rate_effective = base_rate * |C1| / (Km + |C1|)
+    At low C1: rate ~ base_rate * C1/Km (linear, slow)
+    At high C1: rate ~ base_rate (saturated, maximum)
+
+    Per-type params layout: [M1, M2, consumption, production, ar_p1, ar_p2, ar_p3, ar_p4]
+    """
+
+    PARAMS_DOC = {
+        "model_name": "MichaelisMenten",
+        "literature": "Michaelis & Menten (1913) Biochem Z 49:333-369",
+        "description": "Saturating concentration-dependent consumption/production kinetics",
+        "equations": {
+            "field_to_particle": "v = (M1*grad_C1 + M2*grad_C2) * dir",
+            "particle_to_field": "dC1 = -consumption * |C1|/(Km+|C1|) * w(r), dC2 = production * |C1|/(Km+|C1|) * w(r)",
+            "particle_to_particle": "f = (p1*exp(-d^(2p2)/(2sigma^2)) - p3*exp(-d^(2p4)/(2sigma^2))) * dir"
+        },
+        "params_mesh": [
+            {
+                "row": 0, "description": "C1 field parameters (shared with mesh model)",
+                "slots": [
+                    {"index": 0, "name": "D1", "description": "Diffusion coeff for C1 (mesh model)", "typical_range": [0.01, 0.5]},
+                    {"index": 1, "name": "Da_c", "description": "Damkohler number (mesh model)", "typical_range": [1.0, 50.0]},
+                    {"index": 2, "name": "A", "description": "Brusselator param A (mesh model)", "typical_range": [0.5, 5.0]},
+                    {"index": 3, "name": "B", "description": "Brusselator param B (mesh model)", "typical_range": [1.0, 10.0]},
+                    {"index": 4, "name": "mu", "description": "Morphological parameter (mesh model)", "typical_range": [0.01, 0.1]},
+                    {"index": 5, "name": "M1", "description": "Mobility coefficient for C1 gradients", "typical_range": [-16, 16]}
+                ]
+            },
+            {
+                "row": 1, "description": "C2 field parameters + MM control",
+                "slots": [
+                    {"index": 0, "name": "D2", "description": "Diffusion coeff for C2 (mesh model)", "typical_range": [0.1, 1.0]},
+                    {"index": 1, "name": "M2", "description": "Mobility coefficient for C2 gradients", "typical_range": [-16, 16]},
+                    {"index": 2, "name": "mm_Km", "description": "Michaelis-Menten half-saturation constant (0=constant rate)", "typical_range": [0.1, 5.0]}
+                ]
+            },
+            {
+                "row": 2, "description": "Particle-field coupling",
+                "slots": [
+                    {"index": 0, "name": "Pe", "description": "Peclet number", "typical_range": [0.5, 2.0]},
+                    {"index": 1, "name": "consumption", "description": "Particle consumption rate of C1", "typical_range": [10, 200]},
+                    {"index": 2, "name": "production", "description": "Particle production rate of C2", "typical_range": [-200, -10]},
+                    {"index": 3, "name": "influence_radius", "description": "Gaussian influence radius for pf coupling", "typical_range": [0.01, 0.1]}
+                ]
+            }
+        ],
+        "particle_params": {
+            "description": "Per-type params from simulation.params (one row per n_particle_types)",
+            "slots": [
+                {"index": 0, "name": "M1", "description": "Per-type mobility for C1"},
+                {"index": 1, "name": "M2", "description": "Per-type mobility for C2"},
+                {"index": 2, "name": "consumption", "description": "Per-type consumption rate"},
+                {"index": 3, "name": "production", "description": "Per-type production rate"},
+                {"index": 4, "name": "ar_p1", "description": "Attraction strength"},
+                {"index": 5, "name": "ar_p2", "description": "Attraction exponent"},
+                {"index": 6, "name": "ar_p3", "description": "Repulsion strength"},
+                {"index": 7, "name": "ar_p4", "description": "Repulsion exponent"}
+            ]
+        }
+    }
+
+    def __init__(self, aggr_type='mean', p=None, particle_params=None, bc_dpos=None, dimension=2, sigma=0.005):
+        super(PDE_D_MichaelisMenten, self).__init__(aggr=aggr_type)
+
+        self.p = p
+        self.particle_params = particle_params
+        self.bc_dpos = bc_dpos
+        self.dimension = dimension
+        self.sigma = sigma
+
+        self.M1 = p[0, 5]
+        self.M2 = p[1, 1]
+        self.consumption_rate = p[2, 1]
+        self.production_rate = p[2, 2]
+        self.influence_radius = p[2, 3]
+        self.Pe = p[2, 0]
+        self.repulsion_strength = 50
+        self.repulsion_range = 0.04
+
+        self.mm_Km = p[1, 2] if p.shape[1] > 2 else 0.0
+
+        print(f"initialized PDE_D_MichaelisMenten with parameters:")
+        print(f"  mobility: M1={self.M1.item()}, M2={self.M2.item()}")
+        mm_val = self.mm_Km.item() if hasattr(self.mm_Km, 'item') else self.mm_Km
+        print(f"  Km={mm_val:.3f} (rate = base * |C1|/(Km+|C1|), Michaelis-Menten 1913)")
+        print(f"  Pe={self.Pe.item():.3f}, sigma={self.sigma}")
+        print(f"  particle->field: consumption={self.consumption_rate.item()}, production={self.production_rate.item()}, influence_radius={self.influence_radius.item():.3f}")
+        if particle_params is not None:
+            print(f"  multi-type support: {particle_params.shape[0]} particle types")
+
+    def forward(self, data, direction='fp'):
+        x, edge_index = data.x, data.edge_index
+        edge_index, _ = pyg_utils.remove_self_loops(edge_index)
+
+        if self.particle_params is not None:
+            particle_type = x[:, 1 + 2*self.dimension].long()
+            max_type = particle_type.max().item()
+            n_param_rows = self.particle_params.shape[0]
+            if max_type >= n_param_rows:
+                raise ValueError(
+                    f"PDE_D_MichaelisMenten: particle_params has {n_param_rows} rows but found "
+                    f"particle type {max_type}. Need {max_type + 1} rows in simulation.params."
+                )
+            parameters = self.particle_params[to_numpy(particle_type), :]
+        else:
+            parameters = None
+
+        if direction == 'interpolate':
+            result = self.propagate(edge_index, x=x, mode='interpolate', parameters=parameters)
+            pos = x[:, 1:self.dimension+1]
+            in_box = ((pos >= 0) & (pos <= 1)).all(dim=1, keepdim=True)
+            result = result * in_box.float()
+            return result
+        elif direction == 'fp':
+            result = self.propagate(edge_index, x=x, mode='fp', parameters=parameters)
+            pos = x[:, 1:self.dimension+1]
+            in_box = ((pos >= 0) & (pos <= 1)).all(dim=1, keepdim=True)
+            result = result * in_box.float()
+            return result
+        elif direction == 'pf':
+            result = self.propagate(edge_index, x=x, mode='pf', parameters=parameters)
+            return result
+        else:
+            result = self.propagate(edge_index, x=x, mode='pp', parameters=parameters)
+            return result
+
+    def message(self, edge_index_i, edge_index_j, x_i, x_j, mode=None, parameters_i=None):
+        pos_i = x_i[:, 1:self.dimension+1]
+        pos_j = x_j[:, 1:self.dimension+1]
+
+        d_pos = self.bc_dpos(pos_j - pos_i)
+        dist = torch.sqrt(torch.sum(d_pos**2, dim=1))
+        dist_safe = torch.clamp(dist, min=1e-6)
+
+        if mode == 'interpolate':
+            C1_mesh = x_j[:, 6:7]
+            C2_mesh = x_j[:, 7:8]
+            weight = torch.exp(-dist / 0.01).unsqueeze(1)
+            return torch.cat([C1_mesh * weight, C2_mesh * weight, weight], dim=1)
+
+        elif mode == 'fp':
+            fields_i = x_i[:, 6:8]
+            fields_j = x_j[:, 6:8]
+
+            dC1 = fields_j[:, 0:1] - fields_i[:, 0:1]
+            dC2 = fields_j[:, 1:2] - fields_i[:, 1:2]
+
+            kernel = torch.exp(-dist / 0.05)
+            dir_norm = d_pos / dist_safe.unsqueeze(1)
+            domain_scale = 32.0
+            grad_C1 = (dC1 * kernel.unsqueeze(1)) / (dist_safe.unsqueeze(1) * domain_scale)
+            grad_C2 = (dC2 * kernel.unsqueeze(1)) / (dist_safe.unsqueeze(1) * domain_scale)
+
+            if parameters_i is not None:
+                M1 = parameters_i[:, 0:1]
+                M2 = parameters_i[:, 1:2]
+            else:
+                M1 = self.M1
+                M2 = self.M2
+
+            velocities = (M1 * grad_C1 + M2 * grad_C2) * dir_norm
+            return velocities
+
+        elif mode == 'pf':
+            weights = torch.exp(-dist**2 / (2 * (self.influence_radius/3)**2))
+
+            if parameters_i is not None:
+                consumption = parameters_i[:, 2]
+                production = parameters_i[:, 3]
+            else:
+                consumption = self.consumption_rate
+                production = self.production_rate
+
+            # Michaelis-Menten modulation
+            if self.mm_Km > 0:
+                C1_local = torch.abs(x_i[:, 6]) + 1e-6
+                mm_factor = C1_local / (self.mm_Km + C1_local)
+                consumption = consumption * mm_factor
+                production = production * mm_factor
+
+            field_updates = torch.zeros((pos_i.size(0), 2), device=pos_i.device)
+            field_updates[:, 0] = -consumption * weights
+            field_updates[:, 1] = production * weights
+            return field_updates
+
+        else:  # mode == 'pp'
+            if parameters_i is not None:
+                p1 = parameters_i[:, 4]
+                p2 = parameters_i[:, 5]
+                p3 = parameters_i[:, 6]
+                p4 = parameters_i[:, 7]
+
+                f = (p1 * torch.exp(-dist ** (2 * p2) / (2 * self.sigma ** 2))
+                     - p3 * torch.exp(-dist ** (2 * p4) / (2 * self.sigma ** 2)))
+
+                forces = f[:, None] * d_pos / dist_safe.unsqueeze(1)
+            else:
+                forces = torch.zeros_like(pos_i)
+                in_range = dist < self.repulsion_range
+                if in_range.any():
+                    dir_norm = d_pos / dist_safe.unsqueeze(1)
+                    repulsion_mag = self.repulsion_strength * torch.exp(
+                        -5.0 * dist[in_range] / self.repulsion_range
+                    )
+                    forces[in_range] = -dir_norm[in_range] * repulsion_mag.unsqueeze(1)
+
+            return forces
+
+    def update(self, aggr_out, mode=None):
+        if mode == 'interpolate':
+            C1_weighted = aggr_out[:, 0:1]
+            C2_weighted = aggr_out[:, 1:2]
+            weight_sum = aggr_out[:, 2:3]
+            weight_sum = torch.clamp(weight_sum, min=1e-10)
+            return torch.cat([C1_weighted / weight_sum, C2_weighted / weight_sum], dim=1)
+        else:
+            return aggr_out