Fix DM21 functionals incorrectly defaulting to DM21M (#589)

[nsrawat0333]

Summary

Fixes Issue #589 where all DM21 functionals (DM21, DM21MC, DM21MU) were incorrectly defaulting to DM21M regardless of which functional was specified by users.

Problem

Users reported that when attempting to use different DM21 functionals in their neural network research, all functionals would run as DM21M instead of the requested functional (DM21, DM21MC, or DM21MU). This prevented researchers from accessing the full suite of DM21 density functionals.

Root Cause

The issue was caused by improper TensorFlow session and graph isolation between different NeuralNumInt instances:

• TensorFlow v1 sessions were not explicitly bound to their respective graphs
• Multiple functional instances shared state, causing interference
• Lack of proper cleanup methods led to resource leaks and persistent state

Solution

1. Enhanced Session Isolation

• Modified session creation to explicitly bind to functional-specific graphs
• Ensures each functional has isolated TensorFlow computation

2. Added Resource Management

• Implemented close() method for explicit session cleanup
• Added __del__() method for automatic resource management
• Prevents memory leaks when switching between functionals

3. Improved Error Handling

• Added validation for graph context integrity
• Enhanced error messages for failed model loading
• Better debugging information for troubleshooting

Testing

• ✅ Comprehensive test suite validates functional isolation
• ✅ Verified each functional loads correct neural network model
• ✅ Confirmed proper resource cleanup and session management
• ✅ Backward compatibility maintained

Usage Impact

# Now works correctly - each functional loads its specific model
functionals = [
    dm21.Functional.DM21,    # ✅ Full training dataset
    dm21.Functional.DM21m,   # ✅ Molecules only  
    dm21.Functional.DM21mc,  # ✅ Molecules + fractional charge
    dm21.Functional.DM21mu   # ✅ Molecules + electron gas
]

for functional in functionals:
    mf._numint = dm21.NeuralNumInt(functional)  # Each loads correct model
    # ... perform calculations ...
    mf._numint.close()  # Clean up properly

[polarbe]

Dooray! Failure Notice Failure Notice Your message sent to ***@***.***) has failed to be delivered. Please refer to the below for details. * Recipient :  ***@***.***) * Sent time :  2025-08-11T06:13:49 * Subject :  [google-deepmind/deepmind-research] Fix DM21 functionals incorrectly defaulting to DM21M (#589) (PR #647) * Remote host said :  Your mail was denied from the receiver. This message was sent from a notification-only address that cannot accept incoming email. For more information, please contact ***@***.*** © Dooray!.