Disabling Query Optimization in MarabouCore to Preserve Intermediate Variable Bounds

[ace-shifu]

Description

When using the pip-installed maraboupy library, MarabouCore appears to optimize queries during preprocessing, causing intermediate variables (x2, x3, x4, x5 in my case) to lose their bounds information (showing min/max float values instead). How can we disable this optimization to preserve bounds for intermediate variables?

Minimal Reproduction

from maraboupy import MarabouCore
from maraboupy.Marabou import createOptions

# Initialize query
inputQuery = MarabouCore.InputQuery()
inputQuery.setNumberOfVariables(8)

# Mark input variables
inputQuery.markInputVariable(0, 0)
inputQuery.markInputVariable(1, 1)

# Define equations (x2 = x0+x1, x3 = x0-x1)
equation1 = MarabouCore.Equation()
equation1.addAddend(1, 0)
equation1.addAddend(1, 1)
equation1.addAddend(-1, 2)
equation1.setScalar(0)
inputQuery.addEquation(equation1)

equation2 = MarabouCore.Equation()
equation2.addAddend(1, 0)
equation2.addAddend(-1, 1)
equation2.addAddend(-1, 3)
equation2.setScalar(0)
inputQuery.addEquation(equation2)

# Add clip constraints (x4 = clip(x2,0,3), x5 = clip(x3,-0.5,1))
MarabouCore.addClipConstraint(inputQuery, 2, 4, 0, 3)
MarabouCore.addClipConstraint(inputQuery, 3, 5, -0.5, 1)

# Set input bounds
inputQuery.setLowerBound(0, -1.1)
inputQuery.setUpperBound(0, -1.0)
inputQuery.setLowerBound(1, -2.1)
inputQuery.setUpperBound(1, -2.0)

# Calculate bounds
options = createOptions()
exitCode, bounds, stats = MarabouCore.calculateBounds(inputQuery, options, "")

Observed Behavior

Preprocessing reduces the equation count (from 10 to 8) and variable count (from 16 to 11). Intermediate variables show unbounded behavior:

bound of x2: (-1.79e+308, 1.79e+308)
bound of x3: (-1.79e+308, 1.79e+308)
bound of x4: (-1.79e+308, 1.79e+308) 
bound of x5: (-1.79e+308, 1.79e+308)

Expected Behavior

Intermediate variables should respect their natural bounds derived from:

x2 = x0 + x1 → (-3.2, -3.0)
x3 = x0 - x1 → (0.9, 1.0)
x4 = clip(x2,0,3) → [0,3]
x5 = clip(x3,-0.5,1) → [0.9,1.0]

So, What configuration option(s) can disable the preprocessing optimization that removes these variable bounds? We need to retain bounds information for intermediate variables in our analysis pipeline.