Decision Matrix

A lookup table for the question "given my objective, my bounds, and my formulation — is the problem well posed, which solution will I get, and can I trust it?" Read the prose reasoning behind every entry in Bounds, Branches, and Feasibility.

How to use this
  • Dataset authors: find your intended objective and confirm the instance lands in a green row before shipping.
  • Validators: this is the map of where your formulation should agree with the benchmark and where it is expected to diverge. Divergence inside a "✗ not physical" cell is not a bug in your code.

Objective × bounds × formulation

Legend: ✓ = well behaved / exact / physical; ⚠ = conditional, verify; ✗ = ill posed / inexact / non-physical.

Objective familyBounded? (no op. limits)Bounded? (full op. limits)Branch favouredRelaxation exactness (SOC/SDP)Recommended model
Min generation cost (non-neg. coeffs)⚠ coercive only if strongly convex (e.g. strictly convex quadratic)High-voltage✓ under standard conditionsRelaxation, verify ex post
Min losses / min slack power✓ (coercive)High-voltageRelaxation
Feasibility (constant objective)✗ vacuous: every profile feasible, iterates drift (not "unbounded" — a constant objective can't be)Undetermined among feasible pts⚠ one-sided certificate onlyNonconvex or verified relaxation
Min import from gridHigh-voltage⚠ with dispatchable DER (drives local injection up → upper v bounds)Relaxation, watch upper bounds
Max Σ voltage magnitude⚠ unbounded above without High-voltage⚠ fails when upper v bounds bindRelaxation, watch upper bounds
Max PV / DER injection⚠ bounded but boundary-seekingDrives to upper v bound✗ typically inexactNonconvex; see list
Max load served / loadability⚠ drives to the noseToward low-voltage / nose✗ inexact near collapseNonconvex + continuation
Max transfer / ATC / interface flow⚠ boundary-seekingMaximizes currentNonconvex
Max export / feed-inReverse-flow current maxNonconvex
Max storage throughput / arbitrage volumeCycling-current max✗ (plus SoC-constraint inexactness)Nonconvex
Max revenue / profitMore generation & flowNonconvex
Min cost with a negative coefficientInverted → toward low-voltage✗ data-inducedAudit signs; nonconvex if real
The columns are not independent — and none of them is a certificate

"Bounded," "branch," and "exact" are three readings of one underlying property (see §5). A row that is ✗ on exactness is almost always boundary-seeking and low-voltage-leaning too. The green family at the top — objectives non-decreasing in generation/losses — is green in every column for the same reason.

The "branch favoured" column records which branch an objective pulls toward, not which branch a local solver will return. Confirm the sheet a solve actually landed on with the diagnostics; a green row is a reason to expect an operational answer, not evidence that you got one.

Formulation cross-cut

The table's exactness column assumes a single-phase, radial-or-mild-mesh setting. Adjust as follows:

  • Linear approximations (LinDistFlow, network/DC): always bounded and convex given bounds; never "inexact" in the relaxation sense, but they cannot represent collapse or branch multiplicity at all. Use for screening, not for any question on this page.
  • Convex relaxations (SOC-BFM/BIM, SDP): the exactness column applies. On meshed networks the radial guarantees weaken; on multiphase / unbalanced networks treat all ✓ as ⚠ and verify (Bernstein et al., 2018).
  • Nonconvex AC: no exactness question, but every ✗/⚠ row can still converge to a low-voltage or locally optimal solution depending on the start point. Use a high-voltage (flat or warm) start and check the diagnostics.

See also: Bounds, Branches, and Feasibility · Objectives that imply loss maximization · Diagnostics & validation · Trusting the solver · References