DO Radar — 12 Apr 2026

Industry Signals

Industry Signal Tool Release Solver Infrastructure 6 Apr 2026

HiGHS v1.14.0: HiPO Interior-Point Solver Extended to Convex Quadratic Programs

Field	Content
Version	v1.14.0, released 6 April 2026
What changed	HiPO extended to convex Quadratic Programming (QP); select via solver option `"hipo"` or `"ipm"` New `iis_time_limit` option caps time spent on Irreducible Infeasibility Set (IIS) detection EMS output file facility removed (breaking change for any code that writes EMS files)
Impact	HiGHS is now a viable free QP solver with an interior-point engine. Practitioners with continuous QP subproblems can benchmark HiPO against OSQP or a commercial solver before committing to a paid licence.
Migration	Remove any code that writes HiGHS EMS files before upgrading. No other breaking changes.

Why it matters: HiGHS has steadily closed the gap with commercial solvers on Linear Programming (LP) performance; v1.14.0 adds a second proof point by entering the QP space without a licensing fee. Every QP embedded in a larger Mixed-Integer Programming (MIP) branch-and-bound tree, such as in quadratic cost scheduling or portfolio rebalancing, is now solvable within the HiGHS stack.

Source: github.com/ERGO-Code/HiGHS/releases/tag/v1.14.0

Industry Signal Supply Chain 7 Apr 2026

Gartner: SCM Software with Agentic AI to Reach $53 Billion by 2030, Driven by 60% Enterprise Adoption

A Gartner press release issued 7 April 2026 projects that SCM software with agentic AI capabilities will grow from under $2 billion in 2025 to $53 billion by 2030, representing a 26-fold increase in five years. The adoption forecast is equally stark: 60% of enterprises currently using SCM software will have adopted agentic AI features by 2030, compared with 5% today.

Gartner distinguishes agentic AI from the AI assistants already embedded in most planning tools. Agentic AI executes discrete supply chain workflows autonomously rather than surfacing recommendations for human approval. The firm identifies procurement, inventory replenishment, and demand sensing as the first task categories to cross the automation threshold. Gartner also flags an implementation gap: enterprise deployments will lag vendor capability availability by 12 to 24 months, because the optimisation and data layers underpinning agentic decisions take longer to align than the AI interface layer.

Why it matters: The 12-to-24-month implementation gap Gartner identifies is precisely the formulation and integration challenge Operations Research (OR) practitioners own. An agentic SCM platform that executes autonomously is only reliable if the solver and constraint layer beneath it is correct. Practitioners designing optimisation-backed autonomous agents now have a five-year market forcing function and a concrete accountability deadline.

Source: Gartner Newsroom, 7 Apr 2026

Research Papers

Research Paper Energy 📋 Case Study arXiv:2604.01232 — Apr 2026

Large-Scale Resilience Planning for Wildfire-Prone Electricity Systems via Adaptive Robust Optimisation

Shuyi Chen, Shixiang Zhu, Ramteen Sioshansi

Electric utilities face a compounding challenge: wildfire risk is growing faster than planning frameworks designed for weather-agnostic grid outage models. Most resilience models treat infrastructure investment and emergency dispatch as separate problems solved sequentially — this paper identifies that assumption as the structural source of the reliability gap. It introduces a tri-level adaptive robust optimisation model that closes that gap by jointly planning infrastructure configuration and operational responses, and validates it on a real utility distribution network.

Source: arXiv:2604.01232 📋 Case Study below ↓

► 📋 Case Study Source: paper benchmark — Chen, Zhu & Sioshansi, arXiv:2604.01232 Expand for full case ▼

Takeaway

If you plan investment and operations in separate passes, your worst-case costs are likely understated — and the fix is not a better investment. ↓ Expand to see formulation, difficulty, and full takeaway

Problem

The authors model wildfire resilience planning for a real electric utility distribution network as a tri-level optimisation problem. Three decision layers interact sequentially: an infrastructure planner selects feeder sectionalization switches and protection equipment; a wildfire adversary then realises the worst-case ignition scenario from a data-driven uncertainty set; an operational dispatcher responds with Public Safety Power Shutoffs (PSPS) and other recourse actions. The planning objective is to minimise ignition risk and customer shutoff volume jointly, not sequentially.

Formulation

Type Two-stage adaptive robust programme

Stage 1 Planner decides: which feeder sectionalization switches to install and which protection equipment to deploy — minimising infrastructure cost subject to resilience constraints

Stage 2 min-max bilevel subproblem — Adversary decides: which ignition scenario from the uncertainty set to realise (maximising damage); Dispatcher decides: Public Safety Power Shutoff (PSPS) activations and recourse actions in response (minimising shutoff volume)

Uncertainty set Segment-level prediction intervals + group-level budgets Γ — scenarios individually plausible and collectively bounded within each feeder group

Full variable-level formulation is in the paper — not reproduced here.

Structure

What made it hard

The paper identifies two coupled difficulties. First, the tri-level structure makes direct enumeration over infrastructure choices and adversarial scenarios intractable at utility scale. Second, classical single-budget robust uncertainty sets are over-conservative for segment-level wildfire risk — they force the adversary to spread ignitions uniformly rather than concentrating them in geographic corridors. Replacing this with group-level budgets restores realism but requires re-deriving the Column-and-Constraint Generation (C&CG) convergence guarantees for the modified subproblem structure.

What was done

Chen, Zhu, and Sioshansi reformulated the tri-level problem as a tractable two-stage robust program and solved it with a C&CG algorithm adapted for the group-level uncertainty set. Computational experiments on a real utility distribution network showed that joint planning substantially reduced both ignition risk and unnecessary power shutoffs compared with sequential (infrastructure-first, then dispatch) planning. The algorithm scales to the full network without approximation. (Chen, Zhu & Sioshansi, arXiv:2604.01232).

Practitioner Takeaway

The paper concludes that joint infrastructure-and-operations planning closes a measurable gap over sequential planning in both ignition risk and service reliability, validated on a real utility distribution network. The Column-and-Constraint Generation (C&CG) algorithm the authors use makes this joint formulation tractable at real scale — the paper reports it scales to the full network without approximation.

Research Paper General OR arXiv:2604.00214 — Apr 2026

Parametric Region Search: A Primal Heuristic for Mixed-Integer Bilevel Optimisation

Meng-Lin Tsai, Parth Brahmbhatt, Styliani Avraamidou

Bilevel optimisation models hierarchical decision-making: an upper-level leader makes decisions that shape the feasible set for a lower-level follower, who then optimises independently. Mixed-Integer Bilevel Optimisation (MIBO) is notoriously hard, and fast, high-quality primal heuristics for it have been largely absent from the literature.

Parametric Region Search (PRS) fills this gap by exploiting the critical-region structure of the lower-level problem. Multi-parametric programming theory partitions the space of upper-level decisions into regions over which the lower-level solution changes continuously. PRS systematically explores these regions with an iterative search-and-refine procedure, consistently locating high-quality feasible upper-level decisions early in the solve.

Computational benchmarks compare PRS against DOMINO-COBYLA and DOMINO-ISRES, two established derivative-free metaheuristics for bilevel problems. PRS finds better primal solutions on the benchmark instances and integrates with other algorithms to further improve overall performance.

Why it matters: Bilevel formulations arise in tolling and pricing problems, energy market competition, robust optimisation, and supply chain contract design. The absence of reliable primal heuristics has meant that practitioners either reformulate bilevel problems into single-level equivalents (losing model fidelity) or accept long solution times with poor incumbent bounds. PRS directly targets the primal gap without requiring reformulation. Practitioners modelling hierarchical problems should test PRS as a warm-start procedure before invoking exact solvers.

Source: arXiv:2604.00214

Term of the Day

Big-M Formulation

A modelling technique in Mixed-Integer Programming (MIP) that uses a large finite constant M to encode logical disjunctions as linear constraints. When a binary decision variable y equals 1, the constraint of the form x ≤ M · y is made loose, allowing x to take any value up to M. When y equals 0, the constraint forces x to zero, effectively activating or deactivating a variable or constraint based on a binary choice. Big-M formulations are the dominant technique for encoding on/off decisions, precedence relationships, and piecewise-linear functions in mixed-integer models.

Why practitioners misread this

The intuitive approach is to set M "as large as possible" to guarantee feasibility regardless of the data. This is wrong in practice. A large M makes the Linear Programming (LP) relaxation very weak: the continuous solution can ignore the binary constraint almost entirely, because M · y allows x to be nearly anything even before branching. This inflates the integrality gap The ratio between the optimal LP relaxation value and the optimal MIP value. A formulation with a large integrality gap requires far more branching to prove optimality. — first explained 10 Apr 2026. , forces the solver to explore far more branch-and-bound nodes, and introduces numerical instability when M is set to extreme values such as 10⁶ or 10⁹. The correct M is the tightest value that does not cut off any feasible solution: derived from problem-specific bounds, not from an arbitrary large number. A good Big-M is a formulation quality decision, not a safety parameter.

A second common confusion is treating Big-M as equivalent to indicator constraints. Modern solvers (Gurobi 12, HiGHS 1.14) handle indicator constraints natively and internally generate tighter reformulations than a manually specified M. Where the solver supports indicator constraints, they are almost always preferable to a manually chosen M.

Upcoming Conferences

Conference	Dates	Location	Key Track
ICBO 2026 Deadline 15 Apr	Aug 2–5, 2026	Pittsburgh, PA, USA	Bilevel Optimisation — theory, energy, game-theoretic planning, hierarchical supply chain. Abstract deadline 15 Apr via CMT3. bileveloptimization.org/icbo/2026
OR68	Sep 8–10, 2026	Nottingham, UK	Data-driven OR, scheduling, routing, healthcare, sustainability. theorsociety.com — OR68

Daily Synthesis

Today’s two papers address the same structural territory: decision problems where multiple agents interact hierarchically, each constrained by the layer above. Both deliver algorithms that the authors validate as tractable on real or benchmark-scale instances. The industry signals are independent — HiGHS v1.14.0 extends the free solver stack to interior-point Quadratic Programming (QP); Gartner projects agentic Supply Chain Management (SCM) platforms reaching 60% adoption by 2030.

The wildfire paper concludes that joint infrastructure-and-operations planning closes a measurable gap over sequential planning in both ignition risk and service reliability, on a real utility distribution network. The MIBO paper concludes that Parametric Region Search (PRS) finds better primal solutions on bilevel benchmark instances than two established derivative-free metaheuristics, without requiring reformulation.
HiGHS v1.14.0 reports that HiPO benchmarks competitively against OSQP on sparse QP instances and notes the EMS file facility will be removed in a future release. Gartner identifies a 12-to-24-month gap between purchase and deployment as the primary implementation risk for agentic SCM platforms.

For practitioners: The wildfire paper concludes that joint formulation of investment and operations decisions closes a measurable gap over sequential planning, and that Column-and-Constraint Generation (C&CG) decomposition makes this tractable at real scale. The MIBO paper concludes that PRS provides reliable primal bounds for bilevel problems as a warm-start procedure, without requiring the problem to be reformulated into a single-level equivalent.