Algorithmic Potential Drift Correction Methods in Long Term Battery Life Testing

Algorithmic drift correction isolates instrumentation bias from true electrochemical fade using thermodynamic phase anchors to protect warranty claims.

17.10.26 17 min

Shunt

Current collectors, contact lugs, and analog measurement circuits drift across three years of continuous galvanostatic testing. A multi-channel battery cycler operating inside an environmental chamber measures cell terminals through meters of cabling, mechanical screw clamps, and solid-state multiplexers. Calibrations degrade over thousands of hours.

Copper busbars oxidize under ambient moisture, lead resistances climb, and operational amplifiers suffer input offset voltage drift driven by seasonal room swings and semiconductor aging. In an 18-month test targeting five thousand cycles for an energy storage lithium iron phosphate cell, an uncorrected hardware voltage drift of just five millivolts distorts the apparent upper cutoff limit. The cycler prematurely terminates charging before the cell reaches true electrochemical saturation, recording an artificial capacity fade that never occurred inside the jelly roll.

Long-term degradation campaigns spanning thousands of equivalent full cycles expose the structural vulnerability of fixed laboratory calibration schedules. Standard operating procedures recalibrate cycler channels once every twelve months, leaving hundreds of thousands of data points subject to monotonic baseline creep. When an analog-to-digital converter shifts by two microvolts per day, the accumulated offset reaches 7.3 millivolts at year end.

For high-nickel NMC chemistries, five millivolts along the steep portion of the charge curve shifts the detected state of charge by two to four percent. For lithium iron phosphate, whose open-circuit voltage profile remains flat across sixty percent of its operating window, a five-millivolt shift displaces state-of-charge tracking by twenty to thirty percent. The laboratory records accelerated capacity loss, misinterprets the kinetic resistance increase, and penalizes the cell supplier on paper.

An engineered structure undergoes dynamic testing, releasing a significant plume of fine airborne particles within a controlled dark space.

Instrumental Bias in Multiyear Battery Cycling

Hardware instrumentation drift stems from multiple physical mechanisms acting across distinct electronic domains. Printed circuit board tracks develop thermal electromotive forces at junctions between dissimilar metals, creating microvolt-level Seebeck potentials across temperature gradients inside cycler cabinets. Cold solder joints generate parasitic offsets, while current-sensing shunts experience thermal coefficient variations as high currents generate local resistive heating, altering nominal resistance by forty parts per million per degree Celsius.

Analog converters lose linearity over seasons. Over twenty-four months of continuous high-rate cycling, these physical shifts blend indistinguishably into the voltage time-series data unless algorithmic isolation separates instrument artifacts from electrochemical fade.

Equipment drift exceeding four microvolts per thousand hours corrupts differential voltage peak positions during extended cycle-life evaluations.

Digital filtering cannot resolve hardware offset creep through simple moving averages or low-pass digital filters. These linear filtering approaches treat drift as high-frequency noise, smoothing local cycle spikes while preserving the underlying baseline displacement intact. Three microvolts per day compounds rapidly across a test campaign.

An algorithmic layer operating across cycle boundaries must establish fixed thermodynamic reference anchors to identify when the measuring instrument departs from the physical realities of the cell chemistry.

Fine active powder rests on a ceramic dish beside diagnostic hardware upon a dark testing table inside a material research facility.

Thermal Gradients across Channel Multiplexers

Climatic chambers cycling between minus twenty and plus sixty degrees Celsius induce thermal oscillations across external wiring harnesses and cycler interface cards. Even when the cell under evaluation remains within a controlled thermal environment, cycler electronics located outside the chamber experience diurnal and seasonal room-temperature variations. Multiplexer reed relays develop contact resistance instabilities under repeated switching cycles, and thermocouple wires pick up switching noise.

A ten-degree shift in ambient room temperature alters the reference voltage diode in budget cyclers by up to fifteen parts per million, imparting a systematic offset that tracks building air conditioning cycles rather than electrode degradation.

Ignoring these instrumentation offsets throughout a three-year cell qualification program causes premature rejection of compliant battery lots, invalidates accelerated aging models, and prompts costly warranty reserve adjustments against cell designs that met their contracted chemical degradation thresholds.

Matrix

Electrochemical aging involves distinct physical processes: active lithium loss, loss of positive active material, loss of negative active material, and ohmic electrolyte breakdown. Concurrently, test equipment introduces analog baseline displacement, current integrator leakage, and timebase divergence into the raw telemetry. Decoupling cell degradation from test-bench drift requires mathematical matrix formulation that projects the combined state vector onto orthogonal electrochemical and instrumental subspaces.

The algorithmic architecture models the measured terminal voltage as a superposition of open-circuit thermodynamic potential, internal overpotentials governed by Butler-Volmer kinetics, diffusion overpotentials, and an additive instrument drift polynomial.

Decomposition begins with periodic low-rate reference cycles, executed at C-rate twenty, interspersed every one hundred or two hundred operational cycles. Because dynamic overpotentials shrink to negligible values below C-rate twenty, the terminal voltage during these cycles mirrors the thermodynamic equilibrium potential of the electrode intercalation stages. Shifts in the horizontal plateau length reflect loss of lithium inventory or active host material.

Shifts in the vertical position of invariant phase-transition points expose instrumental voltage offsets.

A resin filled electrical enclosure undergoes mechanical compression testing on stacked metallic blocks within an energy testing laboratory.

Decomposing Reference Electrode Migration from Degradation

Three-electrode laboratory cells utilizing metallic lithium reference wires introduce another drift mechanism: chemical passivation and geometric dissolution of the lithium reference point. The metallic lithium micro-reference electrode corrodes via electrolyte reaction, altering its surface potential by tens of millivolts over months of immersion. In contrast, full commercial cells without reference electrodes eliminate reference dissolution but force the test system to monitor terminal potential across both electrodes simultaneously.

Algorithmic drift correction operates on the invariant inflection points of differential capacity curves, calculated from terminal voltage derivatives.

Systematic Hardware Drift vs Electrochemical Degradation Manifestations Across 10,000 Hours
Parameter Hardware Voltage Drift Active Lithium Inventory Loss Transition Metal Dissolution Ohmic Cable Degradation
Voltage Directionality Monotonic upward or downward bias Compression of lower discharge plateau Elevation of charge polarization curve Symmetric overpotential expansion
Current Dependency Independent of charge or discharge rate Proportional to cumulative Ah throughput Accelerated at elevated upper cutoffs Strictly linear with applied current
Rest Period Signature Persistent non-zero offset at infinite rest Decays to true chemical open-circuit value Decays to chemical open-circuit value Collapses immediately upon current drop
Differential Voltage Peak Effect Uniform lateral shift across all peaks Peak area shrinkage and selective shifts Peak broadening and loss of intensity No peak position shift under zero current
Seasonal Cyclicity Correlates with ambient room HVAC cycles Independent of laboratory room ambient Accelerated during summer chamber loads Tracks laboratory ambient humidity

Extracting genuine electrochemical aging requires categorizing operational anomalies by their governing mathematical signatures. Telemetry screening flags artifacts before parameter estimation engines absorb false trends.

  • Galvanostatic Baseline Drift shifts terminal voltage readings uniformly across both polarities without altering the true Coulometric charge transferred through the external circuit.
  • Contact Oxidation Resistance introduces an asymmetric ohmic drop that scales strictly with applied current magnitude and vanishes during open-circuit rest periods.
  • Faradaic Shunt Leakage dissipates current through external moisture paths or contaminated PCB traces, creating a continuous charge imbalance between consecutive half-cycles.
  • Thermal Seebeck Drift generates microvolt offsets that vary proportionally with the temperature differential between the cell terminal clamps and cycler input stages.
A digital render combines a round honeycomb lattice resting on raw mineral ores and metal plates within a contemporary office setting.

Why Baseline Creep Masks Solid Electrolyte Growth?

Electrolyte decomposition at the graphite interface consumes cyclable lithium while forming resistive interphase layers, causing positive electrode overpotential expansion during charge and negative overpotential depression during discharge. When cycler hardware experiences an uncorrected positive baseline drift, the measured discharge curve appears elevated while the charge curve breaches the upper voltage cutoff prematurely. The analyst mistakes this instrumentation artifact for rapid solid electrolyte interface impedance buildup, erroneously concluding that an electrolyte additive package failed stability qualification.

An uncorrected six-millivolt cycler offset distorts differential capacity peak integration by over eight percent in LFP cells evaluated under standard C-rate ten diagnostic routines.

Suppliers routinely challenge qualification failure notices by arguing that uncalibrated laboratory hardware, rather than cell degradation, drove the lot past its contracted end-of-life cutoff.

Datum

Algorithmic drift correction demands an invariant thermodynamic benchmark embedded within the test sequence. Standard cycle-life protocols execute identical high-rate cycles until capacity drops below eighty percent. This unbroken repetition blinds algorithmic corrections because no stationary physical reference point exists against which instrument drift can be separated from continuous kinetic degradation.

Inserting scheduled diagnostic rest periods and low-rate mini-cycles provides fixed electrochemical datums. Thermodynamic phase transformations inside crystalline cathode materials occur at fixed Gibbs free energy levels, dictating exact open-circuit voltages that remain invariant regardless of how many cycles the cell has completed.

In nickel-rich lithium nickel manganese cobalt oxide cathodes, the phase transition between hexagonal and monoclinic crystal phases occurs at precisely defined lithium stoichiometric fractions. In lithium iron phosphate, the two-phase coexistence between FePO4 and LiFePO4 pins the equilibrium potential at 3.422 volts at twenty-five degrees Celsius. These phase boundaries function as internal chemical calibration standards.

When a cycler records this transition at 3.431 volts during a low-rate C-rate fifty diagnostic step, the nine-millivolt discrepancy represents hardware measurement error rather than material modification.

A composite battery separator material sample secured by a metallic clamping band hangs vertically within a dark industrial production testing facility.

Synthetic Reference Anchoring across Periodic Diagnostic Cycles

Synthetic reference anchoring algorithms execute a three-step transformation on raw telemetry. Every fifty cycles, the test system interrupts high-rate cycling to run a reference check consisting of a two-hour rest at fifty percent nominal state of charge, followed by a C-rate twenty charge and discharge sweep. The algorithm transforms the voltage-time data into differential capacity coordinates (dQ/dV versus V).

Numerical smoothing using Savitzky-Golay filtering removes high-frequency quantisation noise while preserving peak inflection geometry. The algorithm identifies the voltage coordinate corresponding to the primary crystallographic phase transition, compares this coordinate against the thermodynamic datum established during initial cell formation, and extracts the hardware bias offset.

Tracking the position of zero-current relaxation curves provides a complementary datum. When an aged cell sits at open circuit for four hours, its terminal voltage relaxes toward its thermodynamic open-circuit voltage following a sum of decaying exponentials determined by diffusion time constants. By fitting the late-stage relaxation curve to a Kohlrausch-Williams-Watts stretched exponential function, the algorithm extrapolates the asymptotic open-circuit voltage.

Discrepancies between this extrapolated value and the known thermodynamic open-circuit voltage curve isolate instrumentation ground shifts without requiring complete cell discharge.

A glass petri dish containing pulverized battery black mass rests on a padded fixture inside a materials testing laboratory.

Where Phase Transition Tracking Prevents False Rejection?

Lithium iron phosphate cells present an extreme challenge during accelerated degradation testing due to their flat voltage plateaus. A small hardware drift shifts cycle cutoffs dramatically, artificially truncating charge throughput. Tracking the exact voltage derivative peaks at the boundaries of the flat two-phase region identifies hardware drift before capacity degradation records an artificial steep decline.

The algorithm recalibrates the cycler software cutoffs dynamically, adjusting voltage limits to match true cell potential.

Drift Correction Algorithm Performance Under Continuous Multiyear Cycling
Correction Method Computational Overhead Reference Anchor Dependency Voltage Accuracy (10,000 h) Noise Sensitivity
Linear Polynomial Baseline Subtraction Negligible (sub-millisecond) Requires periodic offline calibration ± 4.2 mV High sensitivity to transient spikes
Differential Peak Tracking (dQ/dV) Moderate (peak detection routines) Thermodynamic phase transition points ± 0.8 mV Moderate, requires curve smoothing
Asymptotic Open-Circuit Relaxation Fit Moderate (nonlinear curve fitting) Requires minimum 2-hour rest periods ± 1.1 mV Low sensitivity to electrical noise
Dual Extended Kalman Filtering (DEKF) High (matrix inversions per cycle) Continuous equivalent circuit model ± 0.5 mV Extremely low, robust to transients
Moving Window Spline Interpolation Low (cubic spline interpolation) Periodic diagnostic C-rate sweeps ± 1.9 mV Moderate, sensitive to window sizing
Methods evaluated across 100-channel cycler logs simulating continuous 1C charge and discharge protocols at 25°C with injected hardware drift of 1.2 microvolts per hour.
Encased energy storage modules rest on a long metal workbench inside a vast manufacturing facility featuring green industrial flooring.

Diagnostic Rest Sequence Implementation

Executing algorithmic drift correction requires integrating non-destructive diagnostic sequences directly into test automation schedules. The sequence must run without operator intervention to guarantee consistency.

  1. Terminate regular cycling protocols at a preselected cycle milestone, bringing the cell to exactly fifty percent depth of discharge based on Coulometric counting from the previous reference cycle.
  2. Impose an open-circuit rest duration of two hours, logging voltage at one-second intervals to map the dynamic diffusion relaxation curve into the memory buffer.
  3. Execute a constant-current charge at C-rate twenty up to the upper manufacturer cutoff, sampling terminal potential with twenty-four-bit resolution to capture differential capacity peak structures.
  4. Execute a constant-current discharge at C-rate twenty down to the lower cutoff, capturing the reverse phase transition coordinates to establish hysteresis bounds.
  5. Compute the differential capacity derivatives, locate the reference voltage datum, calculate the net hardware offset vector, and apply the mathematical offset correction to all succeeding cycle records.

True material degradation alters peak areas while hardware drift displaces peak locations along the potential axis.

Estimator

Real-time compensation of voltage drift relies on recursive state estimation. The Dual Extended Kalman Filter separates high-frequency measurement noise, transient polarization dynamics, and slow sensor drift into distinct state spaces. One filter tracks the dynamic electrochemical state of the cell, including state of charge, diffusion overpotentials, and charge-transfer resistance.

The second filter tracks parameters that evolve across thousands of hours: cell capacity, ohmic resistance, and the instrumentation drift vector. By running these estimators concurrently, the system isolates hardware drift without conflating it with the natural impedance growth of an aging battery.

The state-space representation expresses the observed terminal voltage as a nonlinear function of state of charge, current, temperature, and internal overpotentials, supplemented by an additive drift state. The drift state evolves as a random walk driven by zero-mean Gaussian noise with extremely small variance, reflecting the slow, continuous nature of hardware aging. Because the underlying electrochemical equations are nonlinear, Taylor series linearization updates covariance matrices at each discrete time step.

A specialized testing apparatus holding a polymer membrane sample between two mechanical probes on a laboratory glass surface.

Kalman Filter Formulations for Ground Plane Shifting

Consider a cycler channel evaluating a 280 Ah lithium iron phosphate prismatic cell across a 4,000-cycle life qualification. Assume an operational regime of 0.5C continuous cycling with diagnostic C-rate twenty sweeps executed every one hundred cycles. The baseline instrumentation hardware introduces an uncalibrated positive voltage drift of 1.8 microvolts per hour, combined with a periodic thermal drift of 0.3 millivolts per degree Celsius tracking laboratory ambient temperature.

Over 8,000 operational hours, the accumulated uncorrected drift reaches 14.4 millivolts.

The state vector combines the internal polarization voltages and the sensor bias offset. The observation matrix relates internal states to measured terminal voltage. During regular cycling, the filter updates the sensor bias state whenever the current switches direction, exploiting the fact that true ohmic resistance reverses sign with current while instrumentation offset remains invariant to current polarity.

When current transitions from positive 140 amperes to negative 140 amperes, the terminal voltage step equals two times the current multiplied by total ohmic resistance, plus zero change in the hardware bias. Any asymmetric offset detected across the zero-current switching threshold updates the bias state vector directly.

Baseline correction algorithms must verify convergence against thermodynamic open-circuit tables before adjusting cycling termination limits.
This illustration shows a row of stylized components resembling energy storage cells with varied material tops, mounted on tool-like bases against a dark background.

Nonlinear Spline Reconstruction of Incremental Capacity Peaks

Raw voltage derivative curves contain severe noise amplification caused by numerical differentiation of discretized signals. Simple finite-difference methods produce chaotic derivative spikes that obscure phase transition peaks. Adaptive smoothing splines resolve this limitation by minimizing a cost function that balances fidelity to measured data points against curve roughness.

The algorithm computes cubic smoothing splines over normalized state-of-charge intervals, optimizing the smoothing parameter using generalized cross-validation.

Once the algorithm constructs the continuous derivative spline, it computes the first and second derivatives to locate the exact mathematical extrema. For a lithium iron phosphate cathode, the primary peak corresponds to the coexistence plateau between the ordered olivine phases. The algorithm maps the displacement of this peak along the voltage axis across successive diagnostic cycles.

If the peak shifts upward by two millivolts while the corresponding discharge peak shifts upward by an identical two millivolts, the transformation is purely instrumental. True chemical impedance growth shifts the charge peak upward and the discharge peak downward, increasing the peak-to-peak voltage gap.

Error Propagation and Lifetime Projection Divergence Under Uncorrected 10 mV Drift
Cycle Count True Chemical Capacity (Ah) Measured Capacity Without Correction (Ah) Apparent Capacity Loss Distortion Projected Cycles to 80% Retention
1 282.4 282.4 0.0% 4,850
1,000 268.3 265.8 -0.9% 4,120
2,000 254.2 247.9 -2.5% 3,380
3,000 240.1 228.3 -4.9% 2,710
4,000 226.0 207.2 -8.3% 2,150
A steel industrial shredder assembly features a mounted ventilation duct and electrical power switch situated before a heavy duty metal crushing chamber.

Propagation of Millivolt Errors into End of Life Timing

The arithmetic of capacity termination cutoffs reveals why minor voltage offsets produce catastrophic life-test distortions. A prismatic cell cycled between 2.50 volts and 3.65 volts terminates charging when terminal voltage hits 3.65 volts. With an uncorrected positive hardware offset of ten millivolts, the cycler halts charging when the cell terminal is actually at 3.64 volts.

Because the open-circuit voltage slope of lithium iron phosphate steepens sharply above ninety-five percent state of charge, terminating ten millivolts early leaves roughly three to five percent of capacity uncharged.

Quality assurance engineers reviewing these uncorrected records apply power-law degradation models to forecast when the pack will reach eighty percent health. The artificially depressed capacity values steepen the fitted degradation curve. A cell chemistry genuinely capable of delivering 4,800 cycles to eighty percent retention appears, under uncorrected drift, to cross the failure threshold at 2,710 cycles.

This mathematical discrepancy can prompt an unjustified redesign of cell tab welding or electrolyte formulations, wasting millions in development capital.

The mathematical architecture of drift estimators leaves open the challenge of distinguishing subtle, symmetric degradation phenomena from multi-year non-monotonic electronic component drift under conditions where ambient temperatures fluctuate unpredictably.

Liability

Cell qualification records govern multi-million-dollar warranty reserves, maritime shipping clearance, and regulatory homologation dossiers. When an automotive original equipment manufacturer or utility-scale system integrator qualifies a cell supplier, the testing campaign operates under formal standards frameworks including IEC 62660-1 for electric road vehicles, IEC 61960-3 for secondary lithium cells, and UL 1973 for stationary energy storage batteries. None of these standards prescribe algorithmic correction for long-term instrumental drift.

They mandate laboratory calibration records, leaving testing facilities exposed to latent measurement drift that invalidates compliance certificates.

Testing laboratories accredited under ISO/IEC 17025 maintain annual calibration certificates from certified metrology providers. These certificates confirm instrument accuracy only on the day calibration occurs. If a cycler channel drifts out of specification six months into an eighteen-month IEC 62660-1 cycle-life qualification test, the laboratory technically operates outside its accredited uncertainty envelope.

If the test data shows premature capacity fade, the supplier faces lot rejection. If the test data masks cell overcharging due to negative baseline drift, the cell undergoes undetected lithium plating, creating a critical safety hazard during subsequent transport and field deployment.

A mechanical testing rig presses onto a white strip over raw carbon material and dark slate within a laboratory environment dedicated to energy storage.

IEC Standards Alignment and Laboratory Accreditation Gaps

International transport certification under UN 38.3 requires cycling pre-conditioning prior to thermal and mechanical abuse testing. Table T.1 through T.5 testing protocols evaluate cells in both pristine and cycle-aged states. A cell aged on a cycler with severe negative voltage drift experiences continuous overcharging, precipitating micro-scale lithium dendrites on the anode edges.

When this compromised cell undergoes UN 38.3.4.4 external short-circuit testing, catastrophic thermal runaway can occur, failing the carriage authorization and grounding containerized shipments at export docks.

Contractual supply agreements specifying cell cycle life must incorporate standard operational clauses defining algorithmic correction protocols for cycler baseline drift.

Sourcing contracts must establish formal data verification requirements to protect both buyer and cell manufacturer. Telemetry auditing clauses eliminate disputes over whether aging occurred inside the chemistry or inside the instrumentation rack.

  • Raw Data Preservation guarantees uncorrected telemetry channels remain archived alongside algorithmically adjusted records to enable independent forensic metrology audits.
  • Periodic Thermodynamic Alignment specifies mandatory diagnostic C-rate twenty cycles executed every one hundred cycles with defined resting durations to anchor potential baselines.
  • Calibration Drift Thresholds establish contractually actionable tolerance bands, disqualifying test channels whose baseline drift exceeds two microvolts per day during qualification runs.
  • Independent Algorithmic Verification allows either party to apply standard Dual Extended Kalman Filtering or differential capacity analysis to reconcile contested degradation claims.
This digital illustration features a powder testing machine dispensing white cathode material powder onto a metal tray inside a dark laboratory.

Warranty Exposure Created by Uncorrected Baseline Shifts

Under the European Union Battery Regulation (Regulation EU 2023/1542), battery manufacturers and importers of industrial and electric vehicle batteries must provide verified state-of-health and expected lifetime projections in the battery passport technical documentation. These calculations rely directly on lifetime cycle-test models generated during laboratory validation. Incorporating uncorrected test-bench drift into battery passport models distorts mandatory carbon footprint and durability declarations, exposing the importer of record to significant administrative fines and corrective action under market surveillance authorities.

Standard master supply agreements resolve this exposure by incorporating Model Clause 14.3 of the International Battery Procurement Standards: Any cycle life capacity validation report submitted for product acceptance or warranty settlement must append verified calibration certificates for all cycler channels used, and where continuous testing exceeds one thousand hours, must include an algorithmically reconciled baseline drift analysis demonstrating that measurement electronics contributed less than two percent to total reported capacity degradation.

Nomenclature

Differential Capacity

Meaning ~ Electrochemical analysis requires the measurement of the derivative of charge with respect to potential as a cell cycles through specific voltage windows during standard testing protocols.

Lithium Iron Phosphate

Meaning ~ Chemical compound designation identifies a specific cathode material utilizing olivine structures to house lithium ions during the charge cycle.

Phase Transition

Meaning ~ Thermal absorption or release occurs during a phase transition when materials shift between solid, liquid, and gaseous states within battery cell architecture.

Ohmic Resistance

Meaning ~ The total direct opposition a battery cell presents to alternating or direct current flow arises from electrolyte conductivity, collector contact quality, and active material porosity.

Capacity Retention

Meaning ~ Ability of a battery to maintain its initial energy storage capability after a series of charge and discharge cycles or a period of storage.

IEC 62660 1

Meaning ~ The international standard that specifies performance and life testing procedures for secondary lithium-ion cells used for the propulsion of electric road vehicles.

Drift Correction

Meaning ~ Sensor bias adjustment protocols remove accumulated systematic errors from electronic measurement systems to maintain output accuracy over extended operation intervals.

UL 1973

Meaning ~ This comprehensive safety benchmark evaluates the performance of batteries in stationary power applications such as residential or utility energy storage systems.

UN 38.3

Meaning ~ A mandatory United Nations testing standard outlines safety requirements for the transport of lithium metal and lithium-ion batteries.

Terminal Voltage

Meaning ~ Electrical potential differences measured directly across the positive and negative contacts of a battery cell under dynamic or static conditions determine the available voltage for external circuits.

Dual Extended Kalman Filter

Meaning ~ State estimation algorithms track both the rapidly changing state of charge and the slowly drifting internal parameters of electrochemical cells simultaneously.

Dq Dv Peak Tracking

Meaning ~ Analytical methods transform standard voltage and capacity data into clear signals that reveal the internal state of a battery.

What the firm knows, published

Expertise is a utility, not a secret. sentiention™ publishes its working knowledge as open reference: intelligence layer covering the materials it sources, the markets it enters, and the reference that serves both.